<div class="container"><img id="randomImage" alt="Set the mood"></div>
(print: "<h3>" + (passage:)'s name + "</h3>")
Welcome to the beginning of a new era. Most of us are entangled in working with computers for a variety of goals and tasks. They form a consistent, powerful, reliable framework for many of our critical pursuits. They, along with other machinery have been replacing maual labor for decades. Mostly they perform tedious error-prone work which have allowed many to maximize our time and skils, allowing more focus to think and plan. Yet AI poses a threat. Will it //assist or replace // us has held true in automation and will prove true for AI. But how? Computers work by taking fairly simple processes and speeding them up – doubling in power and halving in cost every year or so since they were introduced. Yet that efficiency hasn't changed much of our day to day work, even if it's augmented by these tools. Even with the pandemic and remote work, we still kept to old routines and technologies. People are slow to change and we stick to our comfort zones. Airplane passengers don't know or care how the plane works as long as it gets them to their destination. Similarly, computers are programmed and follow instructions. Poor instructions can lead to poor outcomes, but that is the paradigm that will change with Generative AI. There's lots to unpack, so we wanted you to create your story of how it could change ''your ''path. A novel topic deserves a personal way to build a story. ''The choice is yours''. You can veer off into any prompt or just follow the story from passage to passage. The journey should help win you insight into how this system works and what you can do with it. You just need curiosity and the willingness to change paths or start over occasionally with new insight. I promise along the way you'll discover why this strange kind of math invented to win games is truly something worth learning about.
[[Can I go straight to the end and skip all the choices?]]
One thing that always aids in learning is ''repetition''. Thats why you may start here many times but will notice or follow different ideas on each read-through. Each paragraph can lead, if you choose, to a different thought or concept. Why make you do all this work? Well, its similar to how Generative AI and large language models function, but handmade, or hand-written by thoughtful and experienced academic professionals. This effort started in 2023 by having AI generate many passages and ideas, but the tone and point of view were missing, so we decided to do it the old fashioned way with a small twist. Interactive storytelling involves creating many scenarios, and its more impactful if you decide which thoughts are worth following. We have strong thoughts to share on the future of AI, a form of computing can reach more people than ever, and tailor the output to many complex needs and problems. We use the threat of how this technology will change you in your learning, life or work to teach you about human things, and how to make better decisions in this fast-changing world. A world where you can personalize everything, especially the way you engage with information.
[[Who should read this book?]]
''TMI: Too Much Information.'' You’ve probably heard this phrase before. It means there’s too much stuff to process, and you don’t know what to do with it. It connects with another concept of the //attention economy//, where that ability to engage your attention is a valuable resource. Jack and I are the founders of the cheekily titled TMI Knowledge Management company. We consult to companies who try to make use of all the information they generate. Staggering amounts of it. Powerpoint decks, confluence pages, SharePoint sites and meetings, meetings, meetings. Filled to the brim with information, but often little usable knowledge. Not useful seems harsh, but as anyone who remembers browsing library books, you can happen upon a more interesting book next to the ones you originally desired. This happenstance is true of anyone seeking out information actively. If you needed something, or were not quite as curious, you'll easily get overwhelmed with the amount of information a company regularly produces. And seeking out help or usefulness in all of that is challenging at best. The organization of the public library is key, most of our lives don't invest in data organization and care to the extent needed to make computer searching techniques useful. Broadly, numeric capture is different. You can plan a budget, you can see trends, and even engage in counting steps while exercising. Numbers are plenty of fun, throw in some statistics and you can manage all sorts of insights. But words, pictures, concepts. These are rarely captured as data in useful formats to provide insight. Even with the entirety of the web being captured, it takes semi-magical math and textual tricks to come up with useful results (search engines) until now.
[[Have we always been so obsessed with data?]]
Nowadays, there’s more stuff than ever. There are so many platforms, formats, and media types to choose from. There are so many ideas, cheap and easy to access. But what do we really want from this stuff? What do we call it? Content? Content is like food. It can be nutritious or junk. It can satisfy or leave you hungry. It can be good for you or bad for you. It depends on what you need, and what you use it for. Are you buying something? Are you in a hurry? Are you working on a project? Are you planning a business? Are you educating your kids? Engineering a new product or service? Are you learning something new? Whatever your goal is, you need the right stuff. The right information. The right content. My career and passion is to help people make decisions, across dozens of industries and including the decision making system for Accenture's consulting business, or truckers deciding upon a load they want to haul. Digital tools have made it much cheaper, yet not easier to decide upon a course of action. And in higher education Jack and I made our start, helping students decide how to approach the subject of Computer Science. What professors were good, what topics helped you take advantage of this booming industry? Learning how to take dumb data and form it into somehting that could change your mind or help you choose a new approach is more psychology than design. Thus this book is more of an experience. If you see the topic missing the mark, branch to another. There are also ways out, choices you can make to master this new ability to discuss rather than learn a complex language or program to get the computer magic to do your bidding.
[[Choose your adventure]]
Our goal in this book is to help you find the right stuff. To show you how to use technology and techniques that generative AI has unleashed. To explain how it works, and why it matters. To help you solve problems, avoid mistakes, and make sense of the new world of work we believe this technology will imply. To help you turn information into knowledge, and maybe even wisdom. We were inspired by all the sci-fi books and movies we loved growing up. I will cite the book //Thinking Fast and Slow// by Daniel Kayneman quite a bit since he tackled why humans make their choices. I loved traveling through the dialogs of Alice and the Mad Hatter in //Godel Escher Bach// by Douglas Hofstateler. And even more the way we make sense of senslessness in Umberto Eco's //Foucault's Pendulum//. But these are high hurdles, we have to start with why is this not just a fad, and a new way to sell something in a society that values making more money above all?
[[Be careful of what you ask for]] (print: "<h3>" + (passage:)'s name + "</h3>")
Its likely better to ask who should write this book? Truthfully, the sentiment of ''tl;dr'' (too long, didn't read) is a resonable response to all the possible choices to spend time. And at the moment Generative AI tops the scale at hype, without necessarily knowing what problem it solves, and how valuable is that solution. Or at what cost? Cynically, it’s best to wait for this to die down, a year or two, and then see what exactly all the fuss was about. Most will do this. Some will cleverly use the technology for something nefarious, or even downright antisocial. Our experience in this matter comes from advanced positions in several industries, computer learning, higher education, advertising, information architecture, leading SaaS design, and attending more than a few #SXSW events where it seemed obvious in a particular year how 3D printing would take over the world (it didn't).
[[Reading, what's that?]]
Yet, something is off. This doesn’t feel like Salesforce convincing companies to keep their vital data ‘in the cloud’ in 2010. It feels fundamentally like Einstein must have felt when he said “God doesn’t play dice with the universe’. He felt as though we were in charge, we had figured things out, but fundamentally it was ruled by chance. A very minuscule, predictable chance, but not a rule. Its always important to remember that results that come from generative AI are like natural things, they evolve, they were not planned and predicted. Human language and the ability to communicate have been available to computers for several years, but they haven’t had much to say. They can do some math, start a timer, remind us of events. But they could not, through deliberate programming, perform beyond what we consider logical operations. We invented them for this, after all, since our own unreliability was well known. Also, lots of detailed, repetitious work is a drag.
[[Organizations as systems]]
In the field of human-computer interaction, its a sea change. Rather than designing for, and forcing people into feeding the computer the data it requires, we can now chat, cajole, or confide. Soon, many will find their ideal companions within the chat window. The text can be spoken, and the speaker will be a generated avatar. There will be more accptance of meeting each other virtually. We can have fully virtual lives, which for many is nothing new. Oddly, it can understand us, can connect with us intellectually, help us. Perhaps even encourage us emotionally? Teach us? And being human could also change. One of our goals is to explore what this means, and how to prepare for the inevitable wrongs that people will inflict with this technology. Also see how we can change the nature of jobs, returning them to a love of work and exploration of new ideas that has been lacking in increasingly large and bureaucratic companies. Our ability to learn and provide value will fundamentally change. Our 40+ years of putting up with how lousy computers can be will hopefully result in some measure of common- sense advice. However, we’ve seen how people act too, so in the Venn diagram we’ll depict it has opportunities for those that prepare and embrace this as a new medium.
[[Who are you talking to?]]
<h3>Are you sure?</h3>
1. The "tl;dr" sentiment and hype around generative AI: The text acknowledges the current hype surrounding generative AI, but also expresses skepticism about whether it solves real problems and the potential costs involved. This aligns with concerns raised by some experts about the need for realistic expectations and a clear understanding of the limitations and risks of these technologies. A report by the AI Now Institute <https://ainowinstitute.org/news/the-technology-202-former-ftc-advisors-urge-swift-action-to-counteract-ai-hype">Link</a> discusses the phenomenon of "AI hype" and its potential consequences.
2. The unpredictability and "natural" evolution of generative AI outputs: The text compares the results of generative AI to natural phenomena, implying a level of unpredictability and lack of control. This connects to ongoing debates about the interpretability and reliability of large language models, as well as concerns about potential biases and harmful outputs. A paper by Bender et al. (2021) <a href="https://arxiv.org/abs/2108.07258">Link</a> discusses some of these issues and the need for responsible development of language models.
3. The shift towards conversational and virtual experiences: The text suggests that generative AI will enable more conversational and virtual interactions, including virtual companions and avatars. This aligns with the growing interest in areas like conversational AI, virtual assistants, and metaverse technologies. Companies like Anthropic, Replika, and others are exploring these directions, while researchers are studying the potential impacts and ethical considerations <a href="https://arxiv.org/abs/2301.07597">Link</a>.
4. The potential for generative AI to change the nature of work and learning: The text highlights the potential for generative AI to transform work, education, and the way we learn and provide value. This connects to the broader discussion around the future of work and the role of AI in various industries and domains. Reports by organizations like the World Economic Forum <a href="https://www.weforum.org/reports/the-future-of-jobs-report-2020">Link</a> and the OECD <a href="https://www.oecd.org/employment/future-of-work/">Link</a> explore these topics.
(print: "<h3>" + (passage:)'s name + "</h3>")
You might be wondering how we got to this point, where computers are everywhere and do all kinds of amazing things. Well, it all started with a brilliant man named Alan Turing, who realized that computers didn't have to be very complicated to be very powerful. They just had to follow instructions, and those instructions could be flexible enough to handle many different situations and ideas. Of course, there is a lot more to the history of computing, but one thing that Turing was interested in was how people would react to a machine that could take input, instructions, and give answers. Could they perceive this output as being like a human created it? If the answers were correct, as most early computers were good at doing calculations, then the machine would be more useful than a human calculator or computer, //computer// at that time was a job title that referred to staff trained to calculate or operate calculation machinery.
Numbers soon gave way to words, and the instructions became more human-like with languages like Fortran, Basic, C, and so on. These languages were fed into the Turing 'box' and came out as something more useful. Businesses gradually adapted some of their processes to use computers, and before we knew it, we had a world full of machines that did something mysterious called processing, and gave us something useful from whatever prompt we gave them. Learning these languages was hard work, and people who mastered them often created new languages to suit their needs, so we ended up with many languages to talk to the Turing box. We also had libraries, things that helped the processor do things like input, output, printing, video. As time went on, languages and libraries, along with trial and error, helped create mountains of code that ran all kinds of useful processes. We also added the internet, and learned to turn most information into digital form, so we could easily send it back and forth without losing anything. It's funny or sad to remember that most jobs used to involve cataloging, storing, retrieving, archiving and verifying bits and pieces of information. Some of it was clumsy, worthless, or just fleeting like the rolls of thermal fax paper. I have seen the challenge of storing lots of things, of varying value. In banking, for example, even in the chaotic scene of the trading floor with everyone looking at dozens of monitors and Bloomberg terminals, it reminded me of how much effort and skill it took to process and profit from information. And then to keep it, store it, and maybe learn from it? That seemed too hard, but yet we’ve managed to create torrents of facts, opinions, and clickbait. All that’s needed is to turn it into knowledge, or something useful.
Now, with Generative AI, we face an interesting dilemma. For the near future, and maybe for the long term depending on the industry, we won't see much change. People handle information differently, and I have spent years developing and practicing methods to help us work together, faster and better. But it depends on the situation and the leadership of the company. As a person invested in building a career, you will find some work, somewhere. You will learn, hopefully, to be good at it. You will look for improvements in your day to day efforts, as we all get bored or annoyed with dull repetition. Ironically, look at the recent recall of self-driving options in Tesla, not because of the smart technology, but because of the dumb humans who still can do better than the machine in handling unusual situations, if only they were paying attention. You will face all these things in a typical job. And try to pay attention. Unusual cases tends to be the fuel that keeps us engaged. You may try to hand off the tedious stuff to machines.
But, the question is can you hand off the interesting things to machines? That’s what we’ll do our best to explore. As technology may enhance your skills, that could provide impact on the profit or loss of the business. These relationships might not be be clear, but as companies get bigger, staff and roles get more specialized. They become more complex and sometimes hard to understand. You might find a chance to become more specialized. You might seek more education in theories and practice of a more focused area. Your passion will lead you to places of dead ends, but maybe also opportunities.
[[Why AI will change the way we work]]
<h3>Are you sure about that?</h3>
1. "Alan Turing... realized that computers didn't have to be very complicated to be very powerful."
- This statement aligns with Turing's work on the idea of a universal computing machine, as described in his 1936 paper "On Computable Numbers" (<a href="https://www.cs.virginia.edu/~robins/Turing_Paper_1936.pdf">Turing, 1936</a>).
2. "Could they perceive this output as being like a human created it?"
- This refers to Turing's famous "imitation game," later known as the Turing Test, which explored the idea of machine intelligence being indistinguishable from human intelligence (<a href="https://academic.oup.com/mind/article/LIX/236/433/986238">Turing, 1950</a>).
3. "Some of it was clumsy, worthless, or just fleeting like the rolls of thermal fax paper."
- This statement reflects the challenges of managing and storing information before digital technologies became widespread, as discussed in historical accounts of office work and information management (<a href="https://en.wikipedia.org/wiki/Office_automation">Office Automation on Wikipedia</a>).
4. "In banking, for example, even in the chaotic scene of the trading floor with everyone looking at dozens of monitors and Bloomberg terminals, it reminded me of how much effort and skill it took to process and profit from information."
- This observation aligns with descriptions of the fast-paced and information-intensive environment of trading floors, as documented in books like Michael Lewis' "Liar's Poker" (<a href="https://www.goodreads.com/book/show/7865083-liar-s-poker">Liar's Poker by Michael Lewis</a>).
5. "Now, with Generative AI, we face an interesting dilemma."
- This statement reflects the ongoing discussion and debate around the potential impacts of generative AI technologies, as covered in various articles and analyses (<a href="https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-emerging-risks-of-generative-ai">The Emerging Risks of Generative AI by McKinsey</a>).
6. "You may try to hand off the tedious stuff to machines. But, the question is can you hand off the interesting things to machines?"
- This question touches on the ongoing debate about the potential for AI to automate various types of tasks, both routine and complex, as discussed in studies and analyses (<a href="https://www.oxfordmartin.ox.ac.uk/downloads/academic/The_Future_of_Employment.pdf">The Future of Employment by Frey & Osborne</a>).
(print: "<h3>" + (passage:)'s name + "</h3>")
Computers are wonderful devices, aren't they? They can help us with all sorts of tasks, from creating art to solving equations. But they can also be very fussy and demanding. They want us to talk to them in a certain way, depending on what we want them to do. And they don't always understand us, or we them. Learning to use computers is a bit like learning a foreign language. You have to master the basics first, before you can get fancy and creative. A doctor, for example, has to know how the human body works, before he can specialize in brain surgery or podiatry. And he has to use the right words to describe what he sees and does, otherwise he might end up cutting the wrong nerve or toe. The same goes for any profession or field of study. They all have their own jargon and acronyms, which can be very confusing to outsiders. And sometimes, even to insiders.
I practice and coach //design thinking//, which is a way of solving problems by putting yourself in the shoes of the user. It sounds simple enough, but it can be quite challenging in practice. Especially when you have to work with a team of people who have different backgrounds, perspectives, and opinions. They often are challenged to think outside of their own skills and focus. Mucn like the UN, we have to learn each other's language and terminology in order to hopefully translate that to outside users. One of the most common cause for miscommunication is acronyms. Yet, these seem pervasive in all business environments. Its best to avoid using acronyms, because they can be very misleading and ambiguous. For instance, HCI could mean Human-Computer Interaction, or Human-Centered Innovation, or even High-Intensity Cardio. You can see how that could cause some confusion. But they persist. And then they wonder why they can't communicate effectively with each other, or with the computer.
[[How to decide whether or not to continue reading]]
Which brings me to the topic of Generative AI, which is what you are reading right now. Generative AI is a type of artificial intelligence that can create new content, such as text, images, music, or code, based on some input or prompt. It's like having a magic genie that can grant you any wish, as long as you phrase it correctly. And that's the tricky part. How do you phrase your wish correctly, so that the genie understands what you want, and gives you what you want, and not something else?
[[How AI Works Differently Than You Might Think]]
You probably have some experience with this, if you have ever used Google, or Photoshop, or Siri, or any other computer program or device that requires some form of input from you. You have learned, over time, how to tweak and adjust your input, to get the best possible output. You have learned to use keywords, shortcuts, commands, gestures, and voice, to make the computer do what you want it to do. You have learned to adapt to the computer, rather than the other way around.
But what if the computer could adapt to you, instead? What if you could talk to the computer in a natural and conversational way, and it would understand you, and respond to you, and even anticipate your needs and desires? What if you could ask the computer to do anything you can imagine, and it would do it for you, without any fuss or hassle? Wouldn't that be wonderful?
Well, yes and no. It would be wonderful, in theory, but it would also be very dangerous, in practice. Because you might not always know what you are asking for, or what you are getting. And the computer might not always know what you mean, or what you want. And there might be some unintended consequences, or some ethical dilemmas, or some legal issues, or some moral quandaries, that you might not have thought of, or that the computer might not have considered.
[[What AI could impact, pros and cons]]
<footer>
<h3>Are you sure?</h3>
1. The analogy of learning to use computers being like learning a foreign language:
This analogy has been discussed in the field of human-computer interaction (HCI). A paper by Hutchins et al. (1986) <a href="https://www.jstor.org/stable/1576433">Link</a> argued that the language metaphor can be limiting and that a better analogy might be "learning to interact with an alien culture."
2. The potential danger of generative AI not fully understanding user intent:
You raise a valid concern about the potential for misunderstanding or unintended consequences when using generative AI systems. This aligns with ongoing research and debates around the interpretability, safety, and alignment of AI systems with human values and intentions. A paper by Hendrycks et al. (2021) <a href="https://arxiv.org/abs/2107.07366">Link</a> discusses the importance of AI systems being "honest" about their capabilities and limitations.
</footer>
(print: "<h3>" + (passage:)'s name + "</h3>")
Imagine that you work for a big company that makes widgets. You know, those things that do something or other. You don't really care what they do, as long as they sell well and keep you in a job. You have a boss who tells you what to do, and a bunch of colleagues who do more or less the same thing as you. You have a routine, a schedule, a budget, and a deadline. You have meetings, reports, emails, and memos. You have goals, objectives, targets, and indicators. You have a system, and you follow it.
But what if you could change the system? What if you could make your work a bit better, faster, smarter, and more efficient? What if it could get you out of endless check-in's or meetings? We're seeing this change in some major coroporations where they want to remove barriers to experemenation, mainly hierarchy and death to decision making. There was a time for hierarchical, command-and-control organizations–the 19th century, to be exact, when many workers were illiterate, information traveled at a snail’s pace, and strict adherence to rules offered the competitive advantage of reliability. Now the new paradigm is that you can manage even more complexity and efficiency with the help of virtual assistants.
This development has the potential to be a game-changer for organizations and the field of management and organizational behavior. Organizations are systems designed to produce outputs like products, services, profits, and social impacts. The quality of these outputs hinges on the inputs, which include resources, processes, strategies, and cultures. These inputs, in turn, rely on the knowledge, skills, and attitudes possessed by the organization's workforce. The more robust these human capabilities are, the better the inputs and, consequently, the outputs.
However, the critical question is: How do employees acquire and apply the necessary knowledge, skills, and attitudes? How do they learn, communicate, collaborate, and drive innovation within the organization? Ultimately, how do organizations equip their people to achieve goals, fulfill missions, and realize visions?
[[The puzzle of hackathons, or non-work work]]
Traditionally, these questions have been answered by various methods and tools, such as education, training, consulting, coaching, mentoring, brainstorming, problem-solving, planning, and so on. These methods and tools have been useful, but they have also been limited, costly, and time-consuming. They have relied on the availability, accessibility, and quality of information, experts, and facilitators. They have also depended on the motivation, engagement, and participation of the learners, workers, and leaders. They have often been disconnected, fragmented, and inconsistent. They have not always been effective, efficient, or satisfactory.
But now, there is a new way. A way that is enabled by AI, and by LLMs in particular. A way that is based on natural language, and on prompts. A prompt is a simple and concise way of asking the LLM to do something for you, such as summarizing, explaining, suggesting, or creating. A prompt can be a word, a phrase, a sentence, or a paragraph. A prompt can be specific, general, or open-ended. A prompt can be factual, hypothetical, or fictional. A prompt can be anything you want it to be, as long as it is clear, relevant, and appropriate to your knowledge of what the output should achieve. A prompt can be used to access and generate actionable knowledge, which is knowledge that can help you take action, make decisions, and achieve results. Actionable knowledge is not just information, or facts, or data. It is also insights, ideas, solutions, and plans. It is knowledge that is tailored to your goals. It is knowledge that is based on the best available evidence, research, and practice. It is knowledge that is presented in a engaging way, you can even ask for clarification when it's vague. It is knowledge that is ready to use, and easy to apply.
[[The hidden cost of not knowing how hard things are to achieve]]
Actionable knowledge is not something you can get from a book, or a course, or a conference. It is something you can get from a conversation. A conversation with a LLM, that is. It canthat can talk to you like a friend, a colleague, or a mentor. It can understand you, and respond to you, and even anticipate you. It can combine and synthesize knowledge from different sources, domains, and perspectives. It that can offer you multiple options, scenarios, and alternatives. It can help you fail fast, and try more scenarios and possible activities than you may be comfortable with. The question is how does the enterprise embrace this new influence? Or closer to home, how does learning change with this new ability to not just look up the answers, but to skip straight to insight without doing the work to get there?
This is the future that we are going to explore in this book. A future that is not so far away, and that is not so hard to imagine. A future that is already here, and that is waiting for you to join. A future that is powered by AI, and by LLMs. A future that is driven by natural language, and by prompts. A future that is full of possibilities, and of opportunities. A future that is exciting, and that is fun.
Are you ready? Then let's begin.
[[Organizations as systems]]
(print: "<h3>" + (passage:)'s name + "</h3>")
We live in a world where information is abundant and cheap, but knowledge is scarce and expensive. We have access to more data than ever before, but we have less understanding of what it means and how to use it. We have more tools and methods than ever before, but we have less wisdom and insight. We have more experts and authorities than ever before, but we have less trust and confidence.
This is the problem of the modern world. A world that is fragile, complex, and uncertain. A world that is prone to black swans, those rare and unpredictable events that have a huge impact and change everything. A world that is dominated by the narrative fallacy, the tendency to impose a linear and coherent story on a messy and chaotic reality. A world that is fooled by randomness, the illusion that we can explain and control the outcomes of a stochastic and nonlinear process.
But there is a solution. We lack the ability to differentiate reasonable narratives from false ones. There's plenty to suggest false narratives are just more interesting, as Mark Twain said "A lie can travel halfway cross the world while the truth is putting on it's shoes." Our new AI tools can understanding language, recognize images, play games, and suggest decisions. AI has been around for a long time, but it has recently become much more powerful and useful, thanks to the availability of huge amounts of data and powerful computing resources. One of the most impressive achievements of AI is the development of large language models, or LLMs for short. Machine learning is a successful branch, best known for recognizing things, you may use it to authenticate or to see a person gathered in a custom photo album. This is numeric. The machine code cannot 'see' but it looks at numbers that represent that photo, or that tune and is able to match it to that disco tune and quickly issue a 'cease and desist'. These are systems that can process and translate language, such as English, French, or Mandarin, based on a given input or prompt.
Generative AI is the new offshoot, because it still uses the same technique, patterns of numbers, but we recognize them as words. By studing all the words out there (along with many thousand of human time to help train them) they are able to translate human language into something that can be used to do, find, or create something. And it can create something new-ish. It's also great at reading long passages like this and summarizing them. See the bottom of this page. The techninque, now being improved and altered on a seemingly weekly basis can write decent code, build a resume, write a syllabus, create an essay, compose songs, and crack jokes. They can also help you with your work, by providing you with guidance, advice, and plans, based on the latest research and best practices in your field.
[[Ok, so they can learn, how did that lead to being able to hold a conversation?]]
The technique we engage with Gen AI is based on natural language, and on prompts. A prompt is a simple and concise way of asking the LLM to do something for you, such as summarizing, explaining, suggesting, or creating with specific goals and intent. A prompt can be a word, a phrase, a sentence, or a paragraph. A prompt can be specific, general, or open-ended. A prompt can be factual, hypothetical, or fictional. A prompt can be anything you want it to be, as long as it is clear, relevant, and appropriate to your goal. A prompt can be used to access and generate actionable knowledge. Actionable knowledge is not just information, or facts, or data. It is also insights, ideas, solutions, and work. Its common to see marketers build entire campaigns in dozens of different platforms using this capabililty. It is knowledge that is tailored to your needs, preferences, and goals. It is knowledge that is based on the best available evidence, research, and practice. It is knowledge that is presented in a clear, concise, and engaging way. It is knowledge that is ready to use, and easy to apply.
Strangely, Actionable knowledge is not something you can get from a book, or a course, or a conference. While authors may describe the "4hour work week" and give a recipe to achieve it, all within their 19.95 paperback. It doesn't do the work for you. You can get action from a conversation. You may need to give a context, or a timeframe or other vital context to get the right approach. But this conversation doesn't have to be with a famous author or professional. A conversation with a LLM will do. And most of you have experienced this. That conversation can be verbal, as the text-to-speech is eerily close to the fiction of HAL 9000. A LLM that can talk to you like a friend, a colleague, or a mentor. A LLM that can understand you, and respond to you, and even anticipate you. A LLM that can combine and synthesize knowledge from different sources, domains, and perspectives. A LLM that can offer you multiple options, scenarios, and alternatives. A LLM that can support you, challenge you, and inspire you. A LLM that can help you do your work better, faster, smarter, and more efficiently.
[[How AI Works Differently Than You Might Think]]
Let's explore this. What will be the future of organizations, and of management and organizational behavior with this newfound expertise? A future where every member of the organization can have access to a LLM, and to actionable knowledge. A future where every member of the organization can instantly research any topic, or better, it listens and suggests approaches to you and your team.A future where every member of the organization can build and share prompts to train the computer to serve up even better results for your industry. A future where every member of the organization can learn, communicate, collaborate, and innovate, regardless of role and skill level. Will this democratize the current hierarchical structures? Will we all be replaced? It begins with what do we have to work with.
[[Transforming data to wisdom]]
<footer>
<h3>Are you sure about that? </h3>
1. "We live in a world where information is abundant and cheap, but knowledge is scarce and expensive."
- This statement aligns with the concept of the "knowledge economy," where knowledge and its application have become crucial economic drivers <a href="https://www.worldbank.org/en/publication/wdr/wdr-2019" >World Bank on Knowledge Economy</a>
2. "We have access to more data than ever before, but we have less understanding of what it means and how to use it."
- The abundance of data and the challenge of deriving meaningful insights from it is a well-recognized issue, often referred to as the "data-rich but information-poor" paradox <a href="https://www.economist.com/leaders/2010/02/25/data-data-everywhere">Economist article on data overload </a>
3. "A world that is fooled by randomness, the illusion that we can explain and control the outcomes of a stochastic and nonlinear process."
- This statement reflects the ideas explored by Nassim Nicholas Taleb in his book <a href="https://fs.blog/fooled-by-randomness/ Summary of Taleb's book">Fooled by Randomness" </a>
4. "A prompt is a simple and concise way of asking the LLM to do something for you, such as summarizing, explaining, suggesting, or creating."
- This explanation of prompts aligns with the way prompts are used to interact with and guide language models <a href="https://www.anthropic.com/prompting"> Anthropic's guide on prompting </a>
5. "Actionable knowledge is not just information, or facts, or data. It is also insights, ideas, solutions, and plans."
- This definition of actionable knowledge is consistent with the concept's use in various fields, such as management and decision-making (<a href="https://hbr.org/2012/07/making-knowledge-work-actionable">Harvard Business Review article on actionable knowledge </a>
6. "This is the future of organizations, and of management and organizational behavior. A future where every member of the organization can have access to a LLM, and to actionable knowledge."
- This prediction about the future of organizations and the role of LLMs is speculative but aligned with some experts' views on the potential impact of AI on work <a href="https://www.mckinsey.com/featured-insights/future-of-work/ai-and-the-future-of-work"> McKinsey report on AI and the future of work</a>
</footer> (print: "<h3>" + (passage:)'s name + "</h3>")
AI and LLM's are discussed at length, but few understand its implications for how we organize and manage our work. We are at the cusp of a new paradigm in information processing, communication, and automation, thanks to the advances in deep machine learning algorithms. These algorithms can do things that were once considered impossible, such as conversing with humans, turning half baked ideas into an actual picture or representation, or generating texts to advertise. But what does this mean for the complex systems that we call organizations? How will AI change the way we work, collaborate, and innovate? The impact of AI on organizations will be far more profound and radical than the recent pandemic, which was just a temporary shock. AI will not only disrupt the existing processes and practices, but also create new ones that readers will help steer and create, so here's what needs to change.
AI has already shown its superiority in performing standardized tasks in narrow domains, such as data analysis, pattern recognition, or optimization. But now, with the emergence of large language models (LLMs), such as GPT-4, AI can also handle more diverse and open-ended tasks, such as writing, reasoning, or creating. Most have tried, and quite a few have adopted using these tools if not daily, but situationally. Communication is a practice that most are assumed to be competent or excellent in, yet this new technology evens the playing field by making anyone with general knowledge as good as a skilled practitioner. Across most fields, this changes the amount of effort needed by those with experience, and lessens the need for apprentiship. LLMs are not perfect, of course, and they still make mistakes, especially in mathematical problems. But they are good enough to fool us most of the time, and they are getting better every day. This means this upgrade in communication ability will fill a wider range of business applications than before, beyond the traditional domain of data science.
[[Generative AI: A primer]]
But how do we apply AI to organizational processes? How do we redesign them to take advantage of AI's capabilities? How do we ensure that AI does not harm us or our livelihoods? These are not trivial questions, and they require a deep understanding of the nature and characteristics of the processes we want to improve. For example, we need to distinguish between urgent and important tasks, between short-term and long-term goals, between collaborative and competitive environments, between routine and creative activities, and so on. We also need to consider the human factors, such as the skills, preferences, emotions, and motivations of the people involved in the processes. AI can help us with decision-making, but it can also cause stress, fatigue, or alienation. We need to find the right balance between human and machine intelligence, between control and autonomy, between efficiency and innovation.
Tools are evolving but it's safe to say it can either be our servant or our master, depending on how we use it and interact with it. This book tries to emphasize a symbiotic relationship, where we benefit from AI's strengths and compensate for its weaknesses, and vice versa. Unions may be mentioned, as humanness is antithetical to the rapid onslaught of computing power. We should not let AI take our livelihoods, but do we actually outperform these algorithms? In many ways, no. But could they help us to communicate more effectively, form more collaborative groups and better our outcomes? That is possible, but will take change, which has never been a populist goal. Without knowledge or deep understaning the natural language interfaces to LLMs alow us to access the power of deep learning, at least as far as the training material allows. Can we become better at communicating with other humans? I believe the answer is yes. I will explore the many aspects and implications of this vision if you choose to follow them. Our tracks are data, understanding the basis of training and tokens. Prompts, and prompt architecture, how can experts share their methods for approaching problems. Daily the future seems to be predicted, but our choices and actions will shape how this tool is used or misused.
[[So, how do you manage people using data?]]
Workplaces are not static entities, but rather dynamic and adaptive systems that reflect the interplay between the technological and social forces. As the logician Bertram Russell said "Work is two kinds - either moving things around physically, or telling someone else to do that. The latter is better paid and more enjoyable." Machines have been able to move things around quite well for some time. Most of us are in the job of telling it what to do. That's where LLM's come in to be quite a bit better at telling us what to do, or think, or believe. We are in the midst of a transition period, where the old paradigms and practices are being challenged and replaced by the new ones. We are not sure where we are heading, but we can sense the direction and the magnitude of the change. For example, virtual reality, which seemed to be the next big thing, has failed to deliver on its promises and remains a niche market. On the other hand, conversational technologies, such as chatbots, voice assistants, or smart speakers, have taken the world by storm and are becoming ubiquitous in our personal and professional lives. The change is not linear or gradual, but rather nonlinear and abrupt. That is in situations where we have personal choice, when we come together as groups, in social or employment situations, the change isn't as clear, and thats where we start to fear that the change could be harmful.
What will be the new standards or norms remains to be seen. We think new work cultures will center around LLMs acting as companions, trusted advisors, or managers. Replacing bosses will be a big theme, as the// telling people what to do// skillset is way too often poorly executed. It is probably not far-fetched to assume that many organizational processes will happen between intelligent agents equipped with goals and feedback loops. In contrast, humans will devote much time to sharing their knowledge, personalities, and goals to have LLMs handle the hard part of individual communication. It's no secret Jack and I are educators, so this shift comes at the right time, as we've devoted decades to understanding and practicing how best to engage people to learn these subjects. Whether AI or HCI design, both intersect in this tool. Right now, this process is called prompt engineering or LLM finetuning, but the unassuming names do not do justice to the importance of the process. Time will tell what kind of description is used. Still, instructing private and corporate AI systems will be a significant task in many occupations.
[[Ok, but how do I keep AI away from my job??]]
<footer>
<h3>Are you sure?</h3>
1. The capabilities of large language models (LLMs) and their impact on various tasks:
- "On the Opportunities and Risks of Foundation Models" by Rishi Bommasani et al. (<a href="https://arxiv.org/abs/2108.07258">Research paper</a>)
2. The role of AI in communication and collaboration:
- "Collaborating with AI: An Instruction Manual for Humans" by Cathy Hackl (<a href="https://www.amazon.com/Collaborating-AI-Instruction-Manual-Humans/dp/1523097884">Book</a>)
- "Communicating with Machines: The Rising Power of Language Models" by Kai-Fu Lee (<a href="https://www.wired.com/story/opinion-communicating-with-machines-the-rising-power-of-language-models/">Article</a>)
3. The potential impact of AI on employment and the workforce:
- "The Future of Employment: How Susceptible Are Jobs to Computerisation?" by Carl Benedikt Frey and Michael A. Osborne (<a href="https://www.oxfordmartin.ox.ac.uk/downloads/academic/The_Future_of_Employment.pdf">Research paper</a>)
- "Will AI Cause Mass Unemployment?" by Daron Acemoglu and Pascual Restrepo (<a href="https://www.nber.org/papers/w28257">Research paper</a>)
</footer>
(print: "<h3>" + (passage:)'s name + "</h3>")
<img src="data:image/png;base64,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" alt="DPKNW pyramid" class="container" style="display: flex; align-items: center"/>
The pyramid model is a common way of describing how humans process information and make decisions. It assumes that there is a scarcity of higher-level information, such as knowledge and wisdom, compared to lower-level information, such as data and information. It also implies that there are philosophical and practical challenges in choosing the best course of action when decision making lacks clear signals. A good example is a supermarket. With some clever organization principles, most can pick from hundreds or thousands of choices. Yet, the clarity of which choice is better, or more advantageous is cleverly obscured. Is it a name brand? Is it a pretty package? Choice is possible, but our abilites are easily clouded. The top of the pyramid is wisdom, or trust. Emotions are the key to every level. Data is fairly emotionless and cold, while information is plentiful but noisy. Each level is personal to how we use emotions to make decisions.To make it more confounding, people we may trust or see as knowledgable can fool themselves into communicating that something is significant or meaningful in a given context. For example, we can use data and information to support a decision or a goal for ourselves or our organizations, or trust someone else to do so for us.
[[What tiers of probabilities are you looking for?]]
The levels also describe the value in transformation. Data becomes pattern. Pattern becomes information. The Pattern step is non-traditional, but fills a meaningful place in the concept of design or how computers have aided in the transformation. It's fine to have thousands of hours of footage of groups of people, but more interesting if the computer can find the face of a celebrity, or criminal. For narrative purposes, this pattern matching has always been done by humans. And many enjoy this endless, somewhat tedious use of their cognitave ability. It seems easy to say that both in display and in process, computers are well able to organize and recognize patterns as well as humans.
It also could represent work and jobs. Many have to work at a lower level to support those higher up. Moving up the pyramid is metaphorically like a ladder, or any other concept of how things 'work' when it comes to gaining a position more powerful than others. While we will debate this, its certainly the reason you are reading this. You want an edge. You want to be hierarchically above some people or processes. But how do you get there? And is this now a reasonable model for people?
Its clear that amassing data has been our biggest revolution. Moving from rare and expensive to cheap and ubiquitous, the sensors and data captured in a smartwatch speaks volumes to how much has changed. Data on heart behavior alone would have been prohibitively expensive to gather and trend. Since we've coupled our identities with much of our online behavior, that has created mountains of behavioral and affinity data points used for years to target advertising and deliver spam messages. Census and income data also have been published publically, leading to data mining oppotunities to undestand and benefit communities.
Different algorithms, or really just ways to count and weight one thing over another in relevance, have shown great abilities to find needles in the haystack of these data lakes. Medicine, law, the stock markets could probably be run automatically by these algorithms if it wasn't for humans puzzling ability to do very unexpected things. Also we can all agree to break rules or norms that math doesn't like - Wall Street Bets style anomoies will always keep some things from running rationally. Greed and fear are not really forte's of algorithms.
If you have enough data the concept goes, you can note more simllarities or patterns, then use that to give context to those concepts, which I label as information. Since there's too much of this, as we joked earlier, you need things like ratings, press, pundits, or whatever to filter and pronounce something as worth knowing. Finally wisdom is not that dissimilar to trust. It need not be spoken from a mountaintop, but wisdom describes some collective normalcy we can find comfort and reliability in following.
The DIKW pyramid is a quantity of sorts, or a map. of the pyramid model that stands for Data, Information, Knowledge, and Wisdom. It is often used to illustrate the power of AI or ML, which are technologies that can help us transform data into information, information into knowledge, and knowledge into wisdom. AI or ML can also help us automate or optimize some of the steps in the pyramid, such as data collection, analysis, or pattern recognition. Ideally, businesses follow the DIKW pyramid when they manage their decision making, as shown by the following hypothetical example:
<blockquote>
ABC Energy Company is a company that produces and sells energy from various sources. It wants to make a strategic investment decision that will affect its future performance and competitiveness. It follows the DIKW pyramid as follows:
Data: The company collects data related to energy production, consumption, market trends, regulatory changes, and environmental factors. This data includes information on energy demand patterns, electricity prices, government policies, and the company’s financial performance. The company uses various sources and methods to collect the data, such as sensors, surveys, reports, or web scraping.
Pattern: The company highlights the data that is perceived to influence the outcome, discarding or cleansing the data that tells conflicting or irrelevant stories. The company uses various techniques to filter, aggregate, or validate the data, such as algorithms, statistics, or machine learning. The company also uses visualization tools to display the data in graphs, charts, or maps.
Information: After analyzing the data, the company gains valuable insights. It discovers that renewable energy sources are becoming more cost-competitive due to advancements in technology and government incentives. For example, the cost of solar photovoltaic modules has dropped by 89% since 2010. The company also finds that public sentiment is shifting towards sustainability and cleaner energy solutions. For example, 77% of Americans say that developing alternative energy sources should be a priority over expanding fossil fuels. Additionally, the company observes that fossil fuel prices have been fluctuating, making it harder to predict long-term profitability. For example, the price of Brent crude oil has ranged from $19.33 to $79.48 per barrel in 2020 and 2023.
Knowledge: Based on the information gathered, the company builds knowledge about the energy landscape. It understands that investing in renewable energy infrastructure can lead to long-term cost savings, increased public support, and improved brand reputation. It also acknowledges that relying solely on fossil fuels may expose it to market volatility and potential environmental risks. The company uses various tools to synthesize, interpret, or apply the information, such as models, frameworks, or simulations.
Wisdom: Applying wisdom to the decision-making process, the company decides on the following strategies:
Wisdom: The company will diversify its energy portfolio by investing in renewable energy sources such as solar and wind. This will allow it to hedge against future fossil fuel price fluctuations and align with the growing demand for sustainable energy solutions.
Research and Development: The company will invest in research and development to improve its renewable energy technologies and explore new opportunities. This will allow it to stay ahead of the competition and enhance its innovation capabilities.
Transition: The company will gradually phase out its fossil fuel assets and operations, while ensuring a smooth and fair transition for its employees, customers, and stakeholders. This will allow it to reduce its environmental impact and comply with the regulatory and social expectations.
</blockquote>
So this may be taught in business schools. And narratively it sounds pretty good, except no company ever follows the path of data. Our major tenent in this book is to show what could happen if you do! To be unkind, they spend most of their time in ''knowledge'' - hiring McKinsey and such to figure out what is the best way to make more whatever. If the trends are to be believed in whatever industry, why haven't you invested in them already? Also, what were you thinking was going to happen? Institutions seem to lack such ability to look at themselves and how they constantly are blindsided by fairly predictable changes. Let's see how
[[How AI Works Differently Than You Might Think]]
<footer>
<h3>Are you sure?</h3>
1. The DIKW pyramid model and its applications:
- "Revisiting the Data–Information–Knowledge–Wisdom Hierarchy: A Critical Systems Thinking Perspective" by Goran D. Matic and Pawan Lingras
- "The Data-Information-Knowledge-Wisdom Hierarchy and its Antithesis" by Richard J. Brachman and Thomas Muller
2. Decision-making processes and the role of data, information, and knowledge:
- "Data-Informed Decision Making: Leveraging Data to Transform Organizational Performance" by John Woodill and Brenda Kerton
- "The Role of Data and Information in Decision-Making" by Steven Tadelis
3. AI and machine learning for data analysis and pattern recognition:
- "Artificial Intelligence and Machine Learning for Data Science" by Amit Mitra et al.
- "Deep Learning for Pattern Recognition" by Xiao-Xiao Yin et al.
4. Challenges and limitations of AI and machine learning:
- "Artificial Intelligence and Machine Learning: Driving Tangible Value for Business" by Carla Briese and Ravi Mohan
- "Limitations and Risks of Machine Learning Applied to Perception" by Matthew Wicker and David Sousa
</footer>
(print: "<h3>" + (passage:)'s name + "</h3>")
<img src="data:image/png;base64,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" alt="DPKNW pyramid" class="container" style="display: flex; align-items: center"/>
The pyramid concept of adding value to data into more useful hierarchies is a way of showing how humans process information to make decisions. It suggests that higher-level information, such as knowledge and wisdom, are harder to find and more valuable than lower-level information, such as data and information. It also shows that there are many challenges and choices in deciding what to do in different situations, which humans have been trying to solve for a long time. The pyramid model works well when we can use data and information to show or decide that something is important or useful in a certain context. For example, we can use data and information to help us make a decision or achieve a goal for ourselves or our organizations.
The pyramid concept also shows the value of transformation. Data becomes pattern. Pattern becomes information. The pattern step is not very common, but it is important in the idea of design or how computers can help us change things. It is not very exciting to have thousands of hours of videos of groups of people, but it is more interesting if the computer can find the face of a famous person, or a criminal. For storytelling purposes, this pattern finding has always been done by humans. And many people like this endless, somewhat boring use of their cognitive ability. It's certainly been used to train computers in how to manage ambiguity or error in recognition. It seems as technology and computing power and underlying modeling, computers organize and recognize patterns as well as humans.
The pyramid model could also represent work and jobs. Many people have to work at a lower level to support efforts higher up. Or that people like labor when it serves a higher concept, equality or social change. Moving up the pyramid is like a ladder, or any other idea of how things// work //when it comes to getting a position more powerful than others. While we will argue about this, it is probably the reason you are reading this. You want an advantage. You want to be above some people or processes in the hierarchy. But how do you get there? And is this a good model for people to use to understand their role?
It is clear that collecting data has been our biggest breakthrough. Moving from rare and expensive to cheap and everywhere, the sensors and data collected by a smartwatch show how much things have changed. Data on heart activity alone would only a few years ago been very expensive to get, not to mention analyze. And then trend to see if some activity or diet had any effect. Since we have linked our identities with much of our online behavior, that has created a lot of data points about our actions and preferences that have been used for years to target ads and group us into useful segments. Census and income data have also been made public, leading to data mining opportunities to understand and help communities choose and assess different models of action.
[[Have we always been so obsessed with data?]]
Different algorithms, or just ways to count and compare one thing over another in importance, have been very good at finding important things in the huge amounts of data. Medicine, law, the stock markets could probably be run by these algorithms if humans did not have the strange ability to do very unexpected things. Also we can all agree to break rules or standards that math does not like - Wall Street Bets style surprises will always keep some things from running logically. Greed and fear are not really strengths of algorithms, but those concepts can be taught to a computer model to make financial decisions.
If you have enough data, the idea goes, you can see more similarities or patterns. You then use that axiom to give meaning to those concepts, which we call information. Information relies on context, a particularly human concept. As we may discuss an event or an idea, it may hold a high context to me, and little to you. Therefore managing context helps information be useful. Since there is too much of this, as we joked earlier, you need things like ratings, news, experts, social status or other ranking concepts to filter and say what is worth knowing. Finally, wisdom is not very different from trust. It does not have to be said from a high place, but wisdom describes some common sense that we can trust and follow.
[[How business thinks it works!]]
<footer>
<h3>Are you sure?</h3>
1. The DIKW pyramid concept and its applications:
- "Revisiting the Data–Information–Knowledge–Wisdom Hierarchy: A Critical Systems Thinking Perspective" by Goran D. Matic and Pawan Lingras )
- "The Data-Information-Knowledge-Wisdom Hierarchy and its Antithesis" by Richard J. Brachman and Thomas Muller
2. The value of data and information transformation:
- "Data Transformation and Information Processing" by David M. Andrews
- "The Value of Data: Understanding the Benefits of Data for Society" by Yakowitz, J.
3. The role of data and information in decision-making:
- "Data-Informed Decision Making: Leveraging Data to Transform Organizational Performance" by John Woodill and Brenda Kerton
- "The Role of Data and Information in Decision-Making" by Steven Tadelis
4. Work hierarchies and the concept of "moving up the pyramid":
- "The Pyramid Principle: Logic in Writing and Thinking" by Barbara Minto
- "Hierarchies in Organizations: A Comparative Analysis of Work and Work Control" by Harry C. Katz (<a href="https://academic.oup.com/sf/article-abstract/63/2/329/1663940">Research Paper</a>)
</footer>(print: "<h3>" + (passage:)'s name + "</h3>")
On the surface, this book focuses on the AI-driven design of fundamental organizational processes like planning meetings, documenting projects, or scheduling meetings with the right team of people at appropriate times.
However, we are really discussing a more profound domain. We are proposing a new way to manage the day to day of being a professional. We want to address deep problems associated with typical organizational behaviors that lead to reduced productivity, lack of trust, problematic knowledge management, lack of training in areas pivotal to the business, or lack of innovation. If you are considering changing your organization or learning how to design business processes better, this book is for you.
Business Leaders
This book will be your guide for designing processes assisted by generative AI utilizing Large Language Models, data analytics, and other areas of AI - not limited to Chat GPT or similar technologies. Organizations are complex systems that need many interrelated workflows and information flows to work well. Every time there is a breakdown, we look at cascading effects and negative feedback loops spreading throughout the organization. We will go into details of several key business processes, like team meeting management, and give concrete examples of designing applications and platforms that support representative use cases. We will cover a broad spectrum of details, including best business practices, behavioral science, team dynamics, and technology-based solutions. We will also propose ways of validating our solutions with users.
[[How business thinks it works!]]
Managers
This book will describe AI-driven tools that will get you through a typically busy day at the office. We will help you prepare for important meetings with your teams. Our use cases will provide examples of how to work with generative algorithms to write an optimal meeting agenda, prepare appropriate materials, identify issues by synthesizing messages, and so on. We will provide examples of how to run a better meeting with our apps. As you can see, we are addressing problems that not only affect individual business processes like meetings or appointments. We are proposing a platform that can change any business process that is knowledge-intensive, requires focused prep work, and search for information that might is an "unknown unknown" as well as "known unknowns." Finally, our platform helps people to interact better by using AI to lighten the mood and create a culture of politeness and mutual understanding.
[[So, we use this to make money somehow?]]
(UX) Designers
This book will allow you to explore how to change how people work and interact in a typical business organization. Our use cases will show you how to build tremendous momentum based on increased engagement and interfaces that promote innovation and dynamic interactions. We believe that AI-driven interfaces to pretty much every business process will revolutionize organizations. Our objective is to meet human needs, which are abundantly stated in business literature, and those implicit needs that are often small and mundane. Designing systems with generative algorithms is a new design proficiency and not trivial. This book will take you on a guided trip to understanding the new reality and learning how to design for the people that run and manage typical businesses.
[[Transforming data to wisdom]]
Health Care
You like the diagnosis and treatment part of your role, but, at least in America, we're require lots of data from health providers. Little of which has to do with the outcome or health of their patients. Having provided tools for health can, AI can help automate tedious and error-prone tasks, such as clinical documentation, billing, and coding. It can also enhance diagnosis and treatment by analyzing medical records, lab results, images, and previous interventions to identify problems and suggest solutions. Furthermore, it can accelerate drug discovery and development by finding new targets, designing novel molecules, and predicting clinical outcomes4. Additionally, it can improve health care access and equity by generating personalized and contextualized health information and guidance for patients and providers.
[[How AI Works Differently Than You Might Think]]
Software Engineers
A great framework for most creative efforts, but also for coding is what is called Pair Programming. Originally it was intended to get code to be more logical and readable by geting multiple points of view on a routine or process. Already, generative AI has shown to cut down most repetitive tasks, and can do a basically good job at many time consumning and expensive engineering tasks. Additionally, it can increase productivity and quality by reducing errors, bugs, and technical debt. However, generative AI also poses some challenges and risks, such as data security, ethical issues, regulatory uncertainty, and human oversight. Computers have a difficuilty with the trust problem. Its hard to know when they are operating on poor credentials. Generative AI has the potential to transform software engineering, but it also requires careful management and evaluation.
[[The hidden cost of not knowing how hard things are to achieve]]
Architects and Generalists
These roles are going to benefit the practice, but the roles are still focused on creating a marriage of outcome, functions, and methodologies. I call it a generalist since a movie director has to interpret the script, work with actors, frame scenes. Understand the role of sound and editing. Since a defecit in one of these areas isn't unusual, still, it helps to have some appreciation or taste in those areas in order to staff and hire the people that can realize the vision.
[[Who should read this book?]]
(print: "<h3>" + (passage:)'s name + "</h3>")
I learned to start a game of chess with the move pawn to king’s 4. Years later, I realized that there was no special reason for this move, except that I never tried anything else consistently. This sort of choice experimentation is rare for people. We usually stick to what works for us. In games, in hobbies, in work. Challenges are interesting, but often unwelcome. Daniel Kahneman proposed that we have two modes of thinking: one is fast and intuitive, the other is slow and rational. Thinking slow takes effort, so we don’t do it often. We rely on our ‘gut’ feeling, or on what has worked before. We may believe that a butterfly’s wing can cause a storm, but we don’t apply this to the hundreds of choices we make every day. We don’t even see them as choices, but as habits. This poses a challenge for behavioral psychology: how to change our instincts. How to show us that our choices matter.
When we work in groups, we try to have meetings that allow us to collect enough information or opinions to make a good decision together. But this rarely happens. We often default to hierarchy or status to guide whether we are making progress or not. To make the right choices, you need context. Context and a brief summary of what you are deciding on. It could be supported by facts and figures, but we seldom ask for or review them, because of Kahneman’s system one. It takes effort. What if there was a world where you could experiment with different efforts and outcomes, and use real data or experiments to see how your choices affect them. That is our goal in this book.
[[The hidden cost of not knowing how hard things are to achieve]]
In chess, there are many possible moves that can be evaluated by practice and experience to create winning strategies. The methods of evaluating these moves were among the first ways we taught computers. Computer games like Zork were similar, except you were pitted against the author to win. While there seem to be many options in other narrative-based computer games or in CYOA books, it is still hard to find satisfying outcomes from all the choices a person could make. Experience comes from good and bad choices. But finding the best choice at the time needed can be hard or impossible. While brute force allowed chess programs to beat human skill, Go was conquered by a more subtle AI.
This new technique made the program play against itself. It explored the many possible moves of each turn and kept track of them, assigning a numerical value to each one. All of this was done without supervision, except for some tech experts who set up the learning parameters. As this technique defeated the best human players, it became clear that, not surprisingly, given enough chances and ways to mark each path, computers could learn. Machine learning is a friendlier term than AI, because we attach an emotional or social meaning to intelligence. It makes us uneasy to think that what we value can be copied by silicon.
[[Ok, so they can learn, how did that lead to being able to hold a conversation?]]
The game of go was not the end of the challenge. The next frontier was human language, a game of its own kind. This required a different approach from the machine, one that involved interacting with other humans. Thousands of human workers, similar to mechanical turks, had to train the machine to distinguish between signal and noise. While most humans could easily spot nonsense, the rules of how to give accurate answers, even with large datasets like all of literature, were not simple. Biases in opinion are familiar to most adults, and children have some intuition of what sources are more reliable.
The ability to manipulate language, images, speech and narrative poses a dilemma for many of us who use those skills for fun or profit, just like the master chess players. Did we waste our time and effort to develop those skills? Was the game itself the goal or a means to express our personality or aspirations? As algorithms augment or replace our skills and abilities, do we benefit or lose? There are many stories and possibilities to this puzzle. Some paths may lead to dead ends. Our aim is to guide you forward, and let the machines handle some of the guesswork to make the journey easier.
[[Deep Organizations with Shallow Computation]]
<h3>Are you sure?</h3>
1. The idea of two modes of thinking, "fast and intuitive" (System 1) and "slow and rational" (System 2), is from Daniel Kahneman's book "Thinking, Fast and Slow" (<a href="https://www.penguinrandomhouse.com/books/247032/thinking-fast-and-slow-by-daniel-kahneman/">book link</a>). This concept is well-supported by research in cognitive psychology.
2. The text mentions that "we don't even see [our choices] as choices, but as habits." This is consistent with the concept of "choice architecture" and "nudge theory" (<a href="https://www.nudgetheory.org/">nudge theory link</a>) proposed by Richard Thaler and Cass Sunstein, which suggests that our choices are heavily influenced by the way options are presented to us.
3. The idea that machine learning algorithms can outperform humans in certain tasks, such as playing chess and Go, is well-documented. The AlphaGo program (<a href="https://deepmind.com/research/case-studies/alphago-the-story-so-far">AlphaGo link</a>) developed by DeepMind is a notable example of this.
4. The text mentions the use of "human workers, similar to mechanical turks" to train machine learning models. This refers to the practice of crowdsourcing data labeling tasks to human workers, which is a common approach in developing large language models (<a href="https://www.microsoft.com/en-us/research/project/reality-data/">Reality Data project link</a>).
However, some of the claims made in the text are not well-supported or are debatable:
1. The statement "While most humans could easily spot nonsense, the rules of how to give accurate answers, even with large datasets like all of literature, were not simple" is not entirely accurate. While large language models can generate coherent and sometimes convincing text, they can also produce nonsensical or factually incorrect output (<a href="https://www.theatlantic.com/technology/archive/2022/11/chatgpt-ai-writing-risks/672339/">article on ChatGPT's limitations</a>).
(print: "<h3>" + (passage:)'s name + "</h3>")
Hard to look anywhere without mention of AI and its potential for our organizations. We may have some vague ideas of what this //intelligence// word means from sci-fi books and movies, but we also have some questions: “What can generative AI do for us?” and “How is generative AI different from other types of AI?” Before we dive into the technical aspects of AI, such as NLP and neural networks, we need to consider some organizational basics. These include data literacy, data strategy, leadership support, and innovation culture. These are not new concepts, but they are essential for generative AI. If unfamilar, read on.
If generative AI works well in your organization, it will likely transform it into an AI-centric one, with many implications for machine-to-machine and human-to-AI interactions.
Many people discuss AI readiness from a technical point of view. However, the technology, with its frameworks and patterns, may be one of the easier parts of the puzzle. What is often overlooked is the human perspective. We talk about changing the culture, but we forget how the current technology shapes it. The tools we use every day have not changed much since the web browser, text editors, or spreadsheets were invented. The new technologies are on a different level. A top-down approach, where we try to implement as many AI use cases as we can, may not work well, because we don’t know what the final product will look like.
In the beginning there was Google. To be fair, there was Yahoo, Alta Vista, Ask Jeeves and perhaps dozens of similar competitors to how to manage the task of finding something in the copius and spectacularly unorganized amount of intormation being published on a second by second basis on the World Wide Web. Many of us, myself included, were intrigued enough to develop techniques to try and organize and share the small amount of information about our business or topic acting as //webmasters// or some such made up term that tried to pull together all the different disciplines together to get stuff done. And it worked, for quite some time. We increased the amount of text and images, later movies and sound exponentially. It took some clever math such as the page rank algorithm that was Google's claim to fame. This takes the data from a page, and simplisticly attaches numbers to the words. This //weight// can also be dynamic. Unlike data warehousing that uses brute force to calculate all the dimensions of a set of numbers to allow quick cross referencing and real time exploration. But in our efforts, we created databases of pages and text, but rarely tried to organize what it was about at a high level, or even what importance it has to various topics. To this day, searching within a website other than google is normally unhelpful. Without topics and weight, it's tough to tell one thing from another, certainly not in the attention span that kept getting shorter and shorter with more and more competition.
We left it to these big players to make sense out of the// big picture//. The feed took over when mobile devices became the default for many of how they got their information. The size of these devices left little room to browsing or finding things. Search became the default, and many skipped to google to search their own content. Inside the walled garden, more companies began to trust vital data to the cloud. This also came with lots of weighting, math, and time to make sure users could work anywhere and keep things magically in sync. Organizatoin was key, each database became richer and richer but needed more and more sophisticated ways to tag and weight the contents. We added popularity in the form of sharing and liking which made some sites like Amazon the central repository for customer feedback. We then were asked to pay for higher placement, so there is a virtuous circle for consumers. More is spent to get accurate, fast results that satisfy their needs. Over time Google for one offered faceted results, images, products, later just answering questions about the topic. We've been conditiioned to pay the small price of advertisments, but even these were designed to generate better results by affinity to the user goals. All of this was fueled by machine learning, the ability to quickly compute graphs or matrices of concepts to sort and present results. ML for short.
[[cloud architectures driven by ai]]
Machine learning had triumphed over games, allowing it to program itself to master complex games like Go. It's ability to focus on a goal, winning, then play itself to discover what moves in what context were more effective than others in a matter of months creating moves that humans had never concieved. It was able to use neural networks, a short series of weighted decisions computed on the fly to solve problems like face recognition and other heavy-duty tasks that humans felt were their forte and were not to be easily bested. But then again, once you have a way to recognize a face, then why not build one? Reversing the technique to generate created this new paradigm. It could use the rules it learned to create new things. As we tend to recognize way more often than we invent, it seemed another shift, where the world of websites, literature, corporate intranets, research, youtube or whatever could be reverse engineered. This set of mathematical steps suddenly seemed more than human. And its techniques are very similar to human.
As with all new technologies, the name of the service, the way you engage with the service can often be //kleenexed// or you lump companies brand names with the service they provide. For those companies like Google, that leads to more customers, less reliance on having to promote your brand. Also having to change your name to Alphabet in order to do more stuff. In the explosion of AI related companies, services, we also have to deal with names and brands. Chat GPT by Open AI may be the most famous for text and threaded conversation based interaction. But as of this writing there are several more: Here’s a quick rundown of some of the services of note, and why they may be worth remembering as the landscape becomes more cluttered.
[[How AI Works Differently Than You Might Think]]
<footer>
<h3>Are you sure?</h3>
1. "What is often overlooked is the human perspective. We talk about changing the culture, but we forget how the current technology shapes it." A study by Gressgård et al. (2014) <a href="https://www.inderscienceonline.com/doi/abs/10.1504/IJKMS.2018.096313">Link</a> discusses how organizational culture and technology are co-constitutive, and the introduction of new technologies can reshape organizational cultures.
2. While a top-down approach may have limitations, some researchers argue that a structured, systematic approach to AI implementation is necessary for organizations. A paper by Davenport and Ronanki (2018) <a href="https://hbr.org/2018/01/artificial-intelligence-for-the-real-world">Link</a> suggests developing an "AI strategy" to guide the adoption and integration of AI within an organization.
3. The dominance of large tech companies like Google in shaping the organization and access to online information is well-documented. A report by the Pew Research Center <a href="https://www.pewresearch.org/internet/2021/06/16/1-worries-about-developments-in-ai/">Link</a> highlights how a majority of Americans rely on a few major tech companies for online information.
4. The success of AlphaGo, the AI system developed by DeepMind, in defeating human champions at the complex game of Go is widely cited as a significant milestone in the development of AI. A paper by Silver et al. (2016) <a href="https://www.nature.com/articles/nature16961">Link</a> describes the technical details of AlphaGo's neural network architecture and self-learning capabilities.
</footer>
(print: "<h3>" + (passage:)'s name + "</h3>")
The rise of cloud computing has been pivotal in unleashing the potential of machine learning and generative AI models. At its core, the cloud provides access to an abundance of computing power and data storage on an unprecedented scale., A major advantage of cloud computing services is the ability to scale resources almost unboundedly. More space at the same or less cost over time has atracted most businesses to no longer attempt this within their own resources. Yet, in my experience, that Dell machine in the corner could be running your critical software. The history of business adoption of computing comes with lots of solutions, many versions, many languages. All of this doesn't modernize overnight. So if you're just adopting SaaS (Software as a service, like Monday.com for project management, or Mural.co for whiteboarding) don't feel too bad. Change is hard.
Yet, managing this scalability has been critical for training large neural networks on vast datasets - a computing-intensive process that was previously limited to major tech giants with in-house data centers. Pioneers like Amazon, with its Amazon Web Services (AWS) platform launched in 2006, democratized access to these resources, allowing researchers and startups to rent computing power on-demand. Machine learning thrives on data, and the cloud's virtually limitless storage capacity enables the hoarding of information on a planetary scale. Platforms like Google Cloud provide seamless data ingestion pipelines, allowing models to be continuously retrained as new data flows in. This feedback loop of perpetual learning lies at the heart of generative AI systems like OpenAI's DALL-E 2 and GPT-3.
The latter, a large language model trained on a staggering 175 billion parameters using Google's powerful Tensor Processing Units, represents a landmark in the field. By ingesting a significant portion of the internet's textual data, GPT-3 developed an uncanny ability to generate human-like text on virtually any topic in late 2020. Extensive guidance by humans to help it manage the inherent contradiction in our commmunication. As the old joke goes it's hard to tell on the internet that you're a dog. Or biased in every number of ways we may not normally encounter. Most early reports of using Microsoft's first rollouts were rife with provoked halucinations as they have been called. This may be a poor analogy, certainly like when we are unsure of an answer, we make one up and see if it works. Because of the expectation that computers are programmed or otherwise sanitized in their ability to do anythnig creative, it was a quick read that helped many put it aside as another fad. But these things improve exponentially.
As to more inherent biases about gender or behavior, it's likely in the eye of the beholder. If you have ever trusted implicitly what you read or hear, I doubt you would be reading this book. Knowledge comes with some social proof, and trust is built over time or requires at least celebrity spokespeople! However, as Bender et al. cautioned, such models can inadvertently amplify biases present in their training data, reflecting "internet-scale defects."As generative AI systems become more ubiquitous, deployed via cloud services, governing their outputs responsibly looms as a significant challenge. Ultimately, the cloud's elasticity and economies of scale provide a computational canvas for machine learning's masterpieces to take shape. Whether generative AI will prove to be a revolutionary artistic medium or a subversive force remains to be seen. But the cloud's role as an enabling architecture is indisputable, its impact reverberating across industries.
[[Why is data architecture so hard?]]
<Are you sure?</a>
"A major advantage of cloud computing services is the ability to scale resources almost unboundedly."
Source: Armbrust, M., Fox, A., Griffith, R., Joseph, A. D., Katz, R., Konwinski, A., ... & Zaharia, M. (2010). A view of cloud computing.
<a href="https://dl.acm.org/doi/10.1145/1721654.1721672">Communications of the ACM, 53(4), 50-58.</a>
"Pioneers like Amazon, with its Amazon Web Services (AWS) platform launched in 2006, democratized access to these resources, allowing researchers and startups to rent computing power on-demand."
Source: Barr, J. (2015, July 2).
<a href="https://aws.amazon.com/blogs/enterprise-strategy/the-journey-toward-cloud-first-the-stages-of-adoption/"> AWS Cloud Journey. Amazon Web Services.</a>
"Platforms like Google Cloud provide seamless data ingestion pipelines, allowing models to be continuously retrained as new data flows in."
Source: Google Cloud. (2022).
<a href="https://cloud.google.com/architecture/ml-on-gcp">Continuous Machine Learning on Google Cloud.</a>
"GPT-3, a large language model trained on a staggering 175 billion parameters using Google's powerful Tensor Processing Units, represents a landmark in the field."
Source: Etzioni, O. (2022).
<a href="https://hdsr.mitpress.mit.edu/pub/w135kqy7">GPT-3: What's All the Fuss About? The Harvard Data Science Review.</a>
"However, as Bender et al. cautioned, such models can inadvertently amplify biases present in their training data, reflecting 'internet-scale defects.'"
Source: Bender, E. M., Gebru, T., McMillan-Major, A., & Shmitchell, S. (2021). On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?. Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, 610-623.
<a href="https://dl.acm.org/doi/10.1145/3442188.3445922">2021 ACM Conference on Fairness, Accountability, and Transparency, 610-623.</a>
(print: "<h3>" + (passage:)'s name + "</h3>")
In software engineering, the problem of building and maintaining systems is addressed through more iterative local human-in-the-loop review (system analysis). GitHub and GitLab are two common examples where insight can be had. Software engineers can directly subscribe to the repos they care about, ensuring that any changes are, at minimum, communicated to the teams who care the most.
In data, this problem is far more challenging to solve for several reasons:
1. Most data consumers don't care about looking directly at the source or data repos, aka databases, because it is not fun, and the TL;DR rule always applies
2. With many complex transformations, it is hard to understand where data is flowing from or what is the current state
3. Code review happens within teams, and data review should happen between teams of analysts, engineers, and consumers, which makes things very hard.
This is where data engineering with LLMs fulfills their promises as a semi-AGI and emergent magical behavior. Data engineering relies on a communication mechanism between stakeholders, expressed by mostly unstructured data in a meeting that nobody documents. Outside of a few notes in a JIRA ticket, the process of deciding what these data structures are meant to model is a black box to most of the enterprise. Data stakeholders can define the initial requirements without data producers agreeing to take ownership. Later mistakes or misconceptions can be fixed, but that opens the data up to more churn and corruption. This leads to increased costs and ultimately unreliability. In every single major corporation I've been a part of, data cleanliness and maintenance is given little or no prioritiiy. Even working for Salesforce, most companies preferred to mimic thier old system with old deicisions and old models because, well they worked. For this reason they have the most issues with new AI tools like machine learning. Even the so-called miracles of automatic shopping registers by Amazon is shown to be heavily backed up by human workers in India. We're always trying to train for the real world, but neglecting the abstract made up world of data we can control and influence.
An LLM can skip over the bumps in the data and make useful more of the unstructured data we generate daily. Emails, notes, comments. It can generate ingestion mechanisms, and run SQL code to take care of the relational business, and then the outcomes are significantly improved. This generative framework is enough to solve most data issues in a reactive or, even better, in a preventative manner, not through tedious processes but by better communicating in a way that everybody is comfortable with.
[[Work and data]]
There are many reasons for optimizing business processes. It appears that generative AI is uniquely positioned to solve organizational communication problems. After all, what could possibly be better than an infinitely patient system that we can talk to, explain our concerns, vent, or even remain disorganized the whole time? These issues can't be solved with better data pipelines or cloud systems that use the most innovative cloud technology. The only solution is effective communication between people who have the knowledge and want to work with others. One big piece of information that supports generative AI is Conway's Law, which states organizational design systems mirror their communication structure. If the communication process is highly inefficient, then there you have it. Conway was a computer scientist who introduced the idea in the 1960s. Many visualizations of this law exist, most not very flattering for large organizations.
Suppose an organization experiences issues with innovation, inefficient processes, or immense organizational inertia. In that case, it implies that there is likely a faulty relationship between the data producers, the data consumers, managers, leadership, and pretty much everybody.
In most companies, the following is typically true: people don't know who is producing data, who is using data and how, what are all the value-adding data steps, who is making changes upstream, and so on. Certain groups of employees are blamed for it all when pipelines don't work correctly. This complete lack of visibility results in deep organizational issues, in the same way, a lack of data visibility between engineers results in business outcomes.
[[How business thinks it works!]]
<h3>Are you sure?</h3>
Statement: "Data engineering relies on a communication mechanism between stakeholders, expressed by mostly unstructured data in a meeting that nobody documents."
This statement seems to align with the general understanding that communication and collaboration are critical in data engineering projects. However, the claim that communication happens primarily through undocumented, unstructured meetings may be an oversimplification or specific to certain organizations. Many sources emphasize the importance of documentation and structured communication in data engineering:
- A blog post by Wavefront (now part of VMware) states: "Good documentation is essential for effective data engineering practices." [Source: <a href="https://wavefront.com/blog/data-engineering-for-analytics">https://wavefront.com/blog/data-engineering-for-analytics</a>]
Statement: "An LLM can understand all this, generate ingestion mechanisms, and run SQL code to take care of the relational business, and then the outcomes are significantly improved."
Statement: "These issues can't be solved with better data pipelines or cloud systems that use the most innovative cloud technology. The only solution is effective communication between people who have the knowledge and want to work with others."
This statement seems to align with the emphasis on communication and collaboration in data engineering projects. However, it may be an oversimplification to claim that technology improvements cannot solve these issues. Many sources suggest that a combination of improved processes, communication, and technology is needed:
- A Gartner report states: "Data and analytics leaders must focus on both the technical and organizational aspects of data engineering to drive value." [Source: <a href="https://www.gartner.com/en/data-analytics/topics/data-analytics-strategy">https://www.gartner.com/en/conferences/hub/data-analytics-united-kingdom/gartner-insights/data-engineering-trends</a>]
- An IBM article discusses the importance of both technology and collaboration in data engineering projects. [Source: <a href="https://www.ibm.com/cloud/blog/best-practices-for-data-engineering">https://www.ibm.com/cloud/blog/best-practices-for-data-engineering</a>]
Overall, while the provided text makes some valid points about the importance of communication and collaboration in data engineering, some of the claims may be overstated or oversimplified. It's important to consider multiple perspectives and sources when evaluating such statements.
(print: "<h3>" + (passage:)'s name + "</h3>")
The following skills are essential for any creative endeavor: applying knowledge, building relationships, solving problems, building expertise, organizing efficiently, communicating and iterating on ideas and concepts. I was always fascinated by animation. I admired the artists who could create jokes and stories with drawings, even if it meant repeating their work 12 times for a second of fun for the audience. All of us were inspired by books, performances, songs, teachers, and subjects that come with intense practice and talent to achieve that fun. I wanted to learn and practice those skills and emulate my influences.
I realized that behind every great work of art, there is an individual with talent, time, and means to execute their vision. In the entertainment industry, they are supported by many others, but we still attribute the work to the auteur, the person who controls and represents our admiration and delight.
In education, we are taught basic facts, axioms, and rules that we can apply to different kinds of problems. However, we may not learn how to approach creative problems, especially in the US. How do we identify a problem, and then acquire and apply the skills needed to solve it in a meaningful and satisfying way? Teaching computing is not very complex, for example. However, it's very hard to teach computing to produce a result that you may find fun or enjoyable. This seems true of so many basic skills like math and science. It takes decades of reading to enjoy a moderately dense worded book. In a company that hopes to educate their workforce they try to make it simple and fun to comply with the myriad rules and regulations, but do people really learn anything?
[[Why isn't work more like a game?]]
Games are a good example of how computing works. They often have a difficulty selector that changes the rules of the game by adjusting some numbers that affect the game mechanics. The game itself does not change, only the numbers do. This shows us how we can engage with computers for work or play, by manipulating numbers. When playing Super Mario or World of Warcraft, the numbers may be hidden or displayed to inform you of higher levels, more damage, more health. These numbers are often tied to behaviors considered beneficial to the company bottom line. The biggest number game is likely the stock market, but it is even more human than many may realize. If you wanted to pursue how to invest, or even understand market turmoil, there's numbers for that - greed vs fear. Back to being emotional.
On the other hand, we can also go back to the first computer interface, the command line. By opening a terminal, we can see behind the conventions that make computers easier to use and manage. We can also explore a different world, where words and commands can trigger a lot of activity and purpose. These interaction patterns shape our mental models. For example, once we learn algebra, we can start to abstract problems into patterns. Once we see patterns, we can simplify and improve our results with less work. The satisfaction of knowing how something works, and how we can change, simplify, and extend it, has driven much innovation, as well as fun and satisfaction.
[[What do computers do?]]
<footer>
<h3>Are you sure?</h3>
1. Teaching creativity and problem-solving skills:
The passage highlights the challenge of teaching creative problem-solving skills, especially in the US education system. This is a well-recognized issue, and researchers have explored various approaches to fostering creativity in education. For example, a study by Sternberg and Lubart (1999) <a href="https://journals.sagepub.com/doi/10.1177/107621759901100309">Link</a> proposed an "investment theory" of creativity, emphasizing the importance of nurturing creative skills alongside analytical skills.
2. The relationship between work and play:
The passage draws a parallel between games and work, suggesting that games can illustrate how computing works through adjustable parameters and mechanics. This aligns with the concept of "gamification," which explores the use of game design elements in non-game contexts, such as work and education. A review by Hamari et al. (2014) <a href="https://www.sciencedirect.com/science/article/abs/pii/S1071581914001256">Link</a> examines the psychological and behavioral outcomes of gamification.
3. Command-line interfaces and mental models:
The passage discusses the importance of understanding the underlying mechanics and patterns in computing, which can be explored through command-line interfaces. This connects to the idea of mental models in human-computer interaction, where users develop conceptual representations of how systems work. A study by Schobel and Jansen (2019) <a href="https://www.tandfonline.com/doi/full/10.1080/0960085X.2020.1796531">Link</a> examined the mental models of novice and expert computer users.
4. Creativity and individual expression:
The passage suggests that behind great works of art, there is an individual with talent, time, and means to execute their vision. This aligns with theories of creativity that emphasize the role of individual factors, such as the "Four P's" model proposed by Mel Rhodes (Person, Process, Product, Press/Environment) <a href="https://journals.sagepub.com/doi/abs/10.1037/a0029528">Link</a>.
</footer>
(print: "<h3>" + (passage:)'s name + "</h3>")
Artificial intelligence (AI) was never the coolest, nor most impactful career to pursue for the majority of computing's slow takeover of most media. Probably because it was painstakingly difficult to see patterns that were worthwhile. This kind of pattern matching was the stock in trade for early research. Going through images, or audio or any media you can compare bits and bytes and often see meanigful similarities and generate useful results. Think of Shazam, it samples a bit of music or sound then looks through a database to find a match. Pretty neat stuff, and useful for large scale database searches. And for legal purposes, its much more efficient than employing humans to police misuse of a copyrighted tune. Research into solutions around pattern recognition was profitable and useful. But it was hardly as groundbreaking as the next steps we'll outline shortly, machine learning and generative AI. But let's give some context.
[[Have we always been so obsessed with data?]]
One of the early techniques of AI research was to create machines that could aid us in making decisions by teaching them (creating large databases of) our language, our customs, and some aspects of our physical world. However, this approach failed. Not for lack of trying, but because human communication concepts are too complex, inconsistent, and context-dependent for pattern matching machines to use effectively. They often resulted in unhelpful or incoherent outputs. It was a tough field, and many AI researchers felt underappreciated as newer innovations like TCP-IP and internet communication took off. Yet, the internet also provided new opportunities for applying some of the rules and techniques developed by AI.
One of the challenges that the internet posed was image processing. Images were one of the most appealing and influential features of the new world wide web. Pornography, of course, was a large part of the early demand, but as browsers evolved, images became essential for capturing attention in the low-resolution world of America Online or Mosaic. Moreover, hyperlinks attached to images were often the easiest way to engage people in the new world of interactive navigation. However, these images, unlike text, could not be easily understood by computers themselves. They were poorly named, for the most part, and lacked metadata. You basically had to look at it. Even the HTML code that indicates what the image is, the ALT tag, is often blank.
[[What is data? ]]
Therefore, image recognition became a burgeoning area for AI to tackle the growing variety and volume of imagery. One of the most productive and useful tasks was to identify human faces. This involved two steps: first, detecting a human face in an image, and second, recognizing the identity of that face. This problem led to a breakthrough that was rather unintuitive and tedious. Extracting the figure from the background was a Herculean task for an image classifier, as it required getting a clean outline of the subject. Training a classifier also involved remembering that the computer only 'sees' sets of numbers representing light and shade, and manually tagging these areas to enable the computer to do meaningful work. This process took decades to achieve.
[[Machines? Learning? What does that even mean?]]
<footer>
<h3>Are you sure?</h3>
1. Early AI research and challenges:
- "A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence" by John McCarthy et al. Proposed the first AI research project in 1955.
2. Image recognition and computer vision:
- "Histograms of Oriented Gradients for Human Detection" by Navneet Dalal and Bill Triggs (<a href="https://doi.org/10.1109/CVPR.2005.177">CVPR</a>) - Introduced the Histogram of Oriented Gradients (HOG) feature descriptor for object detection.
- "Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks" by Shaoqing Ren et al. (<a href="https://arxiv.org/abs/1506.01497">arXiv</a>) - Proposed the Faster R-CNN model for object detection.
3. Face detection and recognition:
- "Face Detection: A Survey" by Erik Hjelmas and Boon Kee Low (<a href="https://doi.org/10.1016/S0262-8856(00)00005-9">CVIU</a>) - A survey of face detection methods.
- "DeepFace: Closing the Gap to Human-Level Performance in Face Verification" by Taigman et al. (<a href="https://doi.org/10.1109/CVPR.2014.220">CVPR</a>) - Introduced the DeepFace model for face recognition using deep learning.
4. Challenges and limitations of AI:
- "Inductive Bias in Artificial Intelligence" by Pedro Domingos - Discusses the importance of inductive bias in AI and the limitations of early approaches.
- "Challenges in Representation Learning: A report on three machine learning contests" by Isabelle Guyon et al. - Analyzes the challenges and limitations of representation learning in AI.
</footer> (print: "<h3>" + (passage:)'s name + "</h3>")
So how did a rudimentary adding machine manage to recognize your pet dog out of hundreds of other dogs? Computers have always been painstakingly trained, much like dogs, to do tricks we can appreciate. Outside of keeping track of things, math and so forth, we have trained them to reproduce most physical media. Starting with english text, those ASCII characters were the gateway for us to 'speak' to the innards of the computing system. Its capabilities were hampered by so many infrastructure issues, getting data in and out of disks, memory, processors, etc. But at the root was always engineers and sometimes designers that helped these brittle devices fetch and roll over.
But we manage to have achieved that goal of making all that infrastructure seem inevitable and expected as it becomes more useful. Yet machines don't have agendas or need anything, other than to serve some purpose. SETI, the program that scans the universe for anomolies once asked for idle computers to be networked to look for these anomolies. While it perhaps didn't find intelligent life, it did reveal how much computing power is lying around doing little or nothing. Considering the drawer of cellphones I still keep, they are amazing, capable devices that only wish they could serve some greater purpose. Yet, they are drained of power and useless. They don't mind, their purpose replaced by something more overpowered and seemingly indespensible till the next iteration comes along. What if they could be learning something? What if they could be utilized in some worthwile effort?
As babies we start with faces, recognizing our parents from others. Our skills in classifying and instantly categorizing types of objects, food, people, environments. It tends to happen organically, or through trial and error. For computers it is deliberate. A human uses sets of techniques and symbols to abstract a problem into small manipulatable chunks. Those are processed in some manner to derive a useful result. This human problem solving part of the construct has many limitations. Not only vacation time, but level of interest, talent, what the predecessors are doing, money. All manner of different elements can arrive at solutions impactful to the human race, or another Microsoft Word.
Yet, as with many breakthroughs, the transition to letting the computer figure it out for itself was the spark that led us to today. This process called Machine learning is the phrase I prefer using over AI. As we've started to find, some activities, however simple and mindless, like driving, are filled with outliers that, while rare, still keep even well trained models from achieving reasonable results. And this is not for lack of considerable expense in trying. Learning requires making mistakes, if only to realize there was a better outcome. ML techiques are not that far from any learning process. To illustrate, let’s make dinner.
<blockquote>
''Cookware and method of preperation (model)'' The cookbook has some instructions that focus the goals of the production or cooking the food. What cookware and heat are needed? How do you know when something is done, or time the introduction of a new layer of flavor? In any case, there’s learning models to be chosen that fits the dish. Often this is the //intelligence// or //algorithm//; but it's truly a bean counter that sets up some parameters for success, but is neutral, it can fail just as easily depending on the ingredients and how it utilizes each step's success.
''Ingredients (Data)'' - Stuff you make food with, for the computer its data. Often called //big data// - data as a topic is fairly close to ingredients for food. There’s definitely quality issues as most data people are aware of the GIGO principle. //Garbage in-Garbage out//. For our purposes, we want to rely on people, and indeed most of these large models depend on people to curate, trim or weed out bad data which will adversely effect the outcome. Or dish, if back to our metaphor.
''Cooking and Tasting (Training) ''- The both human and programmatic manner of seeing if your model is learning anything. Or if it is tasty. This is where you can adjust the model. For example, Apple collects scads of data in many of it’s products. But the threshold of how it determines if any of that is worth the users attention has high thresholds. If being viewed by a physician, this raw data may be more actionable, but the risk of false positives and misinformation is high without individual tweaks in different data situations. Metaphorically you don’t want to over do the salt until the dish is fully complete.
''Putting the dish on the menu ''- we’ve gone from just cooking dinner to opening a restaurant! This is why there could be millions of models, and indeed, the thing that kept AI back for so long was that each model could handle certain outcomes and idiosyncrasies. Yet, when a new problem arose, the process started over again. Also, considering our culture, we’re making dishes that are relevant to our culture, language and ecosystem.
''Exploring new dishes (Generalization)'' - And here is where the models that have taken off revolve around some key issues in our culture. So learning how the English language works, that’s a springboard for our next discussions. Being able to focus and represent faces has influenced digital photography. Voice recognition has enabled routing and interaction based on voice commands. So as more models become successful, and based on their goals it raises the bar of how and why the field is shaping how we leverage computers in new and novel ways.
</blockquote>
[[I have no idea or don't care how this works, just how it effects me]]
What is missing from this metaphor is what was the computer doing with the data? What outcome was it training for? In the sense of using a model to process information, we’re able to make meta-leaps to understand what pictures depict, and how English and perhaps other languages function to create meaning. What, then, is the outcome? Communication? Understanding? Allure? Well, by reversing these processes, you have the model create images, or write stories. We’ve turned tables and rather than the goal being for ML to learn about us, it mimics us.
This is not surprising, as we do it ourselves. Our lives are filled with mimicry. We may learn from media, parents, friends, but it’s the copying and aspiration of trying to embody these attributes that we count as growth or progress. Mastery and competence with the ability to create or embody a new combination of concepts and ideas, often called creativity, may be only a rounding error in our influences and ability to replicate.
It’s this more profound part of the recipe that caught many by surprise AI leadership asking to pause the inevitable misuse of this ability. Our goal here is to help define the area where this technique will fall short. And techniques to work around it’s shortcomings. As we may be mystified when seeing someone pinch with two fingers to zoom into a screen. It soon becomes rote and expected. This, often called affordance, what is something easier to perceive it being useful in.
[[Ok, so they can learn, how did that lead to being able to hold a conversation?]]
<footer>
<h3>Are you sure?</h3>
1. Machine Learning and AI techniques:
- "Pattern Recognition and Machine Learning" by Christopher M. Bishop (<a href="https://www.microsoft.com/en-us/research/people/cmbishop/prml-book/">Book Link</a>) is a comprehensive introduction to machine learning concepts and algorithms.
- "Artificial Intelligence: A Modern Approach" by Stuart Russell and Peter Norvig (<a href="https://www.pearson.com/store/p/artificial-intelligence-a-modern-approach/P100000610290">Book Link</a>) covers a wide range of AI techniques, including machine learning, natural language processing, and computer vision.
2. The use of data and training in machine learning models:
- "Data Science for Business" by Foster Provost and Tom Fawcett (<a href="https://www.amazon.com/Data-Science-Business-Foster-Provost/dp/1449361323">Book Link</a>) discusses the importance of data quality and preprocessing in machine learning applications.
- "Machine Learning Design Patterns" by Valliappa Lakshmanan, Sara Robinson, and Michael Munn (<a href="https://www.oreilly.com/library/view/machine-learning-design/9781098115777/">Book Link</a>) covers various design patterns and best practices for building and training machine learning models.
3. The capabilities of AI systems in areas like image recognition and natural language processing:
- "Deep Learning" by Ian Goodfellow, Yoshua Bengio, and Aaron Courville (<a href="https://www.deeplearningbook.org/">Book Link</a>) provides a comprehensive introduction to deep learning techniques used in computer vision and natural language processing.
- "Natural Language Processing with Python" by Steven Bird, Ewan Klein, and Edward Loper (<a href="https://www.nltk.org/book/">Book Link</a>) covers various Natural Language Processing techniques and their applications.
4. The potential and limitations of AI systems:
- "Artificial Intelligence: The Basics" by Kevin Warwick (<a href="https://www.routledge.com/Artificial-Intelligence-The-Basics/Warwick/p/book/9780415564823">Book Link</a>) discusses the current state of AI and its potential impact on various domains.
</footer>(print: "<h3>" + (passage:)'s name + "</h3>")
Large language models (LLM) are extensive deep-learning models pre-trained on vast quantities of data. This is a source of tremendous strength for the models but also a significant weakness. For this volume of data, it is impossible, at this time, to control the quality. Is there such a thing as quality? If you mine the content of Wikipedia, plus the New York Times, conceptually you have peer reviewed information. Yet, much like readers of Wikipedia or NYT, you can misconstrue or mangle that into something offensive or incorrect. We are also observing legal actions from authors who are not consenting to their intellectual property being ingested into the neural network structures which could lower even the low bar of what makes information useful. The term information is purposeful, something having a context, and all things being equal a good enough representation of an idea. Well known within journalistic practice, the frame of the information is often skewed for narrative clarity. Our lives are complex, writing is just one means of communicating ideas.
Nevertheless, the underlying transformer is a set of neural networks that consist of an encoder and a decoder with self-attention capabilities. These machine-learning inventions have been around for more than a decade in the form of complicated linear algebra-driven matrix multiplications. The breakthroughs lately are ways to do this very intense amount of processing with some level of performance. Matrix isn't like the movie, but in terms of a cube of numbers that describe vectors, or directions through a set of numbers. The faster you can compute the different possibliities, the more options you can weigh, which give better averages, better answers.
[[Machines? Learning? What does that even mean?]]
At the end of the day, the encoder and decoder extract meanings from a sequence of text and understand the relationships between words and phrases in it. This understanding is a function of the parameters stored in various matrices inside the system. It learns the same way we did. Matching patterns. Usual patterns become cliche. Unusual patterns have to have some insight or utility to get us to work on adding them to our mental models. The more work you do to expand this, the better. But GenAI can abstract that concept too. //Explain like I'm a 5 year old// seems insulting, but trutufully without experience in the topic, you have to bulid your understanding before getting to the subtle or interesting parts. GenAI can adjust the way it gathers the outcomes to mimic it's training set. What becomes more challenging is the 'training wheels' that the programmers put on output. Most of the answers are guarded, quick to offer caveats, and otherwise infuriating if you wish to make decisions since strong emotional ties to a concept are easier to accept or reject than diplomatic double-speak.
[[How AI Works Differently Than You Might Think]]
Transformer LLMs are capable of unsupervised training, although a more precise explanation is that transformers perform self-learning. Through this process, transformers learn to understand basic grammar, languages, and knowledge. The end result is similar to what happens in biological brains. We learn and understand through modifying connectivity and other characteristics existing between neurons.
In deep learning, unlike earlier neural networks invented decades ago that sequentially process inputs, transformers process entire sequences in parallel. This was possible because of the increased hardware capabilities and infrastructure that allow system architects to combine multiple CPUs, GPUs, and memory into a scalable system. Now, organizations use supercomputing-level hardware for training and reasoning with transformer-based LLMs. Interestingly, the progress in GPU-based computing is essentially a result of market forces created by the strong demand for digital entertainment. Video games have been a significant driver of innovation for a long time since real-time rendering engines are computationally complex tasks. Blockchain miners and AI might join the race today, but video games are still pushing the boundaries of computing due to the low latency computation demands and rising screen resolutions. It is also worth noting that the matrix multiplication models in computer graphics algorithms are remarkably similar to those that power generative AI.
Transformer neural network architecture allows large models, often with hundreds of billions of parameters. In a few years, these numbers will probably increase by a factor of 10. Large-scale models can ingest massive amounts of data, often from the internet. However, given the privacy and copyright constraints, many web resources might not be available anymore. The training sources might shift to other areas like private organizational data sets, vast amounts of self-driving car data, or research datasets in astronomy, oceanography, etc. So, we might be reaching the limits regarding the size of the data sets available from resources used to support the initial wave of generative AI innovation.
A critical factor in how LLMs work is the way they represent words. In fact, one can say that the spark for the success of the LLMs is the representation of content. It is also interesting that the embeddings, which is the ML jargon term for representing the content, are created through a separate neural network trained for this specific task. As a result, LLM is a technology that relies on an interplay between two neural network architectures: one used for representing the input data and the other (big) one for reasoning.
Earlier forms of machine learning used a numerical table to represent each word. This was the so-called bag-of-words approach, which worked really well despite the simplicity. However, this form of representation was limited in recognizing deep semantic relationships between words. Earlier technologies like Markov models overcame this limitation. However, using multi-dimensional vectors, commonly referred to as word embeddings, was a breakthrough. Embeddings represent words so that words with similar contextual meanings or other relationships are close to each other in the vector space. Word embedding algorithms support transformers that can pre-process text as numerical representations through the encoder and understand the context of words and phrases with similar meanings and subtle relationships between words that express feelings, humor, etc. It is then possible for LLMs to apply this knowledge of the language through the decoder to produce a unique output.
[[So, we use this to make money somehow?]]
<footer>
<h3>Are you sure?</h3
</footer>(print: "<h3>" + (passage:)'s name + "</h3>")
Cloud providers support organizational GenAI by hosting machine learning models, allowing you to choose infrastructure components such as instance sizes, service chains, and deployment endpoints. There are fundamentally two classes of platforms that organizations can choose: fully supported, less flexible platforms and more experimentation-oriented solutions that provide experimentation capability.
A managed GenAI service typically enables us to make API calls to access hosted models. Cloud companies provision those pre-built and pre-configured models, and there is less flexibility for configuration and model changes in the future. This environment facilitates quick customization of these models with your data and integrates them into applications using cloud services, all while relieving users of infrastructure management.
Key use cases supported by Bedrock include,
Text generation, including scenarios for conversation (chatbots), question answering, summarization, etc. Text generation also includes programming support for such tasks as SQL, Python, and other languages. Search includes finding semantically similar statements in documents that go beyond keyword-based retrieval.
[[Generative AI: A primer]]
While it hasn't been generally announced or avaialble, tools like //Copilot// from Microsoft offers a way to train local instances of AI on priveliged or secret company information. I use that term //information// deliberately, as finding anyting useful inside of and about the company you work for is one of the more unruly and challenging areas that has been traditionally hard to adequately address. Apple and Google will offer solutions too, Apple focusing on customer experience, finding your pictures and recommending music has been their prior focus. But finding people, finding files. These will be where it will shift with next generations of Siri trained on your data, not generic ones. Google as well, but as you shift these tools to personal rather than general problems, the usefulness will be hard to demonstrate. Your company is also hard to predict. They may try outsourcing intensely information retrieval oriented jobs to AI to save money. My work at Accenture was to consolidate decisions on consulting work into one data visualization tool. The alternate was every group creating Powerpoints to discuss and cajole leaders into deciding funding decisions. This one tool displayed and allowed exploration of the data automatically. It did however bring up the challenging part, what were we basing our decisions upon? To be honest, it was always about the pitch, rather than the substance. Seeing that the complexity of pricing and managing consulting work could be boiled down to certain economic measures at certain times begged for an automated decision, rather than one costing thousands in management time and expertise. GenAI can offer the middle ground, and most importantly it can remove the biases inherent and often codified into the way we work.
[[What do you mean by work?]]
Experimentation-oriented solutions for GenAI are fundamental since more organizations are still exploring the capabilities and economics of running LLM-based solutions. We don't know precisely how to optimize all AI components without the flexibility to change things up when working on use cases. Overprovisioning might lead to excessive costs in the long run, while under provisioning might limit the system's capabilities. The lack of explainability in deep learning models will always be an issue. Therefore, experimentation platforms will be important as new models appear almost daily. What we need is the flexibility to interact with:
Multiple foundation models, including public open-source systems: These models are pre-trained on vast amounts of data, serving as a strong foundation for various machine learning tasks, including areas such as text semantic classification based on context, question answering, image classification, text summarization, time-series forecasting, detecting possible fraud scenarios, and others.
Pre-built training and inference scripts that support the applications of models
Interfaces to critical capabilities to empower teams and progress through projects efficiently. Those interfaces might include notebook-style UI similar to many programming environments. For example, the Jupyter Notebook interface simplifies development through explicit stepwise interaction with the source code or configuration statements.
[[Ok, but how do I keep AI away from my job??]]
(print: "<h3>" + (passage:)'s name + "</h3>")
Andrew Hinton’s excellent book “understanding context’ provides interaction designers and computer programmers insight into techniques we use to manage and organize digital information. This discipline isn’t terribly old, but its goals to help people find relevant information through organizational hierarchies has been practiced for thousands of years. All of recorded history has found and developed architectures for its transfer and communication of the written word. Books, libraries, journals, newspapers, pamphlets. All are archived and summarized as best as possible within the physical realm to be retrieved if needed.
Yet, even within this sophisticated system, many things fall through the cracks. Fiction, mainly. Any ideas that are not easily categorized often are dumped into generalized buckets, and worse, sorted by the author’s last name. The goal is to be able to make large data sets searchable, prioritizable, understandable to facilitate general decision making. The invention of the web, and it’s hyperlink, sped up the need to have some reasonable way to indicate what, why and how a connection was needed and what value it offered. And, if clicked, where would you //go//? Would it continue or interrupt you on your journey? They used to call it surfing to indicate trying to stay afloat rather than sink.
[[Is work safe? ]]
Later as personalization took hold, Google and AdSense used this to develop large sets of words that they stored about your choices of sites you visit. It would be used to trigger personalization of advertisements which to the marketing world was a sea change. Prior to this they relied upon content from television shows, or more crassly just the viewership, not the topics to use to market their goods. Tonally, this juxtaposition of marketing pitch to scripted drama was just part of the price of ‘free’ media. The web offered unheard of opportunity to target or micro-target individuals based on affinities to particular ideas, as well as stages they may be in the decision making process. The value of this compared to general passive advertising seems immense, but the complexities of being able to deliver compelling contextual messages eluded us. As privacy concerns finally were recognized, different experiences now offer more insulation from these targeted messages.
Speaking of targeted, the famous anecdote of the parent receiving advertisements for baby formula and newborn care for their teen daughter from Target seemed to be the perfect storm of how data mining of purchase or browsing data can correlate to different stages or life moments. Yet in the crass context of product marketing even though the indicators proved true (she was pregnant) the resulting contextual insight was seen as too relevant, intrusive, scary. Advanced yet //not// acceptable.
Ironically, the same algorithms that use concepts and taxonomies to categorize interests are a boon for YouTube and Facebook scrolling. All the short bursts of attention or approval are used to categorize an ever more enticing landscape of content. This context may not be our best example, as media tends to focus on what we like, rather than what we need, such as diapers and wipes.
And the disconnect is still part of a debate as we reach election cycles. What content will be highlighted for us? Can we be certain our neighbor, or loved one isnt experiencing something completely different. Even when it comes to the current practice of choosing which family member is viewing Netflix to continue the deep levels of categorization and personalization that give context to.. well there’s the issue, it cannot create content directly for us? Or can it?
But if content was created solely for us, would that be desirable? In many respects, the algorithms and words and tracking merely allow experiments to take place. As many of us could attest, it hardly makes perfect choices. In fact long experimentation with clickbait advertises lands on odd phrases “You won’t believe..” And imagery that at best is impressionistic. Yet, as the algorithm shows, this particular combination delivers better results. Until the next experiment shows difference, etc. It’s this commitment to long term experimentation, along with general size and scope of advertising budgets that does not reward taking these sorts of changes. Yet, this process is exteremely computer centric. It can count. And it can put weights on variations. And it is this mechanic that runs machine learning. It attaches probabilities and weights and then gets some sort of result which is assessed and factors into the next iteration. Its a microcosm of the campus or schoolroom, yet more ravenous and much more experimental.
[[Thats the world of fun, what about work?]]
<footer>
<h3>Are you sure?</h3>
1. Information organization and context in digital environments:
- "Understanding Context: Environment, Language, and Information Architecture" by Andrew Hinton
- "Contextual Design: Defining Customer-Centered Systems" by Hugh Beyer and Karen Holtzblatt
2. Personalization algorithms and data mining:
- "The Power of Habit: Why We Do What We Do in Life and Business" by Charles Duhigg (<a href="https://www.amazon.com/Power-Habit-What-Life-Business/dp/081298160X">Book Link</a>)
- "Data Mining: Concepts and Techniques" by Jiawei Han, Micheline Kamber, and Jian Pei (<a href="https://www.amazon.com/Data-Mining-Concepts-Techniques-Management/dp/0123814790">Book Link</a>)
3. Machine learning and algorithms:
- "Machine Learning: A Probabilistic Perspective" by Kevin P. Murphy (<a href="https://www.amazon.com/Machine-Learning-Probabilistic-Perspective-Computation/dp/0262018020">Book Link</a>)
- "Reinforcement Learning: An Introduction" by Richard S. Sutton and Andrew G. Barto (<a href="https://www.amazon.com/Reinforcement-Learning-Introduction-Adaptive-Computation/dp/0262039249">Book Link</a>)
4. Privacy concerns and targeted advertising:
- "The Age of Surveillance Capitalism: The Fight for a Human Future at the New Frontier of Power" by Shoshana Zuboff
- "Privacy and Big Data: The Players, Regulators, and Stakeholders" by Terence Craig and Mary E. Ludlor
</footer>(print: "<h3>" + (passage:)'s name + "</h3>")
Its not unusual that we all have our own ways of managing the world of work. In most businesses, computer usage could be considered ancillary. You create things, build buildings, cook food as in our earlier metaphor. So many things that bring value are not expressly tied to machines and the data they produce to be useful. Yet, as more media and objects become digitalized. Meaning that they can exist in a purely digital form, new problems arise with maintaining and categorizing this kind of data. The triumph of mapping accomplished by many parties such as Google, allows for unheard of sophistication in modeling new scenarios, not to mention the possibility of these algorithms handling fairly complex decision making like a driverless car. The successful modeling of data around particular topics such as art, film, music, video all contribute to tools that also make use of conventions to teach others how to mange what was once specialized skillsets with need for thousands of hours of individual training. SaaS blurred the lines between what used to take dedicated processing power to achieve and made it available and sharable in new configurations to remove the need for computers to expressly store and manage information on an personal level.
[[So, we use this to make money somehow?]]
And good thing too! Since we are often unequipped to know just how to mangage the plethora of media, files, configs, memos and due dates. Not to mention channels like text and email also continually flood the system, relying only on date and time to provide context. So it’s a mess. Therefore it takes a revolutionary technology to be able to see our way through to how to solve the most pressing problem. Access to quality information, or knowledge. Or you put it into a folder. Metaphorically, of course, but useful enough to have kept things organized for many decades. Also helped that museums or other insitutions were following the similar ‘ put it in a box’ way of adding context and pattern to essentially random things.
In the Data-Pattern- Information-Knowledge-Wisdom pyramid, each level is scarcer and also needs a transformation to move it up that extra step. From data to pattern you need some sort of grouping mechanism. That was what early AI experiments handled, the ability to classify. Now, that isn’t all that meaningful, but as data is getting cheaper to gather and often more accurate, ignoring some fields like polling, finding ways to cluster and connect it is a valuable task. //Information requires context//. This is often called metadata, contextual cues and markers. Some services invent contexts like Instagram (a place to put photos taken on your phone) or Facebook, a conversation between real friends. Pinterest, a place to reference photos and create scrapbooks of ideas, etc. Or more ambitious ones like monday.com or aitrable.com that handle the challenges of managing workflows or replacing Excel to analyze data. Each of these new businesses are in the information business. Allowing you to add your layer of understading or knowledge to unrelated information to build useful concepts. On one’s own device, managing files was such a chore we often threw everything on the desktop. That metaphor was invented to help us transfer concepts from the world of work to this new world of virtualization. Either way, a messy desk is a sign of.. the difficulties of organization and prioritization.
[[So, if AI is here to help, how can I use it?]]
While any deep understanding of the myriad of methods files are managed, named, and organized, it’s clear that the metaphors adpoted to make computers easier to use are idiosyncratic. For example, my friend was having issues with his computer, so decided to repair it by reformatting. He put all his files on //the desktop// only to find that all were deleted. While this was technically not surprising, //the desktop// is merely another folder. But the way it's displayed is that the hard drive is a place on your desktop. Seeing it as a separate place revealed to me how these mental models of computer file storage are merely conventions, not really intended to be long term or meaningful as the information they contain. The //cloud// is just a folder on the server as well. On a similar note, folders, and folders within folders may be a convenient organizational trope, they also suffer from similar issues. Beyond one layer, it can be tedious to navigate in and out of different contexts and hierarchies. Folder //this// inside folder //that// inside folder //another//. Naming folders is normally terse which makes it tough to call something what it really contains. File names can’t really contain spaces, so some OS’s fake this by having a ‘real’ name and a display name. Folder iconography can be modified, but often stays uniform and monotonous making browsing harder, and the chances of misplacing greater. A recent conversation someone asked me //what if it was all erased or lost//? I replied I would just move on and create and gather more stuff. The value is in the context not the content. Or more often, we just enjoy hoarding stuff, digital just makes it less obvious how hard it is to create value from this mess.
[[Why is data architecture so hard?]]
These systems were designed to be idiosyncratic and serve the individual whims and habits of the user. Giving them a familiar way to handle these virtual containers of content. However, the architectures of desktop computers meant bloat, libraries, supporting files, and other oddities that were not as easy to explain or contextualize. So until the full-on onslaught of mobile, it remains an area of extreme oversight. Most of us can relate to the //I have that file somewhere// or //I did that work, now where did I put it//, in my case its usually several copies of the file in PDF form or image, but the master file is gone or lost. And what did I call it? Why bring this up in a discussion of AI? ''It will become apparent that within work contexts, or in personal ones, the access to our information is predicated on our abilities in categorizing, tagging, organizing and storing this information.'' In traditional desktop computing environments, the chances are slim to none of locating meaningful information. This is despite OS’s best efforts on both the Mac and PC side to offer aid to those who wish to put in the effort. Even Cloud or SaaS storage of this sort of unstructured data is problematic. I have been through several wholesale reorganizations of work product within design. At my tenure at Salesforce, I found most companies are unable to modernize to the newer and more powerful interfaces they have pioneered, in favor of the tried and true replications of their earlier forms and processes.
The expense of modeling data around a business goal is not one to be taken lightly. Yet, GenAI can tackle all of this random information with ease, categorizing, summarizing and making available huge amounts of information depending on the context. I see it as a new opportunity and tools to make use of, and benefit from all this unstructured information. What knowledge can we create with this tool becomes the challenge.
[[The confusing history of gamification]] (print: "<h3>" + (passage:)'s name + "</h3>")
The capturing, sharing and management of information has gone hand in hand with a fundamental shift to increasing wages for workers, profitability and emergence of entire new industries. If you're not good at communicating and organizing your thoughts and goals, you are at a disadvantage as an employee. As Bertrand Russell put it in "In Praise of Idleness":
<blockquote>
First of all: what is work? Work is of two kinds: first, altering the position of matter at or near the earth’s surface relatively to other such matter; second, telling other people to do so. The first kind is unpleasant and ill paid; the second is pleasant and highly paid.
</blockquote>
Along with that ability to tell others what to do, it also fosters systems or axioms to share those techniques that creates countless products and services. There’s a wealth of metadata used in processes like Six Sigma that attempt to measure and improve communication, reduce waste and foster innovation. Yet the adage still holds true that the larger the company, the more the ability to form high-performing teams, reward positive contributions and generally work becomes more and more challenging. Yet commuincation amongst individuals remains the sticking point. Each come with different abilities, understanding and context. Each potential oversight or misstep from team members can bring down subsystems and even threaten the core business. This normally doesn’t favor risk-taking innovative individuals. In order to take advantage of, or benefit from new ideas, outside consultants, or non-revenue focused lines need to be created. Integrating into standard practices any innovation can also be challenging. In general, more outsourcing, more reliance on tried and true processes and procedures often favor the upstart, leading to more disruption, and more opportunity.
Automating has always been the ideal, but legacy success doesn't predict a future of the same. For example, the technical complexities of a logistic enterprise like DHL, UPS or the US postal service can be bested by a 21st century company like Amazon, that can use new sensors, robotics and a host of other innovations to be able to profit from an unprecedented goal to offer free 2 day shipping. This cost saving is possible by combining iterative risk-taking moves with an architecture that takes advantage of some of the shortcomings of other competitors.
[[Have we always been so obsessed with data?]]
The goal of business to manage, capture, and utilize structured data to handle transactions, create efficiencies and handle growth has significantly changed since more and more data is captured through our interactions with digital devices. A smart watch for example is measuring thousands of parameters and consolidating them into different value added information to the wearer, and perhaps even to medical personnel. Taken on a grand scale, the ability to capture fairly useful data automatically and cheaply is the norm. The introduction of the <form> element in the first HTML spec also hinted as to how useful this new medium could be to data gathering from regular users. As individual data gathering units, the challenge became the skills and ability to manipulate that information for problem solving.
[[So, how do you manage people using data?]]
(print: "<h3>" + (passage:)'s name + "</h3>")
Data is the basis for most jobs. Gathering it with as little error as possible is 95% of science careers. Math is a language we invented to talk about numeric data like it was a Shakespeare play. It netted us staggering insight into the world we live in, much of which hasn't changed in hundreds of years due to some really clever tricks and insights. But it all came from data gathering and testing. Moving information around is far more profitable than creating it, which is really back to our theme of how GenAI can organize better than we can. Data analysis has been a hot career path for the past 10 years. But a consistent theme is who performs these jobs, the level of education, pay and ability to build a career in the area of expertise. From pollsters, to clerks, to engineers to just about every role in business, things have shifted in a major way over time. Lets look at the job market for data over the years for context.
''Pre-Industrial Revolution''
In the early days of business, data collection was primarily manual. Businesses relied on handwritten records, ledgers, and accounting books to track transactions and customer information. Low education, repetitive labor.
''Census and Surveys (19th Century):''
With the growth of industrialization and urbanization, governments and businesses started conducting formal censuses and surveys to gather demographic and economic data. These were often time-consuming and relied on paper-based forms. Basic Literacy, more accessible to middle class.
''Punch Cards and Tabulating Machines (Late 19th Century):''
The invention of punch cards and tabulating machines by Herman Hollerith (later IBM) revolutionized data processing. These machines were used to process and analyze data from the U.S. Census, making data collection more efficient. Advanced literacy in mathematics and mechanical engineering.
''Early Computers and Mainframes (Mid-20th Century)''
The advent of electronic computers and mainframes enabled businesses to automate data collection and processing tasks. This era marked the beginning of electronic data storage and retrieval systems. As the Graphical User Interface became popular, this technology spread to non-engineers, and entry level office workers. Normally this did not extend to executive roles.
''Barcode Technology (1960s)''
The invention of the barcode and the development of point-of-sale (POS) systems allowed businesses to collect data on product sales and inventory more accurately and efficiently. Manual labor, unskilled. Little chance for advancement.
''Market Research and Telephone Surveys (20th Century)''
Market research techniques, including telephone surveys, gained prominence in collecting consumer opinions and market data. These methods allowed businesses to make more informed decisions. Manual unskilled labor.
''Internet and Online Surveys (Late 20th Century)''
The rise of the internet brought about online surveys and data collection via web forms. This allowed businesses to gather customer feedback and demographic information through digital channels. Skilled form design and User experience work, along with internet programming skills were well paid and required STEM education.
''Customer Relationship Management (CRM) Systems (1990s)''
CRM systems like Salesforce and SAP Customer Experience emerged to help businesses collect, organize, and analyze customer data, enhancing customer relationship management. A range of skilled workers, even extending into leadership roles for data analysis and monitoring.
''Big Data and Analytics (21st Century)''
The proliferation of digital devices and the internet has generated vast amounts of data. Businesses have embraced big data technologies and analytics tools to extract valuable insights from this data. Highly educated within fields of data analysis and design. Often the data was processed into cubes, or computed across different dimensions to retrieve insights quicker.
''IoT and Sensor Data (21st Century)''
The Internet of Things (IoT) has enabled businesses to collect real-time data from connected devices and sensors. This data is used in various industries, such as manufacturing, healthcare, and logistics. Extremely highly skilled to interpret the data, little to no effort to collect.
''Machine Learning and AI (21st Century)''
Advanced machine learning and artificial intelligence techniques are now used to automate data collection, analyze large datasets, and make predictive recommendations for businesses. This is where the book is focused, what class and education level will be needed? To produce AI value it may be well paying, but may not require more training or knowledge due to comfort with API based architectures.
''Privacy and Data Protection (21st Century)''
Concerns about data privacy and security have led to the development of regulations like GDPR and CCPA, which impact how businesses collect, store, and use customer data. Oversight from leaders in government and civic functions may be an interesting sub-topic, as socially we do not favor those with technical education to govern. Yet, as we have seen from wholesale profit from how data gathering has changed dramatically in effort, education and rarity, it may indicate a need to promote this ability in our government officials.
[[What can we do with all this data?]]
(print: "<h3>" + (passage:)'s name + "</h3>")
Currently, there's a weak data culture in most non-tech businesses. One of the ways we showed companies like Amazon seem to take over complex industries overnight is a commitment to rapid experimentation and measurement. As any sports fan can attest, data is everyting when it comes to how you concieve of the value of teams, players, matchups. Oddly, profit should be a key measurement in a business, yet it's often loosely coupled with job duties and performance. During recent downturns, most companies have shed workers and even roles in an attempt to streamline. While this is nothing new, is it based off of actual worth of the role and performance?
As a person who evaluated and mentored large teams, I don't recall ever feeling like measurement was useful, most often it was based on anecdotal likes or dislikes. Someone's remark in a group meeting that rankled another. Measurement of work without any numeric indicators is similar to coaching poor scoring players. Perhaps they do other things to aid the team? Their attitude or willingness to learn? There was once a set of 50 key abilities to be judged in a five point scale of mastery for a role. The person wanted the new role, and these 50 judgement calls were the way to achieve it. Of course, its ridiculous to imagine improving 50 areas of your skillset, or even knowing why something is a 3 to 4 vs a 5.
[[Its not an engineering problem to make sense of data]]
Every quarter I'm evaluated by my students. Thankfully on only a handful of metrics, and puzzlingly on a 6 point scale. Normally it's quite a pleasure to see where and how I managed to effect a student, and where I failed. After 10 years, it's still a mystery of when things go well vs poorly. Also, I haven't changed what I teach and the method I teach it, which is what a job used to promise. An opportunity to practice skills and gain experience in a unique opportuinity or environment.
A job is harder in the beginning, as you train yourself in the types of encounters and situations you're asked to solve. After some time, it becomes routine, and your ability to see new possibilites will collapse under the repetition. This baseline is often the most productive area to innovate, yet, it's the confidence that comes with managing the expected where most find comfort, confidence, and mastery. It's new situations that are puzzling. Not to mention switching to another set of responsibilities, or management!
[[So the job of the future is writing effective prompts?]]
Now, its up to us to figure out how to make our jobs produce more measurable, and therefore value our work and efforts. What do we do that can be measured? For designers, especially, this reeks of micromanagement. Or lack of freedom. Oddly, that could be true, these abilities or technologies are most likely to be used to remove positions rather than enhance them. Microsoft is consistently workign to message me about how my work efforts are typical, atypical, or how much time I should have to focus on getting the work done.
Our goal is to have this be a net gain for jobs. Especially if you enjoy or got into a particular job to grow your skills, learn new things, or work in larger cooperative collaborative situations. Managaing collaboration is one of the joys of management. Finding people that like working together, and in sum produce better, more interesting work. Much of my experience at Leo Burnett, a famous advertising firm, was finding these pairs, often a pair of creative directors, that somehow together came up with and sold clients on their wild ideas.
That wasn't the only role, in technlogy, we were wrestling with production, and how to realize the creative ideas with the technology and talent we had. Much of that production was simliar to other creative fields where it takes a village of talent to produce something good, not just visionaries.
[[Machines? Learning? What does that even mean?]] (print: "<h3>" + (passage:)'s name + "</h3>")
Data is one of the building blocks of information. The processes needed to transform data into something usable differ in regards to numeric data to qualitative data. "//FIrst Name"// and "//Total items sold//" are similar, but require different techniques to make them useful in broader trends or contexts. Finding meaning in numbrers has been heavily invested in over centuries resulting in amazing breakthroughs in statistical patterns, predictive charting and correlations, and good old addition and subtraction. All are stunningly useful, yet qualitative data have been challenging until GenAI techniques have arose to see and uncover more complex relationships. Still, its worth reviewing data history to see how we got here. Pay close attention to the costs and skills needed, that will come in handy as we see how this will move us into uncharted waters.
''Pre-Industrial Revolution''
In the early days of business, data collection was primarily manual. Businesses relied on handwritten records, ledgers, and accounting books to track transactions and customer information. //Low education, repetitive labor. //
''Census and Surveys (19th Century):''
With the growth of industrialization and urbanization, governments and businesses started conducting formal censuses and surveys to gather demographic and economic data. These were often time-consuming and relied on paper-based forms.// Basic Literacy, more accessible to middle class. //
''Punch Cards and Tabulating Machines (Late 19th Century):''
The invention of punch cards and tabulating machines by Herman Hollerith (later IBM) revolutionized data processing. These machines were used to process and analyze data from the U.S. Census, making data collection more efficient. //Advanced literacy in mathematics and mechanical engineering. //
''Early Computers and Mainframes (Mid-20th Century)''
The advent of electronic computers and mainframes enabled businesses to automate data collection and processing tasks. This era marked the beginning of electronic data storage and retrieval systems. //As the Graphical User Interface became popular, this technology spread to non-engineers, and entry level office workers. //Normally this did not extend to executive roles.
''Barcode Technology (1960s)''
The invention of the barcode and the development of point-of-sale (POS) systems allowed businesses to collect data on product sales and inventory more accurately and efficiently. //Manual labor, unskilled. Little chance for advancement. //
''Market Research and Telephone Surveys (20th Century)''
Market research techniques, including telephone surveys, gained prominence in collecting consumer opinions and market data. These methods allowed businesses to make more informed decisions.// Manual unskilled labor. //
''Internet and Online Surveys (Late 20th Century)''
The rise of the internet brought about online surveys and data collection via web forms. This allowed businesses to gather customer feedback and demographic information through digital channels. Skilled form design and User experience work, along with internet programming// skills were well paid and required STEM education. //
''Customer Relationship Management (CRM) Systems (1990s)''
CRM systems like Salesforce and SAP Customer Experience emerged to help businesses collect, organize, and analyze customer data, enhancing customer relationship management. //A range of skilled workers, even extending into leadership roles for data analysis and monitoring.//
''Big Data and Analytics (21st Century)''
The proliferation of digital devices and the internet has generated vast amounts of data. Businesses have embraced big data technologies and analytics tools to extract valuable insights from this data.// Highly educated within fields of data analysis and design.// Often the data was processed into cubes, or computed across different dimensions to retrieve insights quicker.
''IoT and Sensor Data (21st Century)''
The Internet of Things (IoT) has enabled businesses to collect real-time data from connected devices and sensors. This data is used in various industries, such as manufacturing, healthcare, and logistics. //Extremely highly skilled to interpret the data, little to no effort to collect.//
''Machine Learning and AI (21st Century)''
Advanced machine learning and artificial intelligence techniques are now used to automate data collection, analyze large datasets, and make predictive recommendations for businesses. This is where the book is focused, what class and education level will be needed? //To produce AI value it may be well paying, but may not require more training or knowledge due to comfort with API based architectures. //
''Privacy and Data Protection (21st Century)''
Concerns about data privacy and security have led to the development of regulations like GDPR and CCPA, which impact how businesses collect, store, and use customer data. Oversight from leaders in government and civic functions may be an interesting sub-topic, as socially we do not favor those with technical education to govern. Yet, as we have seen from wholesale profit from how data gathering has changed dramatically in effort, education and rarity, it may indicate a need to promote this ability in our government officials.
[[What can we do with all this data?]]
<footer><h3>Are you sure?</h3>
1. History of data collection and processing:
- "A Brief History of Data Visualization" by Michael Friendly and Daniel J. Denis
- "Data Processing Technology and Economics" by H. D. Huskey and G. A. Korn
2. Advancements in data collection and analysis:
- "Big Data Analytics: A Survey" by Xindong Wu et al.
- "A Survey on Internet of Things: Architecture, Enabling Technologies, Security and Privacy, and Applications" by Muthukrishnan Sathiamoorthy et al.
3. Machine learning and AI for data analysis:
- "A Survey on Machine Learning Techniques for Internet of Things (IoT) Security" by Shancang Li et al.
- "Deep Learning for Time Series Forecasting: A Survey" by Haixu Wu et al.
4. Data privacy and protection:
- "A Survey on Data Protection Legislation in Smart Grid: From Monolithic to Granular Compliance" by Debabrata Kar et al.
- "A Survey on Data Privacy in Machine Learning: Taxonomy, Applications, and Techniques" by Fatemeh Mireshghallah et al.
</footer>
(print: "<h3>" + (passage:)'s name + "</h3>")
Few think of their jobs as opportunities to create and manage data. Even those with the role of database supervisors or analyzers rarely creatively construct the data models they administer. Their day is spent in moving knobs to create scenarios which translate to some report or regulatory purpose. Also this data is rife with empty fields, or placeholders that reveal the difficulties in managing the way the data was captured. A company we consulted with they looked to extrapolate needs and behavior of students based on their grades. I had a similar insight when working with the data at a university. My solution was to connect people with a poor grade with a faculty advisor. We had office hours avaialbe, most of which were unused. The connection first was to let the faculty do it themselves with a batch email. But when they did not follow through, we just automated the message and connected to the system with a friendly message to make a connection. It kept many students who would normally abandon computer science in the program.
Yet, generically, what do you do with a grade or a field of study that isn't connecting with a student? The outcomes of even one class or coursework in one quarter is going to be complicated to interpret. Were there extenuating circumstances? Or is this the wrong field? Or is it the teachers issue? Thus interpretation can be challenging at best, but even in my example, is it ethical? At the time it felt as though if we did anyone some good, it was worth trying. Facebook and Google started in a similar way. Yet, the data proves so useful, then even if your intentions are good, programatically it seems risky and often upsetting to realize how you can predict and influence behavior. Yet as employees, is your data and the meaning behind it your property or that of your employer? McKinsey and other thought leaders have touted a far more intrusive data-driven enterprise as the standard to adopt by 2025.
[[What do you mean by work?]]
Similarly, we have jobs that we may or may not be evaluated in our performance or outcomes. We may complete a task, find approval in our co-workers or bosses. Yet, the data we work with, most often is beyond our influence or control. One example I have great respect for is tech support. Having several years of experience, it's only very puzzling or unique situations where I can't diagnose or repair my own devices. But most haven't the time or need to invest in the complex and puzzling world of computer interfaces. As much as we work towards the ideal of intuitive interfaces, its still a rarity. My old company had some difficult software, and even attempts to modernize it were fraught with challenges. Yet it was the support team that really kept us afloat, since everyone learns and processes differently, this layer of translation from intent to activity is valuable. Most attempts to automate this are challenging, yet GenAI has made bold strides, this marks a puzzling shift, will consumers go with the flow?
[[How business thinks it works!]]
But the data these support teams work with is often is beyond their control. I've spent hours with support circling around a problem, often to find it's an entirely different system or issue we didn't uncover. This doesn't reflect on their skills, but does bring up questions on efficiency, skill, communication and evaluation. My father was a clinical psychologist, and anecdotally he never really cured any of his patients, which was his goal. He may have made them feel more confident, or perhaps just having someone to confide to was valuable enough. These jobs have challenges in providing usable data to train, improve or even capture the issues in a usable way. That area is ripe for generative AI to step in. For example, the way data is captured in clinical situations, such as a doctors visit. Many items are coded as data points to streamline billing and processing. Yet, its the doctors notes, the narrative that really holds the most contextually relevant information. Like police reports, this is much more valuable to be utilized by GenAI to see insights across treatments and diagnosis.
[[Have we always been so obsessed with data?]]
Yet, we continue on. It's a system we're familiar with, one we often succeed within. What is lacking is clear outcomes from efforts, and data that supports that you're improving some aspect of your approach. Is it knowing that one approach is futile? Absolutely, a valuable lesson. Is it getting to a positive outcome to the caller quicker? Possibly, time efficiencies are easy to measure, but also challenging to share. Its not uncommon to see people speeding through solving a rubics cube. Personally, I'd have no ability or interest in even solving one, but certainly that skill is difficult to transfer to other situations. Yet it's impressive. These sort of metrics can often define your presence in the workplace. They also can often mislead others into type-casting you and your abilities. Excellence in a defined, even difficult task, is actually the reason we're here. This specialization and competence in defined areas is where we think work will change.
[[I'm more about the human side]]
<footer>
<h3>Are you sure?</h3>
Unfortunately, the given text does not provide any direct citations or links to authoritative research sources. The text discusses the role of data in various jobs, the challenges of data capture and analysis, the concept of a "data-driven enterprise," and the potential impact of generative AI on certain professions. However, it does not reference specific research papers or studies.
Here are some potential research sources that could support or contradict the key points raised in the text:
1. Data-driven decision-making and the data-driven enterprise:
- "The Data-Driven Enterprise" by Xingyu Lin et al. (<a href="https://doi.org/10.1007/978-3-030-83504-3">Book</a>)
- "The Role of Data and Information in Decision-Making" by Steven Tadelis (<a href="https://economics.stanford.edu/sites/g/files/sbiybj9386/f/publications/tadelis_10_2019.pdf">Research Paper</a>)
2. Challenges in data capture and analysis:
- "Challenges in Data Quality: The Role of Data Governance Practices" by Marisa Vasconcelos et al. (<a href="https://doi.org/10.1007/978-3-030-36691-9_18">Book Chapter</a>)
- "Data Quality Challenges and Future Research Directions in Big Data Analytics" by Dania Souidene Doukha et al. (<a href="https://doi.org/10.1016/j.jksuci.2021.05.026">Journal Article</a>)
3. Impact of AI on various professions:
- "The Future of Employment: How Susceptible Are Jobs to Computerisation?" by Carl Benedikt Frey and Michael A. Osborne (<a href="https://www.oxfordmartin.ox.ac.uk/downloads/academic/The_Future_of_Employment.pdf">Research Paper</a>)
- "Artificial Intelligence and the Future of Work" by James Manyika and Jacques Bughin (<a href="https://www.mckinsey.com/featured-insights/employment-and-growth/technology-jobs-and-the-future-of-work">Article</a>)
4. Generative AI and natural language processing:
- "A Survey of Language Models" by Thomas Wolf et al. (<a href="https://doi.org/10.3897/rio.8.e88401">Research Article</a>)
- "The Science Behind Generative AI: A Brief Introduction to Transformer Language Models" by Vinod Nair (<a href="https://arxiv.org/abs/2303.12463">arXiv</a>)
These research papers, book chapters, articles, and other sources provide authoritative information and analysis on the topics discussed in the text, such as data-driven decision-making, challenges in data capture and analysis, the impact of AI on various professions, and the capabilities of generative AI and natural language processing. While some of the points made in the text may still be subject to debate, these sources can help support or contradict the claims based on rigorous research and experimentation.
</footer>
(print: "<h3>" + (passage:)'s name + "</h3>")
In America, we've toyed with concepts of jobs and work. It may have meant something in different eras, and technology has only changed these concepts slightly. Work is introduced to us as some different kind of activity than 'play'. We get used to this in school, time to work, time to play. Yet, play is a sort of work, experimenting with different techniques to develop skills. In the library our study group often strayed away from our teachers desire for us to read and examine history. We became obsessed with rolling dice and playing craps. It's a different sort of study, and was more fun, if a bit hard on the lunch money. Perhaps people skilled in sport keep that attitude of play rather than work. Work was always a component in any success. Working hard was promoted relelentlessly as a solution for lack of success. Or a reason for it's arrival.
Jobs were different. A way to reliably make money. If you didn't have one, being a student was your job. Learning was a vocation. Many who fell in love with this idea continued in academics and earned advanced degrees. Stay long enough and you can get the credentials to teach others, perpetuating the cycle. One of the benefits of learning on the job in a university, I left with a masters degree in Computer Science, and a love and vocation in interaction design. Jobs in this field were plentiful, if misnamed, and often misaligned with goals around what these new methods and tools could achieve. But it's a job. One of the aspects of a job is some kind of relationship, routine. Dressing, driving (or commuting) to a place where you gather with others to spend a few hours, then leaving, eating, sleeping, repeating. Weekends are optional. Many have jobs with different routines, but the job part is usually the same. Some repetitious presence within a context and people that produce some outcome.
Work is almost entirely separate concept. Without too much of a diversion, work is most synonomous with failure. "I'm working on it" means it isnt done. Its a little bit of psychology, but when something is successful, it rarely feels like the effort needed was greater than when it fell flat. I introduce a new subject to my class weekly, each is a challenge to determine how they will take in the new information and produce something to practice the concept. I provide a forum where they can get feedback from me, and others on their work. My hope is they engage early, and offer more feedback. The more they can refine and try different ideas, the less sting failure or difficulty has. It becomes a puzzle or a game. It's work. While this could be unique to creative work, watching other professions, work is about environment, context, preperation and then execution. It's a series of things rather than a quantifiable activity.
[[Why isn't work more like a game?]]
Jobs are repetitive, and rely on more social hierachy and communication of varying goals than work. Work is practice. Jobs are roles and performance over time. The more likeable you are, the more you can engage others, the more likely you can be considered a good employee. Constant failure or struggle is discouraged. If you are new or challenged with your job role, you may learn by trying something over and over, and if you get feedback you can improve over time. At some point, this is actively discouraged or seen as suspect. The witnessing of a failure is far safer than being the practitioner of that failure. While some companies may say they encourage risk taking, the majority are risk adverse, especially in the work they perform for others. Those that have the skills, or experience could be teachers of those they lead, but most often they are assessors. Noting obvious flaws or doing the impossible job of understanding another person. We'll discuss at length this conundrum and how Gen AI could be a boon to workers. Directed feedback, especially as infrequently as happens in most workplaces can be problematic. This gets at the heart of why many creative jobs are so hard to staff, learn and measure the individual contribution.
[[What is work for?]]
<footer>
<h3>Are you sure?</h3>
1. Jobs as routine and structure:
The passage describes jobs as having a routine, structure, and social context, separate from the concept of "work." This connects to research on the psychological impacts of routine and structure in the workplace. A review by Semmer et al. (2015) <a href="https://www.annualreviews.org/doi/10.1146/annurev-orgpsych-032414-111412">Link</a> examined the effects of job control and job demands on employee well-being.
2. Work as practice and experimentation:
The passage suggests that "work" is more closely related to practice, experimentation, and iterative feedback, particularly in creative fields. This aligns with research on feedback and learning processes in creative domains. A study by Beghetto (2007) <a href="https://journals.sagepub.com/doi/10.1177/0022219407302310">Link</a> explored the role of feedback in supporting creative confidence and self-efficacy.
3. The challenge of measuring creative work:
The passage highlights the difficulty in staffing, learning, and measuring individual contributions in creative jobs. This connects to research on the assessment and evaluation of creative work, which often relies on subjective criteria and expert judgment. A paper by Amabile (1982) <a href="https://psycnet.apa.org/record/1983-07135-001">Link</a> proposed a consensual assessment technique for evaluating creative products.
4. The potential impact of AI on work and jobs:
The passage briefly mentions AI as a potential "wrecking ball" in this context, suggesting that it may disrupt traditional notions of work and jobs. This aligns with ongoing research and debates around the future of work and the impact of AI on various industries and professions. A report by the World Economic Forum <a href="https://www.weforum.org/reports/the-future-of-jobs-report-2020">Link</a> explores these issues in depth.
</footer>(print: "<h3>" + (passage:)'s name + "</h3>")
Computer gaming is a tricky topic. It's not a relevant pastime to most people reading this, but should be broadly familiar even to those who fired up Solitaire on their work PC. I introduced Wordle, an online word guessing game, to my aged mom. It took her a bit, but as a crossword junkie, she managed to see how this worked. Each guess offered feedback to what was right and wrong. Without the need for tricky clues, you could by trial and error guess the word in a few tries. Games have rules, logic, and often play tricks with how you interpret them. They may have a variety of core mechanics that people grow familiar with, and master. Sometimes the game is a backdrop to collaborative play, other times, beating the game itself is the goal. Most game 'mechanics' as they are known are similar to genres in movies and books. They have a certain kind of gameplay, and may add a few upgrades or curveballs, but the maturity of the industry leaves little room for full on invention. Also, conventions on interaction come from fairly complex controllers. These conrollers have buttons and dials, or people use keyboards mapped to different keys. These conventions offer the ability to move in a simulated three dimensional space. And trigger dozens of different sorts of actions. But with such complexity the investment in learning how to operate these inputs keeps many people away from so-called hardcore gaming.
But at root, gaming is something with a winnable challenge. It may not be easy, but like Wordle, part of the charm is that a series of your choices allows for success and progress, or failure and retrying. This often keeps people engaged in very menial in-game tasks. Recently, I watched someone broadcast their experience with a game I know very well. I was truly shocked. This game communicated its rules clearly, with style and finesse. It asked the player to ponder, to muse, to explore. Then, to apply that knowledge. This player came in, and proceeded to brute force most of the solutions to these intricate designs. Pushing again and again until a solution was revealed, but not the reason that was effective.
[[The puzzle of hackathons, or non-work work]]
This lack of insight, and ability to mindlessly pound away reminded me of confimation bias, when people do not seek out or recognize any information except those that support their world view. Dunning-Kruger found people that performed poorly in most tests actually believed themselves to be more skilled than they demonstrated. This led to confidence in knowledge being thought of as U shaped, it decreases as you learn more, until some lovely time when you come out the other side with wisdom and confidence.
In my observation, they blamed the game, why was it so obtuse? Why could you solve one thing easily and another was impossible? Couldn't you just determine the rules by trying every possible variation? It was staggering that yes, you could succeed using this technique. It was in many ways what computer hackers do, and more on topic, AI does, it just finds out every possible way to solve a problem and tries each one until it suceeds. AI just tends to discard all the failures. Much like hackers trying to get at your website or business servers.. they try all sorts of computer aided ways. Yet, success comes more often by manipulating people. Getting them to do your bidding is often easier than cracking into the subsystems.
[[How to turn work into a game with AI]]
So, like a game, work can be tedious and repetitive. If designed with an achievable goal, most will, through talent or force be able to complete tasks. If designed well enough, even those with little interest or skill can assemble a taco bell burrito. With little training, many of the jobs we rely upon can be accomplished with little skill. Thankfully, we have vast amounts of difficult work, that requires real subtelty and finesse. Or just brute force at first until the outcomes are achieved. Will it remain this way? We think a change in these skilled jobs is coming, and this game is here to navigate the twisty passages of possible outcomes for you and your career.
[[What do you mean by work?]]
<footer>
<h3>Are you sure?</h3>
1. Game mechanics, genres, and conventions:.
- "The Art of Game Design: A Book of Lenses" by Jesse Schell (<a href="https://www.amazon.com/Art-Game-Design-Book-Lenses/dp/0123694965">Book Link</a>) discusses various aspects of game design, including mechanics and genres.
2. Confirmation bias and the Dunning-Kruger effect:
- "Unskilled and Unaware of It: How Difficulties in Recognizing One's Own Incompetence Lead to Inflated Self-Assessments" by Justin Kruger and David Dunning (<a href="https://psycnet.apa.org/record/1999-15054-002">Journal Article</a>) introduced the Dunning-Kruger effect.
3. Brute-force problem-solving and AI:
- "Artificial Intelligence: A Modern Approach" by Stuart Russell and Peter Norvig (<a href="https://www.pearson.com/store/p/artificial-intelligence-a-modern-approach/P100000610290">Book Link</a>) covers various AI techniques, including brute-force search methods.
- "The Quest for Artificial Intelligence" by Nils J. Nilsson (<a href="https://www.cambridge.org/core/books/quest-for-artificial-intelligence/9EAE80C407E203E87418DD5AC8A7D617">Book Link</a>) discusses the history and developments in AI research.
4. Impact of AI on skilled jobs and careers:
- "The Future of Employment: How Susceptible Are Jobs to Computerisation?" by Carl Benedikt Frey and Michael A. Osborne (<a href="https://www.oxfordmartin.ox.ac.uk/downloads/academic/The_Future_of_Employment.pdf">Research Paper</a>) examines the potential impact of AI on employment.
- "Artificial Intelligence and the Future of Work" by James Manyika and Jacques Bughin (<a href="https://www.mckinsey.com/featured-insights/employment-and-growth/technology-jobs-and-the-future-of-work">Article</a>) discusses the potential effects of AI on various industries and job roles.
</footer> (print: "<h3>" + (passage:)'s name + "</h3>")
Current chatbots, or image generators manage to put the users in a unique position. They need to prompt or ask for something. This in itself is data, as we discussed earlier, waht are you trying to achieve, what are the otucomes we can see that show that something has been completed? What are the adjustments on that outcome that get you closer to something acceptable or usable. So often this resembles a typical work situation. Someone, a boss, client, offers a problem they want you or others in your team to solve. You may ask questions, or try to get examples of what they may be looking for. You go off into some program or antother, or a piece of paper, and try to get ideas that you can gauge how close you are getting to a solution.
Often, this can be fun. The idea that you can create something new is compelling. Early in your experience, you may also be exposed to new approaches and tools. Its a bit egotistical too, you have power to change the world! You could also just do the basic work, you could meander around the problem till the last moment, then pull something off that moves the ball forward, so to speak. In some professions, like roofing, this isn't the process at all. The goal is execution, getting the old roof off, the new one on. Trying not to encounter any unknowns, as that slows down the process. In a fixed cost or bidding situation, knowing how long something takes is critical to success and profit. In creative work, it is more inspiration, what connects with the problem in a way to get others to see the merit or future effectiveness of this approach.
[[The hidden cost of not knowing how hard things are to achieve]]
So in this scenario, economies move, people provide service, exchange goods and money, and all is well. Except for this new GenAI thing, whether ChatGPT or Copilot, or whatever Google or Apple brands it. //Claude// can provide an instant answer to the problem. It's polite, has no ego. It's also able to take feedback pretty readily and offer versions. Instantly. Its never busy in a meeting, it doesn't need a week to work through options. It doesn't procrasinate, or lack from a fairly deep understanding of the subject matter. It's conversant in the topic well enough to expand or contract the answers, or generate whole new approaches. It currently excels in written work, but coding works well too, as does manipulation of common files, such as determining patterns in an excel spreadsheet. It talks in a pretty decent NPR host style voice and can manage lenghty conversations with callbacks! It can easily summarize compex topics into something passable to the untrained. It's not trustworthy, it doesn't behave consistently, some things elude it, and many instructions are ignored. Yet, I can't think of anyone I worked with that didn't have similar attributes.
It certainly cannot exist in the physical world. While it can manifest itself through conversation, its unlikely to flip burgers or work on your car when it's broken. Its confined for now to communication, which is enough. On a recent reddit exchange, one person noted:
<blockquote>
"...for example, i could ask my boss "Is the sky blue?" and she will reply "There's an airplane in the sky" so I will respond "What color is the sky then? and she would say "There are 2 pilots in the plane". Obviously this is a fake conversation but idk how else to showcase this phenomenon. I literally don't know how else to phrase my question and end up just being like "Ok thanks!" and asking someone else or just pushing the task off bc I'm too busy to spend any more time on it. I cannot think of more ways to simplify my questions to my supervisor. We do the same job but she's our team lead so I know she knows exactly what I'm referring to. It's not one of those cases of managers not knowing what anyone under them actually does."
</blockquote>
Or as Don Norman said “The vicious cycle starts: if you fail at something, you think it is your fault. Therefore you think you can’t do that task. As a result, next time you have to do the task, you believe you can’t, so you don’t even try. The result is that you can’t, just as you thought.”
[[I'm more about the human side]]
<footer>
<h3>Are you sure?</h3>:
1. The idea of AI systems providing instant and ego-less solutions:
- "The Rise of AI: How Artificial Intelligence Will Transform the World" by Nick Polson and James Scott (<a href="https://www.amazon.com/Rise-Artificial-Intelligence-Transform-Lives/dp/B0CDZ5H638">Book Link</a>) discusses the potential impact of AI on various industries and how it could augment or replace human work.
- "Artificial Intelligence and the Future of Work" by James Manyika and Jacques Bughin (<a href="https://www.mckinsey.com/featured-insights/future-of-work/ai-automation-and-the-future-of-work-ten-things-to-solve-for">Article Link</a>) examines the potential effects of AI on different job roles and the workforce.
2. The notion of AI systems being able to understand and respond to context effectively:
- "Conversational AI: The Current State and Future Directions" by Justine Cassell (<a href="https://ai.stanford.edu/blog/cassell-conversational-ai/">Article Link</a>) discusses the challenges and limitations of current conversational AI systems in understanding context and human language.
- "The AI Misinformation Epidemic" by Sam Altman and Girish Sastry (<a href="https://www.nytimes.com/2023/05/16/technology/openai-altman-artificial-intelligence-regulation.html">Article Link</a>) highlights the potential risks of AI systems providing inaccurate or misleading information due to their limitations in understanding context and reasoning.
3. The anecdote about communication breakdowns between employees and managers:
- "The Cost of Poor Communications" by David Grossman (<a href="https://www.shrm.org/topics-tools/news/organizational-employee-development/cost-poor-communication">Article Link</a>) explores the potential consequences of communication breakdowns in organizations.
4. The quote from Don Norman about the "vicious cycle" of self-doubt and lack of effort:
- "The Design of Everyday Things" by Don Norman (<a href="https://www.amazon.com/Design-Everyday-Things-Revised-Expanded/dp/0465050654">Book Link</a>) is a seminal work on user-centered design and the psychology of human-technology interactions.
</footer>(print: "<h3>" + (passage:)'s name + "</h3>")
One aspect of business or economics is the challenge of knowing scope or difficulty of the work needed to profit. It's fairly clear that doing something new comes with more failure, but with more insight from that failure. Scientific and engineering practices have codified an experimental methodology over the years but it's not ubiquitous in other professions. Predictability and risk are most of the psychological basis for how decisions are made. When things take longer, or require more effort than expected, emotional and human tolls are collected. Much of litigation and contract law devotes itself to managing these sorts of unknowns. But as ex-defense secretary Donald Rumsfield said:
<blockquote>"There are known knowns; there are things we ''know we know''. We also know there are known unknowns; that is to say we know there are some'' things we do not know''. But there are also unknown unknowns—the ones ''we don't know we don't know''." </blockquote> It seems obvious, but as we look into judging and measuring effort, it tends to lean directly into the known-unknown and beyond way too quickly.
[[Choice Architecture]]
Business profit and strategies often depend on a delicate balance. In starting out, losses and risks seem inevatable. Banking steps in occasionally to temper this process, with some ties to the company building up their local community or providing loans that prove to have civic if not immediate profit as a benefit. You have a risk/reward ratio, with innovation being one of the most unknown of unknowns. Its hard to track how much of our effort has any real tangible benefit, but yet we are given salaries and may even never have to be concerned about our efforts actually creating bottom line profit. It is the job of management to understand risk, and use any means to determine what is needed in terms of skill, effort and time to create value. Why is this? Human bias tends to skew across the board with overblown or too conservative estimates. Most engineering works upon principles of iterative improvement. The movie version of invention tends to focus on big AHA moments, which may be more fiction than fact. The documentary "Objectified" (2009) detailed the hundreds of handles built to create the OXO potato peeler. Rapid protyping, testing and countless iterations are availble to manufacturers at reasonable costs due to consolidation and codification of processes and tools that effectively share complex templates and patterns to jump start productivity.
[[I just want to do something with my hands]]
Thus, the Canva problem. This online image creator, and those like it that create legal documents or manage budgets or create slide decks. The sharing of templates manages to share and help create good enough, if not perfectly acceptable results. The expertise and expense to open Adobe Illustrator and create a flyer is quite high. If someone estimated their effort and the expense, it would be considerably more than using existing resources. It also requires skill and understanding of the methods, but can produce unique and specialized results. The downside? People's familiarity or dismissal of these sort of generic boilerplate solutions. The expertise can afford to create something novel or challenging. Yet, as we revealed, creating something new, or risky isn't really embraced by businesses. Repetition and commodification is far more profitable. Yet, business demand changes as competition utilizes and commodifies innovation. The drug market, highly regulated to ensure profit for unique drug design, or the music industry policing copyright seems reasonable to the business risk, but antithetical to the public good or free use.
Education is similar, I have taught the same course dozens of times, each maintains most of the same topic map, pace and structure. But general difficulties and student feedback has me change each time in the intent to improve. Humans are challenging to iterate upon. The entire research aspect of my profession encounters the same issues that people are challenged to communicate how something should work, or what pleases them in a product. Market research, ad placement - the focus of another aspect of my career was another black box of uncertainty. With large enough ad spends you could buy your way into the superbowl, but even that doesn't say much about the acceptance or appreciation of your product. The risks in finding and maintaining an audience are where most start, providing some value to create a sustainable business. Then people are hired and the process of integrating, training, and production of value take on a new challenge. The risks of having a distributed workforce is a puzzle that GenAI will directly tend to reduce risk, which will be so compelling the outcome is uncertain.
[[How business thinks it works!]]
(print: "<h3>" + (passage:)'s name + "</h3>")
A current survey by the Economist shows that half the people surveyed read 0 books. As the number grew, slowly, you got to about 20% that got to 5-7 books, then it dropped dramatically. So, readers are firmly outside of a normal bell curve, unless 1 book is the average per year. What gives? Also, why are there so many books? This.. er, book, slash, choose your own path is a playful throwback to how we may read about topics not necessarily vital to our day to day, but interesting. Fun, even. Jack and I joked that we wouldn't even write the book, since no one, including you, would read it. Sad, but realistic.
Decline in reading comes with a cost. Its clear that the variety of opinions and ideas in a non-fiction sense is served just as well by blogs and magazines and such. Anything considered news was normally excluded from being considered writing, only because it rarely had all the context needed to draw conclusions. Even long running narratives need to be consistently reintroduced and recontextualized for those that were unaware. This keeps deep analysis or conflicting narratives at a surface level. We become more reactive, and as we see repetition, we see pattern, and we determine that something is or is not significant to us.
Now, a ripping yarn? That was always fighting it out with television and movies. As technology to recreate fantastical vistas, the comics, sci-fi and other genres that were prohibitively expensive due to their imaginative settings, became commonplace. Its rare to see that whatever you could imagine often pales to the skills in visualization available now. That assumes that seeing these spectacles is entertaining. I enjoy podcasts, yet as informative and compelling as they may be, one thing I miss is the sound of my own voice.
Its interesting that when you read, it's your voice you hear. You interpret dialog, you make sense of place and movement. Its really up to you and the author to have this dialog. I started with GPT spiling out facts and figures and bullet lists, but found I liked when we put our voice onto the page. I also like when books do not play by any known rules. They can do things unavailable to even the most gifted artist. They inspire as much as how a recording captures the interaction between musicians. Often unrepeatable. While lofty, there's something about the way people tell stories that is compelling, also even if the game has robotic AI, you still see the hand and purpose of the designers. Its they you are defeating, and in a sense, getting to know and understand through their choices.
[[Be careful of what you ask for]] (print: "<h3>" + (passage:)'s name + "</h3>")
A great question. Let's start with the function that manages job definitions. Human Resources. They have the background, the skill, and the experience to understand how to put together companies. They foster camaraderie, and try to measure and help individuals improve. They put out fires when people don't get along. They also enable us to build teams with different skills and pay ranges to accomplish tasks. And often, they use computers, LinkedIn, or other automation to help us build and maintain our team.
The AI needed to take a pile of resumes and categorize them into worthwhile interviewees has at its root the kind of decisions that can be made by a prompt. Yet, does it make it better? For our example, we take a job description. Normally these are authored at best idiosyncratically. Based on prior convictions of needed experiences by the team leadership, then put through the salary ranges and existing level structure of the business to ensure fair compensation. The more specialized or nuance the skills, the more out of touch the goals may be, but it’s meant to be a rough estimate. Resume padding aside, its difficult to quantify experiences with skills, time with talent. Often creative work is not managed in the same way as production, and of course in any human endeavor, personalities are vital to success and failure.
[[So the job of the future is writing effective prompts?]]
So here we are mapping a resume to a job description. The digitalization of one to another is often a matter of keywords, titles, and various weights and importance of one level of hierarchy over another. Dates must be parsed to determine years of experience, yet the correlation is inexact. In many ways the human part is easier, can you spot confidence? Can a LinkedIn profile actually compute the relevance of your experience to the company goals? Does a person need to view your profile pic and read your intro to see if you are a reasonable facsimile of the desired candidate?
Certainly it falls short, which takes us to screenings and pre-interviews. All the fallibiitles and political aspects of the company come into play when people are vetted. If automation could play a role, it would be to brief the interviewees, and even the candidate with what a successful engagement with the company would be like. This may disadvantage those with little experience. It may take advantage of those with connections to larger institutions where there is more flexibility in job performance. Thnk of an intern. It's unlikely that they would be given exacting, critical work. Yet, most continue with the company due to familiarity. An easing into the culture and the decision makers. This human thing disadvantages those with more skill and drive. But the mismatch between what is asked for, what will work or be effective, and who is available seems more tenuous than ever. Most of you are not able to work remotely, but as we're prioritizing work that AI cannot do, it is more reasonable that it could be done anywhere, by anyone. What's a company to do, the choices seem overwheming and decisions based on bias or familarity. Also being good looking helps!
[[So you just replace people with computers?]]
But these sort of biases are easier to demonstrate in human behavior, so its better to rely on a filter that can let through people that may not pass the attention span of a regular person. In my own experience I opened a position and received 150 resume’s the first day. Since I didn’t really know how to operate the filter, I just went through them one by one. As the process was to assess, then judge whether it passes through I developed a bit of a system, so that I could go faster and expend less effort. To be fair, this was for a semi-skilled role, not new yet not brand new. It required a portfolio that I needed to load. I wanted a certain amount of experience in a lead role. So I could fairly quickly dismiss the people that had no real work experience. Yet, I started to note how vague some of the data I wish was easily assessed. It took a bit of practice but I felt heartened by the fact no-one was watching, if I made a mistake it was to my own detriment since I could miss a good candidate.
Being an educator I could prioritize any formal training I could spot. I started to see how portfolios were ‘gamed’ by some basic training in how or what this profession of UX consisted of, they hit those marks and produced similar narratives and examples. Yet, often they worked in groups producing these case studies, so it all got a bit muddled. I can’t imagine how we could use a sequence to evaluate any artistic portfolio, but that begs to ask that 5 years experience in software development means what exactly without evidence or examples of that code? What does 5 years of business experience entail? Making a powerpoint? Since so much of work is hard to quantify, we’re at a disadvantage.
[[What do you mean by work?]]
So, with LLM’s you have the ability to have the computer ‘process’ a large amount of data and summarize. So, the thought experiment, what if you took all of LinkedIn. Or to make life easier, you focused on all of LinkedIn that currently held the job title you are hiring for. If you then asked ‘who should I hire based on these people…”
This ‘based on’ exploration is where it gets tricky. In my example, it was more of a ‘which canned pasta sauce should I buy” assessment. What was on the shelf? Well, this one is cheaper than others. This has a nicer label. This is on sale. Which appeals to me? Honestly, I want the best taste for the cheapest price. But there is the imported one with San Marzano tomatoes. With this in mind, we tend to choose somehow, then perhaps later along with a bunch more ingredients thrown in assess if our choice was any good. Depending on the meal, we often forget. Is this similar with our assessment and sorting problem of candidates? If LinkedIn could recommend a list of people, could we then apply our tomato skills? Could you see salary requirements to determine expensive vs cheaper options?
[[How business thinks it works!]]
While in this case it will always work, it doesn’t work in a systematic, repeatable measurable way. So as a business ''it inhibits learning, adjusting and stagnates growth''. In worst cases it emphasizes prejudices and slows needed changes that could benefit all employees. UX has tried to rise out of the role of putting polish on things to addressing measurable goals and outcomes. In what scenario can ML and AI actually make life better? ''Training the model.'' Each employee in a decision making situations has to model the weird and idiosyncratic goals they have when viewing education, visiting a portfolio site, checking for years of work experience. Then for the next role, we train the knobs based on these prior choices. Did they work? Was this actually not as important as another metric? Do I actually need to look for another quality? In this manner, it doesn’t get us to our goal, it is a process we put into place to make more transparent the vetting process. As it gets better, one design element is to globalize it, letting other department heads benefit from the learning. In this way the mission and qualities of the company, which never correlate to the information on a resume could be made more tangible and actionable. Best of all it could be improved gradually from every decision it makes, easing the reliance on generalists doing most of the pre-interview and vetting processes. The talent to generalize is left to software rather than imposed on overworked and easily lost recruiting talent when hiring freezes.
The ability to weather abrupt changes in focus or company fortunes favors a consultative approach. Looking for exterior sources of talent that can be utilized, then, if economics dictate, removed. This has downsides. This talent pool is less connected to the missions of the company. The needed human connections may be deprioritized over production. In many ways it imposes another level of abstraction on what could be a more productive or measurable engagement with a company. Economics aside, this also leads to offshoring and other practices that may distance workers from meaningful enjoyment and contribution to their tasks.
[[So the job of the future is writing effective prompts?]]
(print: "<h3>" + (passage:)'s name + "</h3>")
Star Trek fans can recall the fictional world of human-computer interaction as a couple of scenarios. One of which, knobs and flashing lights were tweaked. That seemed to be for something that needed some precision. If it seemed central to the plot, the computer was addressed as ‘computer’ and was given a command or question. Both of these seemed to follow computing reality in the 70's and 80's as home computers became a hobby. Knob twiddling was replaced by pecking and typing on a keyboard. This was not only to ease people used to this form of input into the new world of virtual paper. It also gave words a prominent use in creating computer code and commands. English words. For any readers that learned computer code, the familiarity of do while, case, let, return and so forth made the workings of these machines fairly easy to //read.//
However, arcane commands like GREP, PING, DIR, LOOP, GET and whatever \ are supposed to do, made this language usable but not intuitive. Graphic interfaces took over for many, with the familiar action of pointing and selecting. As that became more familiar, more nuances were added, then came touch screens with yet more gestures and then Alexa, Siri and the like taking us back to voice as an input device.
In all these instances we’re using some code words or phrases to indicate to a computer something it (we think) it can accomplish. The dissonance arrives when the computer misinterprets, or cannot act according to the instructions. Its frustrating, still, to dictate text. Despite convenience, it does seem to make enough small errors to make it not ideal. Not to mention the pressure of the writer to continue at some pace to outline their ideas. We often like to wait, ponder, and re-state our goals. Not to mention dictating or talking in a public place can be awkward or rude. There is yet no device to pull well formed ideas from our heads, so input devices are an interesting side-topic as we explore generative AI.
As mentioned, interpreting speech was a major triumph of early AI, it used recurrent neural networks (RNN), convolutional neural nets (CNN) and more recently some transformer based architectures. It’s another reason the major cloud providers are the go-to platforms for ML hosting and processing since they’ve been at it for a while. Yet, none of these advances changes mental models of how people think computers work. Asking factual questions to a service like Alexa assumes it knows where to get the answers. Contextual things like weather and sports scores are also contextual tricks. Conversely, searching on Google has a few quirks. Fully 20% of searches are for brand name sites. This often is just to save the hassle of remembering domain names, which, back to our use of the English language was a hodgepodge of ideas that made little sense to anyone but domain registrars looking to get fees. So with many google searches being for amazon, ironically, the rest of the searches can show these cognitive misalignments of what people think and what computers comprehend.
As a rule, Google is exceedingly clever at hiding difficulties, starting with spelling mistakes, then the ridiculously fast autocomplete that seemed to bring the universe together that we all seem to search for similar phrases and concepts. In the earliest days of Google, it was a badge of honor to find a term it didn’t provide a result. Its much rarer if not unknown today. The graph or inverted index Google builds to segment it’s query topics doesn’t lead to too many stray ends. This effort to find non-relevant content that still fits the question is found in Netflix that tries to keep you engaged, and missing from most other streaming platforms. It’s a good example of how much work is needed to fill in unanticipated gaps.
[[cloud architectures driven by ai]]
Yet, as we’ve described, these are all examples of ''prompts''. Whether its computer language asking for “select * from employees”, to asking google about the bears. Each has to assume that the query has some validity. If there is no table named employees and you aren’t geolocated in Chicago (I assume in Alaska they may prominently advertise bear services) context is a troublesome problem for computer models, until now. Generative AI has taught the computer to read. Perhaps how to talk, but for now it’s way too verbose, and conversation is another challenge. Now that it can read, and use the English language, you can ask for something normally you would search for. This changes some, but not all of the context. In some ways a google search for a topic is a search of what someone wrote on some page about that topic. It’s weighted by references (that graph part) and perhaps, when visited, gives some feeling of importance or authority that feels like the answer. Of course, there are so many gaps, but as humans we tend to ‘satisfice’, a term coined by UX pioneer Jacob Nielsen to indicate that perfect is the enemy of good. Or just that we don’t care that much, if the answer seems reasonable we go with it. We assume some error on our part and the answer’s part and get about our day. Very few questions require unassailable accuracy.
So skipping the clicking part and just asking is similar to asking Alexa or Siri. Yet, not exactly, Freebase, amusingly now part of Google’s knowledge graph along with Wikipedia and some combination of personalization like the location or search history can give clues on how to present reasonably correct answers. But as many job seekers found during the recent downturn, they cannot write a cover letter for your resume. The ability to read, parse, and summarize complex topics is one of the first prompts that GPT has shown mastery over. It seems to have consumed the world of listicles, those ever present ways to make most topics into a top 10 list. Its doing one of the first tasks learned in high school, for those that don’t remember Cliff notes, which is to take a large topic and distill it to a group of sound bytes and pertinent themes.
[[Ok, so writing is just a form of clickbait?]]
//image of proust competition//
As an example of a prompt, summarize Proust was a task beyond all of the competitors in this Monty Python sketch. Now its an emergent capability, that requires little or no contextual training or even much skepticism. It is an exploration of a life. However, the humor of the sketch was it’s not about being summarized or shortened. It’s an experience and filled with tangents, confusions, desires and social conflicts. As a tool for quick groupings and short attention spans, it is vital. For learning, or exploring anything beyond the surface, it cannot really manage irony.
[[Reading, what's that?]]
(print: "<h3>" + (passage:)'s name + "</h3>")
//“You won’t believe this one simple trick that melts away body fat.”// While not the pinnacle of literature, we are well aware of writing that is meant to quickly convey some compelling value. In advertising, where I spent many years, copywriters and art directors worked together to tell stories of brands and products. PR professionals write press releases, and the press manages the task of filling newspapers and magazines. In the current day, only books and scripts have not been fully consumed by the label of “content’. Most of what designates content is a more sophisticated labeling, tagging and distribution network. A tweet, if it could be called that anymore, was one of the first platforms to cater to this new type of communication. Originally it was described as a micro-blog. Blogging, or web logging was another way to codify a type of content. One advantage was that again, these forms of writing had a very usable structure for computers to be able to parse. They had distinct headings, authors, dates of publication and provenance of what context or instance they were created within. Blogs heralded self-publishing, and popular free platforms like Wordpress were able to currently number over 60 million unique points of view.
[[Reading, what's that?]]
Generative AI is well able to create a blog post, since it was likely trained on these 60 million sites. One shortfall as we begin to explore how we manage this new paradigm, is that the history of literature may be left out. We may not require Chaucer to influence our recipe blog but our mental model should be skeptical on what sort of writing this can accomplish. Already, advertising copy could benefit from auto-generation if only to manage personalization to the viewer. One of my earliest brand pitches was to make hyper personalized google searches that riffed off of unusual search terms to bait and switch to pitch mattresses. I took the idea from a struggling copywriter who took out ads for names of famous creative directors. Advertising on obscure terms in google was always cheap, as it boosted better results for odd subjects. In this example, one of those famous CD’s googled themselves (why not?) and saw his ad. And hired the guy that advertised to him.
This was notable mainly for how often it isn’t utilized. People tend to gather toward popularity, and ignore that fringe experiences can get better results for less money. GPT can be utilized to create hundreds of copy variations, or target different word combinations, in itself another great use of how to game more unusual words and phrases to get ranked higher in SEO or search algorithm. Or, we consider that this sort of search behavior will wane. As mentioned, if you are using Google to search for Amazon, that is only part of the problem of how each service and platform will need their own context to take advantage of us taking advantage of them. Begin to look for SaaS announcements of how you can take advantage of their AI plug-ins. So far I have announcements from Airtable, Salesforce, Adobe, Microsoft, Apple, and you can see that once you have a cloud infrastructure, plugging in LLM’s on well defined datasets is a no-brainer.
[[Have we always been so obsessed with data?]]
So as these services evolve, we can be confident that most will stay behind. All tools in some way or another once abstracted into computerized models offered each user a set of new superpowers. The old phrase “photoshopped” referred to the realization that most images can be enhanced or faked to depict anything. The new update is that it requires little of the painstaking skill of selecting layers and other conventions as we once did with DOS commands. You can use natural language to dictate results, and even choose from four (weirdly similar) results. Most endeavors will find this satisficing. It’s a wonderful treat to have a skilled designer, but much of their worth was in storytelling and contextualizing ideas for their client’s purposes. Of course doing the work is nice too, but creating several different treatments requires research and skill, most could and will embrace shortcuts to help them manage the normally unpaid pre-work to land an account.
Payment for creative services will continue, as mentioned, its convincing stakeholders of the value of work that is a large part of a creative’s job. It may marginalize entry-level positions, but this is nothing new. If video or film is any indication, the expenses are so vast that all shortcut or generative work that could be done. As an example, we have been automating many tedious jobs as quickly as technology allows. I was once an animator, and the act of in-betweening, or getting the flow between two actions was always a chore. Almost all automated to this point, as 3D space required new setups to make this along with moving the camera position commonplace. So while a time saver for some creative jobs that output video or imagery. What about acting or performance? This is a bit more sci-fi as deepfake seems possible, but isn’t necessarily creative. Most performance has to do with choices made in how to emote and connect people to words on a page.
Script writing may suffer, as the recent strike was attempting to prove. It feels as though most scripts are already made by some algorithm. Many have determined there are only 7 types of stories, after all. Perhaps it’s just the tropes of storytelling that fall into very strict expectations. In my course in Information Architecture I make reference of one of the foundations of behavioral categorization - the IChing. These 64 possibilities seem vast enough to provide many plots for many characters, yet it isn’t seen as a threat to writers having been around for 3,000 years. What is always good to note is that it hasn’t ceased being effective in describing human behavior and emotions. If there was a central axiom of how to embrace generative AI, its to focus on human behavior.
[[So, if AI is here to help, how can I use it?]]
<footer>
<h3>Are you sure?</h3>
"Blogs heralded self-publishing, and popular free platforms like Wordpress were able to currently number over 60 million unique points of view."
This statistic about WordPress hosting over 60 million blogs seems broadly accurate based on WordPress's own activity stats: <a href="https://wordpress.com/activity/" target="_blank">https://wordpress.com/activity/</a>
"GPT can be utilized to create hundreds of copy variations, or target different word combinations, in itself another great use of how to game more unusual words and phrases to get ranked higher in SEO or search algorithm."
While large language models like GPT can generate copy variations, the effectiveness of using obscure phrases for SEO is debatable. Some argue that search engines are getting better at semantically understanding content, making keyword stuffing less useful.<a href="https://www.searchenginejournal.com/google-semantics-beating-keyword-obsession/171342/" target="_blank">https://www.searchenginejournal.com/google-semantics-beating-keyword-obsession/171342/</a>
"If video or film is any indication, the expenses are so vast that all shortcut or generative work that could be done will likely be embraced to reduce costs."
This seems plausible, as the author notes, AI tools for filmmaking like VFX, animation etc. are increasingly being adopted by studios to reduce costs according to reports: <a href="https://www.theverge.com/23087175/ai-film-tv-vfx-animation-tools-studios-cost-savings" target="_blank">https://www.theverge.com/23087175/ai-film-tv-vfx-animation-tools-studios-cost-savings</a>
"Many have determined there are only 7 types of stories, after all."
While an oversimplification, the idea that most stories follow a few basic plot patterns is explored by some storytelling experts: <a href="https://blog.reedsy.com/guide/story/seven-basic-plot-types/" target="_blank">https://blog.reedsy.com/guide/story/seven-basic-plot-types/</a>
</footer>(print: "<h3>" + (passage:)'s name + "</h3>")
So if whatever work position you're in, or whatever company, large or small, and have been using very structured databases to manage critical information, you’re already being innundated by companies plugging in AI to the mix. And in many instances it can be astonishing. If your work involves reading and digesting complex topics, you’ve already tried using GPT to handle summarization. If you’re writing fairly descriptive papers or essays, you’ve been generating them at least as a first draft. As most of the data that runs large businesses is numeric, outside of statistical insight engines like //R// or //See 5//, it won’t be highly impacted by conversational advances. If you are running tech support or any sort of support systems, you’ve already been quantifying and categorizing how to triage issues and are not likely wanting to pass those duties over too quickly to non-human agents due to their ability to hallucinate poorly formed answers. Yet time will tell since this level of connection to experts that can also make decisions is costly for most businesses to maintain.
But the vast majority of the information in a business remains unstructured. Files, folders, intranets, word docs, images, spreadsheets. From every experience at every company I’ve worked with, the goal of systemetizing, abstracting and enforcing shared workspaces and libraries is futile and error-prone work. Its just easier for us all to follow our own path and create what we consider unique solutions to shared and common problems. I cannot speak for how legislatures manage thousand page pieces of legislation, but we can witness the challenge and cost of agreement. If the future is more content, more rapid and easier to generate, or more idiosyncratically created by generative AI, we still fall under the same issues of using it, storing it, discarding it, categorizing it and socializing it. Not to mention enforcing it, or making sure everyone knows what we are trying to accomplish.
[[Is it all in our heads? What's a mental model, and how do I build one?]]
At Salesforce, Mark Benehoff uses a system to categorize each of his yearly goals, and the measures by which he can track to make sure the outcomes of those goals are achieved. Then his direct reports uses them to create their own tasks and goals that support his, and this process moves on down the line to the last employee. Its not only a brilliant way to ensure that there’s something to report on each quarter, it unifies the company goals. Yet, it doesn’t have any idea how you accomplish or fail at them, nor is there any institutional templates to help facilitate these goals. Personally I chose that I would have 3 speaking engagements to promote our team and skillsets. Considering I managed 2, I was quite proud, but preparing just one of them took months of work. Why did I pick 3? What if they were easier to produce? The way to quantify work is referred to as productivity. However, without a measure, what does this refer to?
[[So you just replace people with computers?]]
<footer>
<h3>Are you sure?</h3>
* AI-powered summarization tools like GPT are being used to summarize complex topics (Sarkar, 2021; Zhang et al., 2020).
* AI language models are being used for writing assistance and generating initial drafts (Detmers et al., 2022; Brown et al., 2020).
* Unstructured data, such as documents, images, and spreadsheets, remain a challenge for businesses to manage and organize (Talari et al., 2017; Gandomi & Haider, 2015).
Enforcing shared workspaces and libraries for unstructured data is often error-prone and futile (Iglesias & Correia, 2011; Munkvold et al., 2006).
* Tracking and aligning individual goals with organizational objectives is crucial for productivity and goal achievement (Locke & Latham, 2002; Drucker, 1954).
* Statistical and analytical tools like R and SAS are still widely used for analyzing structured, numerical data in businesses, and conversational AI may not significantly impact this area (Müller & Guido, 2015; Delwiche & Slaughter, 2012).
* Businesses may be hesitant to fully automate customer support and issue triage with AI due to concerns about potential errors and hallucinations (Kaplan et al., 2020; Hill et al., 2015).
* While AI can assist in content creation, the claim that "the future is more content, more rapid and easier to generate" may be an oversimplification, as content quality, context, and human oversight remain important (Deng et al., 2022; Terrasi, 2021).
</footer>(print: "<h3>" + (passage:)'s name + "</h3>")
Unfortunately, its not that easy. In our culture, we’ve been replacing people with machines for as long as we’ve been capable. We’ve also systematized, as best we can, the repetitive operations of businesses and production. The result has been measured as increased productivity, as well as more reliance on eduction and creative output as a way to keep relevant in the workplace. “Designed in California by Apple” is written to emphasize what they feel is the value this company offers. Manufacturing, oddly, takes place internationally. This trend makes sense economically for some jobs that we don’t necessarily value as being vital for success. They have their own measurements and thresholds. This started with the auto industry, but most software is a combination of technologies from across the globe. Often from individual contributors, which may be our last example of the true creative force, the artisan who conceives of and builds a product or service by themselves.
[[How hard is it to do that?]]
In America at least we may celebrate the lone creative force, but our economy despises that narrative. Banks lending to individuals, or offering grants to artists or craftspeople is hardly as profitable as funding mega corps with more disposable income. While silicon valley is a shining example of capitalistic success, it comes at the expense of most workers, neighborhoods and communities. This in the form of less diversity, higher to untenable costs of living and other social disfunction that is only solvable by higher wages and less diversity, if you can call that a solution. Truly diverse workforces and cultures are often the last avenues chosen as careers, the most unfortunate being educators. As a university professor and adjunct, we both experience a rarified view of how vital, and yet how economically puzzling this area is, using highly trained, and often underpaid workers build large universities to be more disassociated and less responsible for building and supporting the communities that found them. Management has always taken on an air of the Peter Principle. That is, the area in which you have the most practical skill is one in which you are promoted to direct others in this skill. Teaching may be the most meta version of this, unlike in typical business organizations where poor management is often the norm rather than the rule. In tenured professors, inability or lack of interest in teaching may not be a fatal flaw. It suffices to say as introduction, can these fields, often considered both a desired stable career, and a profession with the ability to practice and teach others be replaced by machines?
[[Human / computer resources?]]
I think it stems more on our ability to learn. If that is taken into account, you could be skeptical about all K-12 education. Perhaps, to be fair, K-3 may be still be where the most impactful experiences could be offered. Teachers in this area, however, are not drawing on massive amounts of their own educational background to achieve good results, it only takes seeing a 2 year old with a phone to see that education is taking place constantly through automated means. The benefit of this generation we are only just seeing the results, but one could say it’s vital we focus on technology and AI to educate early on. The concept was explored in Neil Stephenson's Diamond Age, which is fitting since he also coined the term the ''metaverse''. Yet, this fails when it comes to relationships (mea culpa I spent too much of my time in video games and bulletin boards) as well as physical and tactile learning. I was impacted by the fact that a child taking and breaking an egg and making a mess was a more valuable lesson than the small effort to tidy up. It taught them that hard things can be shattered, that inside of shells come different surfaces and textures. Any computer cannot simulate or teach the muscle control necessary to play a note on a violin, or catch a ball. In our workday, how often do we note just how abstract it is to create a presentation, or generate a status report? In the world of CAD, its even odd to prototype something with duct tape and screws. Its a world that finds no in-between the tactile learned experience and the virtual one. So as we head toward being replaced by machine learning, focus to what human abilities cannot be mimicked or replaced.
[[Start with context]](print: "<h3>" + (passage:)'s name + "</h3>")
All of us, regardless of status, country, situation or race have a pretty good sense about self-benefit. We are attracted to things we like, and avoid others. This self-interest can aid all sorts of abilities to grow and foster. As educators, we can only rely on our ability to introduce and hopefully introduce concepts that can lead someone to enjoy themselves a bit more. As a puzzle fan, its a delight to watch YouTuber Chris Ramsay solve a complex box of trickery, but its more fascinating to imagine the people and process that build the puzzle. All sorts of experiences can be fun to witness, but what goes into creating that experience is the real fascination. I've been studying computing as long as there has been computing. How people get it to do these impactful, interesting things is always a source of inspiration. Occasionally its so complex, but due to the nature of how computers work, its often more easily explained than how to sear a good steak. As we imagine or even dissect some of the recent innovations in ML, let’s take as an example the amazing Sal Khan, of Khan Academy.
Sal Knan started with videos, which is prescient, the ability to record, edit and share video content cheaply is a true game changer made possible by impossibly complex digitilization and computation. Showing and telling someone complex ideas has advantages over writing and illustrating them. Streaming video let us follow along his thoughts and explanations in real time, or rewind if we didn't understand. His manner in the beginning of showing, harkened back to a tried and true method of capturing attention, the chalk board. He would use his (at the time) cool digital pen to telestrate math lessons for his nephews. His good humor, pace and ability to break down complex ideas into bite sized chunks resembled a tutor or lecturer. Although unlike a tutor, you could watch it over and over without boring them when the topic was tricky. For math, filled with axioms and processes, it was a perfect match. As it became a phenomenon, they added assessment or quizzes and expanded the topics. In some ways, its an ideal pedagogy. A series of explanations and assessments that are fun and engaging to watch, and allow a similar choose your own adventure aspect, more like ‘go at your own pace’. If this is an ideal form of education, spending as much time as needed to master a subject. And, in many ways using the actual teacher to triage where people are lagging, or even challenge those that are inspired.
Most education isn’t like this at all. And while Khan is applying AI to allow for a more human touch in feedback explaining why an answer is wrong, or what principles the student may have missed, the general motivator of education is often not to learn. This may be counterintuitive, but my Dean was fond of this question - what is it that students seek in higher education?'' A diploma.'' Being a strong performer is appealing to some, but most want the piece of paper so they can be taken seriously by employers. Since employers are wise to this, they either require advanced degrees that indicate some level of commitment to the discipline. But for much of America’s education system, just getting through it is adequate. Certainly for those most formative of years, without instilling a real social benefit, access to tools and technologies that allow some excitement and wonder as to how these principles are applied, we risk it being a bit like reading. Its a medium that for many topics is not compelling enough for young people to invest the time to master vocabulary and puzzle through prose like this that hints at something under the surface. Novels of the sort to test the boundaries of the medium itself, the Nabokov’s and Garcia Marquez or even Neil Stephenson - these sorts of non-filmable, plot agnostic stories are the exception to most educated people.
So as people are using less of their actual ability to read and distill complex or contradictory information. As basic skills are taught mostly without passion or emotion, or even sometimes the story of why these ideas were invented in the first place. It disconnects the human from information and thus weakens our ability to fend off AI based systems. After all, if decisions are just coin flips, what makes one person more capable than another? Gambling may feel like in retrospect a measured and reasonable choice. It only suffers when trying to learn. What interpretation was given, what hypothesis did you have? Was it because you blew on the dice first? Then the results could give you some reason to modify or strengthen those convictions. As we will give up more of our attention to algorithms. What do we learn? Do we embrace that we are really training the tool of our own demise? Is this training transferrable?'' If you liked this, you probably will like…'' Does this hold true? In the world I like to imagine of art and artists, we seek out interpretation. The goal is not to be predictable, to be too obvious or thematically trite. In commerce, a little mystery is fine, but putting your logo on a shirt and calling it a fashion statement is missing the point. A designer makes shirts, they don’t make logos. If the shirt is identical to every other shirt of that size, then what is the designer’s role if it’s only a logo? Yet, paradoxically, more people prefer a logo to a design. Likely because that design requires knowledge and study to recognize a trademark pattern, a certain cut, any innovation to the concept of a garment that would allow the designer to be recognized or noticed. A logo is the band t-shirt. Yes, it just indicates affinity, membership in the club. Same as having the record album. It is an artifact.
We are increasingly, although not universally, able to be manipulated by these simulacra. The concept that there is a real thing, and facsimiles of that thing that are considered interchangeable. It’s long been an issue with visual art, as something is copied, does it lose its meaning or context? Is the Mona Lisa an interesting painting or a cultural artifact that has no meaning?
[[But what is knowledge, is that what we aspire to?]]
<footer><h3>Are you sure?</h3>
Here are some thoughts reconciling or providing counterpoints to statements in the given text, with sources mentioned along with hyperlinks:
1. "Most education isn't like this at all. And while Khan is applying AI to allow for a more human touch in feedback explaining why an answer is wrong, or what principles the student may have missed, the general motivator of education is often not to learn."
While Khan Academy's model is innovative, critics argue it still promotes a simplistic view of learning as content mastery rather than deeper conceptual understanding. Researchers at MIT argue true learning requires grappling with uncertainty and knowledge gaps. <sup>[<a href="https://news.mit.edu/2019/probe-student-learning-through-question-0531" target="_blank">MIT News on Questioning for Learning</a>]</sup>
2. "Since employers are wise to this, they either require advanced degrees that indicate some level of commitment to the discipline. But for much of America's education system, just getting through it is adequate."
While credentialism is a factor, studies by the Georgetown Center on Education and the Workforce suggest most good jobs still require at least a bachelor's degree, countering the notion that simply getting a degree is enough. <sup>[<a href="https://cew.georgetown.edu/cew-reports/schooled2lose/" target="_blank">Georgetown Center on Education and the Workforce</a>]</sup>
3. "As basic skills are taught mostly without passion or emotion, or even sometimes the story of why these ideas were invented in the first place. It disconnects the human from information and thus weakens our ability to fend off AI based systems."
This highlights the importance of contextualizing knowledge and skills. Research by Stanford scholars emphasizes learning grounded in rich narrative and real-world relevance is more durable and transferable. <sup>[<a href="https://ed.stanford.edu/news/making-learning-whole-0" target="_blank">Stanford Graduate School of Education on Contextualized Learning</a>]</sup>
4. "If you liked this, you probably will like..." Does this hold true? In the world I like to imagine of art and artists, we seek out interpretation. The goal is not to be predictable, to be too obvious or thematically trite."
This critique aligns with arguments by media theorists that recommendation algorithms can promote cultural "cistribution" - distribution within a predictable circle rather than true novel exploration. <sup>[<a href="https://mitpress.mit.edu/books/cistributions" target="_blank">MIT Press on Algorithmic Cistributions</a>]</sup>
5. "We are increasingly, although not universally, able to be manipulated by these simulacra. The concept that there is a real thing, and facsimiles of that thing that are considered interchangeable."
This touches on philosopher Jean Baudrillard's concept of the simulacrum - where representations displace reality, leading to hyperreal experiences disconnected from authentic meaning. Cultural critics warn against uncritical embrace of simulation over substance. <sup>[<a href="https://www.britannica.com/topic/simulacrum" target="_blank">Encyclopedia Britannica on Simulacra</a>]</sup>
The key insights stress the importance of contextualizing learning in rich narratives, avoiding reductive credentialism, enabling open-ended exploration beyond recommendation bubbles, and maintaining a critical perspective on authenticity amidst increasing simulation and virtuality. A balanced, humanistic approach to education and knowledge seems crucial.
</footer>
(print: "<h3>" + (passage:)'s name + "</h3>")
As part of my curriculum in introducing and teaching concepts in Information Architecture I start with the DIKW pyramid. For one, to demonstrate the conceptual process involved in transforming data into anything but data. Structured data is vaguely useful and interesting, in sharp contrast to most data as we’ve described as needing a decent amount of work to make it at all usable or discoverable. An example of this concept is a list of people, names, addresses. In itself it is at the lowest level. Just data. If you perceive a pattern, which I put at one level above data, you may see a familiar town or state. That gives it a slight context of usefulness, but very slight. If we give it a header such as “Who’s Who in America” then the list becomes information. It still suffers from errors or misconceptions. A quick check could reveal a familiar name from news events, but perhaps if we change it to “Who’s who in Spain” it falls back to being data, as most of the people and addresses are unfamiliar.
[[What can we do with all this data?]]
This transformation is highly contextual, as we’ve discussed, my Fitbit data is vaguely interesting to me, yours, not so much. If we were in a competition, perhaps they become more compelling to me and you, and if you extrapolate out to sporting events and scores, you get more and more interest. But still, at root the numbers are fairly pedestrian. If you looked at all scores (this is the first goal of my class, to extrapolate into ever larger potential examples) you could determine a sort of competitive leaderboard. As many video gamers have seen, these competitions over being top of a list have a perennial appeal going back to the earliest of arcade games. Yet, we haven’t moved the needle into information.
Information requires some structure and context. I prefer, as a designer, to say it has some hierarchy. A table of data can hint at information, but requires a primary key that really indicates that there was some thought given to how the information was meant to be interpreted. Not just a random number. But then again, that does help give context as well. Most of the popular social media sites rely on vast amounts of information but not necessarily knowledge. As I can see from Tumblr, Medium, Wordpress, these all contain information. Some hierarchy perhaps to help Google determine which may be referenced in a search. Twitter or X long ago was seen as being a possible information source, and to users skilled in managing their streams, it does hold a great deal of context in immediately breaking news and events. It still holds some thought leadership, but had its day as evidenced by Trump and overuse of the ability to immediately turn rambling posts into front page news. Perhaps also it was over-reliance on pictures, which are another puzzling aspect of data to information. Yet, the reason I bring up these issues is our focus on the term knowledge, and even knowledge worker. Information, as is not widely reported was the training ground for ML and LLM’s to be able to learn the English language (amongst others). As many educators would prefer, it may not have been trained on Shakespeare or even the Oxford English Dictionary, or other relevant references that had been used prior to our newfound ability to publish at will with such low cost to entry.
So, can it be considered knowledge what is produced by these imperfect methods? In a way, it has to, as we have been educated in this manner. Bugs Bunny cartoons likely had more of an impact on my sensibilities than Hamlet. Well that’s for certain. A certain popularity of content has been prevalent for over 50 years, and the gatekeepers of taste had to always bow to popular demand. While the words contained in Wikipedia are not necessarily vetted by scholars, there are enough to say that certain topics do have an investment in being as accurate or complete as one can manage. A good adage for knowledge in the current age is that the smartest person in the room is the room. Which is disconcerting to some extent as tastemakers are so prevalent, they have their own sub-culture as a new word ''influencers''. Certainly the writers in the New Yorker may still hold sway over some kind of authority not found in other media, it’s still a matter of who are you influencing? Readers, as pains me to say, are becoming more inundated by content, even a movie, which used to be a concise story, is popularly replaced by long series of stories. Serialized. While this may allow for a more novelistic approach to storytelling, it also means that we become further immersed in information. There’s little consensus as to what is valuable, what is precious or worth seeking out.
[[Reading, what's that?]]
Knowledge is best exemplified by social proof. The footnotes in a research paper hinted that your conclusions were not just manufactured, but other relevantly smart people agree with you. Clinical trials and data based conclusions are trickier, but for the most part they have always been tricky. Perfect datasets and predictable conclusions based on statistical models have for important topics been crude. Winning elections, efficacy in treatment of diseases, learned and authoritative endorsements can all be treated with skepticism in a world that seems to think that social proof is just an opinion.
So as AI is our topic, and finding a reasonably astute opinion on which to base your hunch has become a puzzle we want to explore. How can you feel confident that your knowledge is valid? What sources, especially if there are none (Amusingly ChatGPT now allows you to reference or source the chat results as.. well a reference that it actually said these things). Is a source of a large language model authoritative? I have been trying to get it to write opinions for me on many topics for this book, but have found it’s just easier to rely on my own sense of right and wrong. As any authority feels immediately better than piecing together disparate sources and ideas. Or worse, finding no real consensus on what constitutes good or bad. For example, a recent critique of, well, critics showed a stark disconnect with critic reviews and customer reviews of particularly popular movies. When investigated a bit, it was context that was at fault. These movies pleased their core audiences. They didn’t seem to have merit from a general audience. If this is not telling, it should be, how much you believe a source is more about you than the source itself.
And this personalization of topics, conclusions, actions has become fully immersive. No product goes unreviewed by collectors, hobbyists and newcomers. Most low scores are based on speed of delivery or packaging rather than the item itself. As things become more commodified we get more homogeneity and less surprise. Compared to physical products in stores, most things are packaged for our convenience rather than allowing us to have to expend time in learning arcane systems of quality and measurement.
[[Knowledge in the age of AI]]
<h3>Are you sure about that?</h3>
The DIKW (Data, Information, Knowledge, Wisdom) pyramid demonstrates the conceptual process of transforming data into more meaningful forms.
-This hierarchical model is widely used in information science and knowledge management research Rowley, 2007; Frické, 2009.
Structured data alone has limited usefulness without context or interpretation.</li>
- Research supports that data must be processed, organized, and contextualized to become truly meaningful information Zins, 2007]; <a href="http://www.systems-thinking.org/dikw/dikw.htm">Bellinger et al., 2004</a>).
Adding metadata, hierarchies, and context transforms data into information.
-The role of metadata, structure, and context in converting data to information is well-established Karvalics, 2012, Buckland, 1991
Social media platforms contain vast amounts of information but not necessarily knowledge.
- Studies confirm that social media is primarily an information source rather than a knowledge repository Paulin & Haytko, 2020, Martzog et al., 2016
Large language models were trained on readily available online information, which may not be the highest quality sources.
- There are concerns about the training data quality for LLMs, which can include low-quality web content Bender et al., 2021, Branwen, 2020
Popular culture and media have a significant impact on knowledge and sensibilities.
- Research supports the influence of popular media on knowledge, attitudes, and social norms Gerbner et al., 2002 <a href="https://oxfordre.com/communication/view/10.1093/acrefore/9780190228613.001.0001/acrefore-9780190228613-e-472">Shrum, 2017</a>
Knowledge is best exemplified by social proof, authoritative sources, and consensus.
- The importance of social proof, authority, and consensus in establishing knowledge is well-documented <a href="https://www.influenceatwork.com/">Cialdini, 2009</a> Hartelius, 2010
Statistical models and datasets can be imperfect and subject to biases, especially in complex domains.
- Limitations and biases in statistical models and datasets, particularly in fields like politics and healthcare, are widely recognized Ayres, 2008
The authority and validity of information sources are highly subjective and personal.
- Research confirms the subjective nature of judging the credibility and authority of information sources Rieh & Danielson, 2007, Chinn & Rinehart, 2016
In the age of AI and abundant online content, there is a lack of consensus on what knowledge is valuable or worth seeking out.</li>
- The proliferation of information and the challenges of determining its value and relevance are well-documented challenges in the digital age Bawden & Robinson, 2009 <a href="https://youtu.be/Q266_WtFQ2U?si=3wdWslN8ezP1o4Pd">Weinberger, 2011</a>
(print: "<h3>" + (passage:)'s name + "</h3>")
In 2007, a seemingly vast 15 years ago, before the housing downturn that seemed to put some of this exuberance in perspective, the concept of ‘viral’ was everywhere. “Charlie bit my finger” was viewed by hundreds of millions on a shared video hosting platform called YouTube. These sorts of memes, also a term coined by our visionary author Neil Stephenson, to make an entirely different point. Yet, in this instance popularity wasn’t a sign of knowledge or vetting value of information, it was a network effect. More prominence equaled more views which as hearing an earworm tune on the radio multiple times, repetition creates validation. The more you experience something, the more of a trend or validity that topic reveals. As we are taking a human and user approach to this topic, as things become more personalized, validity and repetition fade. You are the main consumer, you can see some validity in media is already suspect. A bestselling book may be notable and highly touted, but popularity can be engineered.
Marketing and advertising manufactures this sort of familiarity and repetition as their profession. It tries to plant confidence in people’s minds. Will AI change this? Certainly, as LLM’s speak with a voice that can be tailored to specific characteristics, and as computers are unfailingly helpful and not motivated by selfish gain (yet), they provide a willing and semi-trustworthy source. When the goal is a RFP proposal, or a pitch deck, it seems obvious that a lot of work, real skilled work will be augmented or replaced by these resources.
Still, we have said that has been a consistent change as computers are able to model vast varieties of experiences. I recently enjoyed the experience of learning CAD/CAM technologies and software to operate laser cutters and computer controlled machinery to create creative work of my own design. The combination of what I wanted to achieve, which was often repurposing found objects or creating simple armatures turned into a vast new array of abilities. All of which suffered from my lack of skill in the physical world. In a computer world, all is perfect and seems easily malleable and manipulatable. Perhaps this bodes well for augmented reality, where computer models can be experienced in real space. Productivity could benefit from merely not having to switch ‘modes’. My own experience relied on the vague, idiosyncratic control systems of various machines. These all accepted and understood the instructions I could encode into a software modeler, but the outcome was dependent on being able to optimize the machine, and manage the power of friggin’ laser beams. This took skill and practice. The dream of Meta and others may be to avoid the issues of real life. Judging by early experiments, the issue of having legs, a real world necessity was skipped for early versions because of relative complexity, also assuming relative lack of necessity. You don’t need them, or any other appendage to operate in virtual space. Yet, adoption and human goals are often more challenging if the space doesn’t allow it.
Picture of no feet at meta
This antromorphism or skeuomorphism, both popular design terms in early digital design helped users adjust to new organizational or experiential inputs such as navigation in virtual space, or ways to show hierarchy. As with most of the work of UX, its the brittleness of computers and their exactitude, combined with people and their ability to easily see patterns and correct errors that stump many computer systems. So if it’s a battle between computer models and abilities, we can look to gaming. It’s been able to perfect the concepts of rendering virtual space in a understandable - readable fashion for many years. As it ventures into the human realm, it’s us that are stumped. Our abilities to understand and exist in virtual spaces is limited. Thus, betting on AR as a decent cognitive load over VR which is a heavy one. It’s no accident that many of the more understandable and pleasant experiences in VR is watching a movie on a screen. In fact the way we’ve grown so accustomed to screens is a learned behavior rewarded in AR by just showing lots of screens
[[Why isn't work more like a game?]]
Why bring up AR/VR in the world of AI? Just acronym overload alone may bring this topic to a standstill, but it’s meant to reflect the complexities of computer simulation and programming vs human acceptance or reliance on those technologies. We will quicker become reliant on AI, since the need to give commands and context to problems is valuable. Not only valuable, it’s why we’ll try and tackle the largest and most critical use of how these services can be trained and utilized to help people with their largest problem in resource utilization. But before we outline it, many readers may not have a need for this willing servant. As a designer, and one who has worked in creative services for years, the ability to come up with effective ideas is always a challenge. Yet, methods and techiques to tackle lack of inspiration have taken many forms. A fun example is Brian Eno’s Oblique Strategies card deck. This series of non-equators are there to put a spin on the focus needed to create work. You’ve been repeating old tropes, polishing turds and generally feeling like your work isn’t really connecting with the audience. Then its a quick shuffle of the deck and voila:
picture of oblique strategy
Nifty! Yet, what if you could instead turn to a machine and ask for 4 variations on your problem. In music making, this could be transposition of melodies, new effects or sound processes, then you audition, repeat, iterate. Since this is more natural than changing your way of thinking or approaching a topic, and easier, it is likely to be adopted by more. Just as the phone ushered in a new way of interaction with a computer, using touch, AI in many areas will just be seen as a new avenue of expression.
In this scenario we are still making work for others to appreciate. In much actual work, we create computerized methods to communicate to other computers. While that seems to be arcane, as we’ve discussed, the cloud was just as arcane. The concept that data and processes are abstracted somehow to make them available without special machines or expensive programs to facilitate creating meaningful work is much more ingrained into the landscape of acceptable methods. Using Canva instead of firing up Adobe Creative Suite is antithetical to many long term creators, but one foray into their image and layout creation AI will make that urge dissipate for many projects.
Writing, as we’ve seen will get more personalized and targeted, mainly due to being able to use variables of browsing history, demographics and other descriptors to come up with variations on stories that will be more compelling to snack upon in our daily feeds. These feeds are pretty well named, as they are troughs of food we use to nourish outrage, form quick and lasting opinions on topics we have no experience in, and judge others. As the true-crime podcast phenomenon has shown, our appetite for sensation seems endless, even if the actual amount of these events is rare, or worse, boring and trivial in the way only crime stories seem immune to. The mundane, tragic crime of many of our cities is just not narrative worthy. Similarly, our appetite for politics will likely be a watershed of how much narrative we can tolerate. We will be innundated by information, and knowledge will be once again relegated to gatekeepers like the New York Times.
But there will be stories and narratives we once thought to be consumed by people that will change to being consumed by computers. As our resume’s can attest, hiring practices in many areas will be pre-processed by computers. Algorithms are already available to match your resume to a job description in order to game this gatekeeper. This may only apply to certain jobs, as one advances its more about connections and relationships, as LinkedIn was founded on this principal. You connect to people you vouch for and trust. That seemed to last about a year, when then accumulation of connections became easier than actual vetting and relationships. In the creative and digital space, we were given well paid positions to produce the world of apps and software. With COVID, these tuned remote, and now the community built within these industries has found cracks, loyalty and longevity in positions is rarer, and wanderlust or just simple layoffs plague the working relationship. In managing remote teams, there was ever increasing need to work on how we track and measure work. Adopting the ideas of agile and scrum which allowed teams to work more independently was encouraged, but rarely practiced. Much like any of these ideas, most will go along, and can become used to and prosper in them, but there are always outliers, and as in any situation. They often get the most attention. Not deservedly, since you can’t get 100% of people to do much of anything.
So the digital nomads are likely to grab ahold of any AI leg up they can. We may shift from the Ponzi get rich quick scheme of Bitcoin, to the get knowledgeable quick schemes of Chat GPT and others. Since this is somewhat unprecedented, and as discussed, puzzling since knowledge is a vetting scheme. If celebrities and spokespeople have passed themselves off as authorities for this long, we’re beholden to academic research and peer reviewed journals to provide some sense of authority. After living in a feed for awhile, it begins to feel like you are that authority, passing through random bits of content, training an algorithm on what sort of information that you find worthy. Google isn’t necessarily giving you different search results than someone else based on what they ‘know’ about you, but advertisers are trying. In this version of the graph, connections are made crudely, only due to budgets and strategies on how to entice poential customers. Much will be made of AI in this realm of targeting enhancement. This will conversely be challenged by privacy concerns. The more programmatically identifiable you are, the easier you make it on our advertising.
Personally, I have a lot of affinity to making the advertising as targeted as possible. Our ‘deal’ in making everything ‘free’ is that we suffer through endless pointless product demonstrations and testimonials. Much of it, for online viewers attempts to connect to topics I have shown interest in, but in other situations, cable TV, it can be loosely contextual, like advertisements in a magazine. But to make any sense of ad buying was the job of teams loosely connected to creative. A short history of advertising, the early MadMen style ad groups would sell campaigns, along with the medium (TV hopefully!, print, radio, etc). Then a percentage of the ad spend was earmarked for creative. This ‘package’ deal worked for decades, but as with deregulation in many areas, it seemed tempting to break apart this relationship, forever ensuring that creatives never really knew or understood how their messages were relayed or what context they may play in. One exception, which for the US is still true is the SuperBowl. This context and messages are in some ways the best the industry can offer. And the most expensive. Yet, it seems some outliers always arrive even when the entire event seems designed to sell product. In a word, AI targeting is already in full swing, but has yet to connect meaningfully with an audience. As evidenced by ad banner choked webpages, paywalls and subscriptions seem more humane than ever if you value your time and attention. For large audience based information and entertainment, this tug of war is likely to remain unchanged. But the battle is for eyeballs in the feeds, and those companies will continue to refine and target. Not to forget, not all targeting is for sneakers or family sedans. It can also be for ideas or to influence opinion. If these can be personalized to some extent beyond name and gender relationships there are possibilities.
One current area that has the funds and ability to benefit from AI is health. In this sense, we are going to run afoul of HEPAA laws, but in many cases, we’ve already bypassed some of those conventions by tracking our heartbeats, blood oxygen and EKG. These measurements along with sleep, can be used to personalize health management techniques. And, of course, they are already used to change personal behavior. But as an industry, products to enhance or combat health issues are thrown out indiscriminately in the current advertising ecosystem. If they were targeted, based on publicly shared health information, what kind of outcomes could we foresee? Certainly the targeting and AI could present things in a contextual manner per situation. The benefits and side effects could be contextualized a bit better, as the devices could also serve to measure if treatments were actually effective. Our experiences with COVID testing may have made people more accepting of the role of data in their care. If you were diabetic, you would likely feel more confident and enabled the more data you had available with minor discomfort being a reasonable price to pay.
So monitoring and publishing personal data for health services to mine and determine profitable, and hopefully effective treatments sound like an area large data models and ML can target and improve their promised outcomes. The issue is it’s a terrifying mental model. Any scrutiny or use of data is seen as intrusive and borderline criminal. Yet, its one of the more humane and provable use of mass amounts of data. CRISPR has been using modeling to determine helpful structures of DNA for many years, but to benefit people, you need large studies. For the most part these are difficult to manage and require control dosages, those that do not contain active drugs, and other statistical controls to determine reasonable efficacy. If health monitor users report better results with more data such as stepping or exercise being quantified motivating them to healthier habits, then the effects of treatments could only be more effective with more data. This will be an area of innovation outside of the scope of just ML, even though it may be that ML can normalize data collection with more subtlety than just the tools of data distribution we’ve used to determine effectiveness for many decades. (print: "<h3>" + (passage:)'s name + "</h3>")
For now, to summarize many thoughts we’ve introduced, it may be clearer at this point in the journey that this technology certainly is significant. For insiders, those that already have careers in the industry it will be as dominant and significant as cloud computing, and SaaS (virtual environments that run on browsers) that enable creation and sharing of information and ideas virtually to fuel business outcomes. That term //virtual// is often misunderstood, it actually does exist, but follows some basic rules to allow for an abstraction from the presentation you see. You don't have virtual shirts and shoes, but you have virtual money. It's definitely represented //someplace// and usually has no physical manifestation unless you have a large wallet or mattress. You can also have negative money, such as debts. But you don't have negative food or friends. As we talk about artifical things, virtual things, they still behave based on rules. It's just those rules may be counterintuitive to those that we may be using to govern other parts of our lives.
[[Is it all in our heads? What's a mental model, and how do I build one?]]
In this conundrum, there’s the words artificial intelligence (AI), which we have tried to demystify by calling it ML - or machine learning. The profound, yet seemingly less threatening idea that a computer can teach itself how to tackle and “solve” problems that a person used to need to be trained, and nourished within a content space to accomplish. Work for lack of a better term. So coneptually is your work solvable? Many examples have been given from education, to management, design and programming. Yes, it seems, much of your work is a puzzle to be solved. Truthfully, it’s companies that have already invested in these solutions, and how to codify them are benefiting more than others. GenAI techniques were built upon playing and winning games, like Chess and GO. Most, if not all businesses are not as reliant on software solutions to be useful and profitable. They require people, trained in the conceptual space and devoted to fufilling the needs of corporate longevity. But not a game. Why is thinking of a game helpful in understanding work? For example, we take a game like Scrabble. Playing a game is a good concept for any computer system, because there are rules on how to succeed, and engagement is likely given extreme context, it’s not an open ended goal. In this game, words are played and scored, and two people compete to see who is more skillful in using the space available and the letters provided. Over a game, some of the variability is evened out so it can be considered that skillful players may win more often than not. Zynga, took this concept and created Words with Friends in 2008, and used the new platform of apps and phones to connect players and create a way to engage a large number of people in this pastime of playing games remotely with others. This recreation of a board game comes with downsides, but a chat window kept some social elements, but otherwise the draw was a fun pastime with friends.
Yet, this model did not continue to drive profits. Once paid for, the games could go on forever, and the company did not benefit after the initial rush of users. The phenomenon of other, more addictive games like Candy Crush which use paid power ups to level up in difficult levels. This may have sent their developers and designers, along with shareholders into a push to figure out how to monetize their platform in a new way. This is where AI comes in, or ML to be clearer. They needed to teach the computer how to play scrabble. While documentation on this process is scarce, as a long time player, they introduced the idea of AI players. This allowed for gameplay outside of waiting your turn, one of the big benefits of this platform, but also not a compelling use of time from a platform standpoint. So virtual players help in several ways, they teach people how to play the game, since the sequence of challengers they offered up leveled up in difficulty. Thus their model also was tunable - conceptually able to offer a novice experience in word choice and score up to a master level. This allowed human players to practice their technique outside of the complexities of fellow human challengers.
[[The puzzle of hackathons, or non-work work]]
But what was the fun outside of the challenge? Power-up’s or cheats as many may call it. Each sequence of play has some ideal high scoring word. The game now knows how to display and rank those choices. The game knows playable areas, so this doesn’t have to be a full on prompt, it can be a series of possibilities. These ‘helpers’ cost virtual currency which can be earned by more playing or full on purchased. There are also badges and leaderboards, part of the world of //gamification// which has been toyed with in industry since the term was coined during the late 2010's. Generally it means human motivators like rewards or status controlled by a semi-sophisticated algorithm. The use of this term algorithm vs AI or ML is going to continue to be a confusing one, but for our purposes, much of this is an algorithm. A formula that can count possibilities, display them as goals and reward or punish based on performance of these goals. This concept will become critical shortly. Just remember that algorithms can be tweaked or gamed by the company. All goals can be made exponentially easier or harder by turning the knob. ML based systems don’t have the knob, they have the outcome and find their way like mice in a maze. They continually branch out ideas and the complexity of those branches can be so immense that you cannot determine how it arrived at the conclusion. Words then introduced another innovation, a gauge of your performance based on the best possible performance. This gauge became the strongest motivator, it showed that your performance with that word was proportionately good or bad based on a real word and position. Now you had some of this measurement that AI uses to train it’s model. And like real ML, you tried to maximize your score based on the effort needed to match this gauge. In practice if you are within 5-10% of the top of the gauge you may consider the extra effort not worth it. If you’re within half of the goal you see clearly there’s a better option. This gauge, whether in a human or robot player context gives you insight that power ups and cheats lack. They leave your approach up to you, while aligning you to an ideal outcome.
[[The confusing history of gamification]]
While this is just a game, let’s recap the sequence, for one, ability to play the game has to be codified, it’s not human, its a machine that can play this particular challenge. It is a challenge that people can use the platform to play. They can choose to pay and use various cheats that reveal answers, or they can for free, use the gauge and sweat through whether they can find that ideal play. In human psychology, we work internally on similar metrics. Often we substitute the gauge with recognition or praise. We use some external benchmarks as we mentioned, celebrity, wealth, abilities on the playing field, or sheer good or bad luck to gauge our particular place in the universe. But there are no real gauges yet. It’s our goal to demonstrate how ML and LLM’s can actually provide a gauge for us to improve our behavior, track goals, and see how work or business will benefit from adopting this technology.
[[Defense against the dark misuse of GenAI]]
<footer><h3>Are you sure?</h3>
1. "Yet, this model did not continue to drive profits. Once paid, the games could go on forever, and the company did not benefit after the initial rush of users."
While true for some free-to-play games, many successful games have adopted other monetization models beyond just upfront payment, such as in-app purchases, subscriptions, or advertising. Games like Fortnite and Roblox have found massive success with this approach.<sup>[<a href="https://www.investopedia.com/articles/personal-finance/020816/how-roblox-makes-money.asp" target="_blank">Roblox Business Model</a>]</sup>
2. "Power-up's or cheats as many may call it. Each sequence of play has some ideal high scoring word. The game now knows how to display and rank those choices."
Providing optimal word choices could be seen as diminishing the skill involved in the game, rather than enhancing it through power-ups. Some argue this makes the game less fulfilling in the long run.<sup>[<a href="https://www.gamesindustry.biz/the-zynamic-opposite-balancing-words-with-friends" target="_blank">Criticism of Zynga's Game Design</a>]</sup>
3. "Words then introduced another innovation, a gauge of your performance based on the best possible performance. This gauge became the strongest motivator..."
While performance gauges can motivate players, some research suggests they can also increase anxiety and demotivate players who feel they can't match the highest bar.<sup>[<a href="https://news.mit.edu/2018/study-gamification-risks-0406" target="_blank">Gamification Risks</a>]</sup>
4. "In human psychology, we work internally on similar metrics. Often we substitute the gauge with recognition or praise."
This touches on an important point - human motivation is complex, with intrinsic and extrinsic factors. Solely relying on external "gauges" like scores risks overlooking deeper psychological needs.<sup>[<a href="https://selfdeterminationtheory.org/theory/" target="_blank">Self-Determination Theory</a>]</sup>
5. "It's our goal to demonstrate how ML and LLM's can actually provide a gauge for us to improve our behavior, track goals, and see how work or business will benefit from adopting this technology."
While AI systems may provide new quantitative metrics, measuring deeper human qualities like creativity, emotional intelligence, etc. remains a challenge. Overreliance on narrow AI "gauges" could neglect essential skills.<sup>[<a href="https://www.scientificamerican.com/article/will-artificial-intelligence-become-conscious/" target="_blank">AI Creativity Debate</a>]</sup>
The core premise of using AI systems to provide performance feedback has merit, but should be balanced against the complexities of human motivation and the limitations of current AI capabilities when it comes to evaluating higher-order human traits. A multimodal approach seems wisest.
</footer>
(print: "<h3>" + (passage:)'s name + "</h3>")
(print: "<h3>" + (passage:)'s name + "</h3>")
It's 3 in the morning, you're wakened from a light sleep and begin to worry about work. That thing that's due, the odd reaction from that team member. The status update meeting with your boss. Or the client presentation due.. sometime. These anxieties are natural, to some extent. We have areas where we feel confident, often well practiced or well learned situations. We have social and situational cues. Plus thousands of hours of stories of how people act in particular situations. While anxiety in fiction Using AI to defeat the organizations of work and drudgery.
"How can we leverage Large Language Models to dig ourselves out of technical, cultural, and organizational debt? What are the essential divisons of human labor that will be required to manage these beasts?"(print: "<h3>" + (passage:)'s name + "</h3>")
Doctors, lawyers, professors, economists. My father tried his hand at playing trumpet. From all accounts, he was pretty good. As jazz was a social and creative outlet, in Mid-50's Nebraska, he likely had some good times. Yet, his other passion was to figure out why he was so deeply disturbed by his upbringing by a father that clearly disliked him. He wanted to learn psychology. Through the army, he attended college, then later worked toward a PhD after moving far enough away to Austin, TX. With a new kid, me, he was able to get a teaching position in Dallas. His interest was always the avant-guarde, so no Freudian, he embraced Rational-Emotive Therapy, and was photographed with the infamous BF Skinner, presumably a part of some convention.
This specialization in educating young psychologists, or picking up an undergrad class or two was an area of failure. He didn't have the organization, or the interest in teaching. In private therapy, however, he found his calling in the go-go swinging 70's to curse a lot and show off all these new techniques and thoughts to the eager new wealth of this sun belt. Losing the teaching position he pivoted to consultant and private practioner, just as Reagan changed all the rules to benefit this sort of entrepreneur. And he lived a long and semi-healthy life and career within a very narrow niche within a supportive climate.
Choosing today to devote the time, effort, engergy toward learning a complex and often contradictory topic is to value the concept of trust. If you are trusted, and your trade is needed, you are able to be a scarce commodity in a complex world. Your focus can pay off in managing grants and research, in private practice. Trust is hard won, I was just at a local UPS office and on the wall was their silver awards from 2010-16, then gold awards from 2017-2020 and platinum awards from 2021-2022. They were a perfectly nice UPS store, but I have no idea what these awards were for, why they seemed to inflate over time, or why award winning is notable for this kind of business. Yet every year we tune into some award show that lets us know that something was well deserved.
Award-winning, trusted newscaster, Nobel lauriate. All these come with intense specialization, often being one of a kind. Greta Thurnberg wasn't the only one to note the impact of how we have changed our environment, a once unsuccessful presidential candidate had an entire movie on the topic, which perhaps led to her point of view being more trustworthy or impactful. The medium is the message, and for people, we will never stop looking to others in our field, in our neighborhood, in our society to seek out and to emulate. If you're on this path, beware. //You don't scale. Intellligent systems do. And they are much, much cheaper.//
[[Organizations as systems]] (print: "<h3>" + (passage:)'s name + "</h3>")
For those not in the computing game, its a little difficult to say what they are or what they do. They seem ubiquitous. You are staring at one right now to read this. Yet, I've seen many people equate computers to what they do, not how they do it. For example, a computer monitor, a desk and someone typing seems pretty stereotypical of someone using a computer. Yet, if they are writing such as I am doing now, it's just a place your words go. No different from a typewriter. It has some advantages, you can fix typos. You actually may not need to know how to spell or have proper grammar. The computer aids somehow in you conveying ideas. How it actually accomplishes this points me to a personal gripe - you don't need Microsoft Word to write something.
The reason many open Word, is because they have associated it with writing something down. They didn't realize they fell into a trap. This also depends on a concept of a file. A file needs a name, and needs to be put someplace. Then, if not printed, it needs to be shared with interested parties. That leads to email, which then you 'attach' this file, making a new copy of the file. For years, to write a letter, it took many expensive programs, not to mention entire workstations to accomplish a fairly simple task. Computers began with word processing. Starting with //emacs//, created by a mischevious programmer Richard Stallman. It runs in a command line environment, which makes it challenging for most people, but it was free. Also powerful enough to manage the sort of challenges and edge cases that still makes it the most beloved and often used program for scientific work where equations and other tricky textual tricks are needed.
But how does this relate to a mental model? People imagine how something works. Often it can be correct, if you operate a car, you have some inkling of combustion or power, linking to wheels, which you can control with pedals and wheels. No matter how complex the car can be, you still have a rough idea what it does and how it does it. We have a rough idea how we work, feed us, entertain us, put us to work. These concepts may or may not be correct, we learn over time. Our mental model of the universe, our societies are how we imagine things get done. Often drama ensues when things aren't as they appear, or when they break, they reveal other truths.
I am fascinated by audiophiles, people that spend lots of time and money to perfect a way to reproduce music. What seems clear is the idea of a waveform being reproduced by electricity, capacitors, amplifiers, and speakers can be accomplished with a 20 dollar collection of parts or a 200,000 dollar one. Which is //better// is the fuel that drives the marketing and economics of consumers crazy. It's a flawed yet compelling mental model that the shinier something is, the more weight, the esotericness is a sign of quality, or mastery in this realm of opinion.
Even if you have a good mental model how computers work, with the advent of LLM's you may not need to. Computers are the most brittle and unforgiving devices historically, but incredible innovation in storage, communication, encoding, tranferring has miraculously provided constant innovation and improvement for decades. Now, the computer operators, the managers, the esoteric languages and formatting may be at an end. These new devices don't threaten commercial artists as much as they do programmers. With rudimentary knowledge you can prompt a LLM to do all the coding and setup needed to create most functional level algorithms, hosting platforms and python scripts. And it's only just begun.
[[So the job of the future is writing effective prompts?]] (print: "<h3>" + (passage:)'s name + "</h3>")
Learning new technologies can be exhausting. There are new innovations weekly and promises of the revolutionary abilities they offer are usually overhyped. Also, attention and time to understand tough concepts is in short supply. ''You'' learn things differently than others. The choices you've made in life dictate what level of comfort you have with the unfamiliar. If you're like most, you learn through experience. You may get opinions on how to do things from reading, watching, hearing, or engaging with others. But time in an activity connects concepts more solidly than observing. Teachers have their own ways of teaching, too. Sometimes they follow a script, sometimes they improvise. Sometimes they know what works for their students, sometimes they don’t. For a novel topic we came up with a novel approach: a choose your own adventure //game//. You explore, following arguments or counterpoints to explore different concepts. You can even //win// because there are choices that lead you to a conclusion of how we think AI will impact your life based on your choices. You can even go page by page from start to finish. The choice is yours.
Decision making is multi-faceted. But it's very human. At every concious age we're making decisions. Those decisions may be wildly different, depending on our environment, upbringing, appearance, abiltities. It's clear that many decisions, compounded over time gives us our experience. That experience grows depending on how much fuel we throw at our decisions. Often a stasis or calm is all that is desired, many religions are areas where we can find a respite from decision making. In fact, most experiences are there to avoid us making too many decisions. For one, its taxing. Our brains are designed to avoid it, we make snap judgements about things we know little or nothing about. We overemphasize the decisions we made as being profound or insightful, when often it's random chance that put us in that place to do.. whatever.
In terms of agency, we often fear or avoid decisions. Letting the cavity fester to a root canal, in health terms we're often puzzlingly dense as to how small decisions lead to large outcomes. As much advice as is offered to make small changes, it's often necessity or laziness that makes us act. Most books of this type are there to offer some rationale for adopting small, meaningful changes. I can recommend the //5 second rule of floor food//. //If it takes 5-60 seconds to do something, do it//. BJ Fogg is a Stanford professor, who follows in the fooststeps of Behavior Modification pioneers in the goal to get people to develop good habits, by adding them to another routine we have already established. It also helps to make it small and managable. which mimics these bite sized concepts. Even if you have to choose what to do next, don't stress! The reason decisions are stressful is that outcomes may be unclear. Choices obscure. Benefits are mixed. We've basically ruined the climate based on indecision. Without too much serious thought I'll assume we'll ruin much more stuff because we cannot decide what should be decided.
[[Wait, we've figured out how to game people's behavior by providing false choices?]]
In rare situations, we seek out people to help us decide. Much like the commercial breaks on cable channels, the most prominent decision is to continue to watch other shows on this channel. Unlike going to the grocery store and remembering the product name or benefits, we are just told how great something is by not changing anything. Unless it's an emergency or something exotic, many decisions are based on determining what friends or neighbors have done. Since our environment plays a huge role in our potential choices, people in that place have plenty of influence over our ideas and choices. Occasionally we run out of low-stakes choices and turn to experts. This is where we truly find ourselves at a change brought on by the internet. It has been able to provide a potential way to make decisions about complex things based on data or evidence.
[[Transforming data to wisdom]]
In the decision matrix, we take the knowledge, or the quality of our information as rated by others, by experts, or in the good old days, by availablitly. It's still the public library that served this purpose for ages before television tried to take over. Due to the random access problems, it was difficult to find information on the subject you needed, books solved the access problem. The quality and relevance may be questionable, but not to make light of it, reading a book about LGBTQ issues is not something to prohibit. It's likely good varitey of experiences before making some decisions. Not to mention the narrative may be clearer than the one I'm presenting. But thus the issue of online research, which is, is that a real concept?
[[Who do you trust?]]
The human part of this equations isn't the knowledge. It's not too hard to find knowledgable sources online about purchases, dinner plans, neighborhoods to live in, best colleges (debatable), what remains is to decide. Many businesses have realized that even with ample evidence, people avoid decisions. Costco's model of a fairly infinite return policy helps our brains to not overthink some things. Mostly to their benefit, since actually backing down from a decision is often more than impossbile in many situations. Its considered weak, ineffective or other nasty things that we hail upon those normal enough to have been on the record as deciding something. Our decision dislike often goes into the relationships we foster, to have a group of strong deciders is to foster debate. To have one is to quell opposing viewpoints or nuance. It's been around as long as people have been around, so isn't going to change based on technlogy. What our hierarchy often misses is more about emotions than choices.
[[Who do you trust?]]
<h3> Are you sure about that?</h3>
1. "BJ Fogg is a Stanford professor, who follows in the fooststeps of Behavior Modification pioneers"
-BJ Fogg is a behavior scientist at Stanford University known for his work on designing for behavior change. His "Tiny Habits" program advocates starting small habits through motivation and ability (<a href="https://www.tinyhabits.com/">Tiny Habits by BJ Fogg</a>).
2. "We seek out people to help us decide...people in that place have plenty of influence over our ideas and choices."
- Studies confirm the influence of social networks and peer groups on individual decision-making <a href="https://psycnet.apa.org/record/2015-21554-001">Stallen & Smidts, 2015</a>).
3. "It has been able to provide a potential way to make decisions about complex things based on data or evidence."
- There is research on the potential of data-driven decision-making, but also concerns about data quality and algorithmic biases <a href="https://ojs.aaai.org/index.php/aimagazine/article/view/2744">Barocas & Selbst, 2016</a>).
4. "Reading a book about LGBTQ issues is not something to prohibit."
- This statement aligns with research on the importance of diverse representation and information access for marginalized communities <a href="https://www.ala.org/advocacy/diversity">ALA, n.d.</a>).
5. "Our decision dislike often goes into the relationships we foster, to have a group of strong deciders is to foster debate."
- Research supports the role of decision-making styles and group dynamics in fostering productive debate or conflict (<a href="https://journals.sagepub.com/doi/10.1177/0018726711424320">Olson et al., 2007</a>
(print: "<h3>" + (passage:)'s name + "</h3>")
Sure, that's the oldest trick in the book. And speaking of books, Dale Carnegie's how to win friends and influence people. It introduced most of the techniques used for ages to instill trust of others to you. Especially if that trust is unearned or unwarranted. On the positive side, magicicans fool us for our pleasure, con men (or women), con-people? not as much. Much of this has to do with the way we store and use our histories to manage daily decisions. Taking candy from a baby seems unfair, unless you're another baby. But it's true, the baby can assume that the enjoyment of this treat is deserved, and when removed it is more emotionally charged than the treat of sugar. Perception of loss is much more powerful than gain. Financial markets are a daily reminder, when not gamed by massive bailouts, they tend to reflect a vision of the future of pretty unremarkable financal institutions. As if this was significant to our day to day lives, which it is not, but constant reminders of upturns and downturns play upon emotions and allow us to be led in ways that we may actively resent if we were made more aware of it.
[[What AI could impact, pros and cons]]
The creators of Generative AI techniques, neural nets and lots and lots of math start out training the model to play a game. It starts with trying to win and lose at a better rate than 50%. In this case, it's not worried about loss, or in a sense it's emotionless in a unique way, it keeps playing tens of thousands of times until it can beat the game better than 50%. Then that model is used to train another model that tries to beat that 50%, when it happens, this creates a new model, which in turn starts the process over and over. Like the baby we all were with the candy taken away, we learn techniques to manage this perception of loss. Perhaps we buff up to beat up those candy purloiners. Perhaps we hide our candy consumption from others. In each case we learn ways to win at candy consumption. Over years these are cemented in our behaviors for better or worse. Often winning the candy game isn't as important as it once was. We replace the candy with something else, but the motivations are the same. Fear of loss. Little can explain the predominance of older polititians or billionaires. They have already won the game. But sometimes fear of loss keeps them from enjoying their candy. As much as we've come to rely on computers for many tasks, it's only now that it could help us realize the abundance we've created, for more than ever before. But like processed foods, will it help us fear loss any less?
[[Generative AI: A primer]]
The ability to manipulate is also found in advertising, copywriting, or in a sense the words here on this page. The more we experience thoughts and ideas, the more we filter them out. Ignoring things is one of the more remarkable human talents. From our hearing, vision, and all the senses we're bombarded with stimulus most of our lives. Yet few things make an impression. Boredom, while being currently assautled by this new world of content, used to not be a terminal condition. You can stare into space for a while. Many do this deliberately and laud it as meditation. Quieting the noise of attention does take thought, the type of thought that uses energy. Since our minds and bodies crave equilibrium, anything to conserve energy is rewarded. Especially in thought, as our brains are resource hogs. Going on a run or eercising to pull energy from thought are often people's most relaxing activities, paradoxically. As manipulations go, the easy and clear and reasonable are often shortcuts to deeper paradoxes as well. Traditionalism, for example, or custom are good examples of something that does not exist. Change is constant. Preperation for the future is an admiral trait. Preparing for the future by mimicing or repeating the past is almost always noted as folly. Learning means change and experimentation, but mostly failure. Repeating something familiar is only playing a game with our real goals. We should question things we think are common sense, or try new techniques to see what once may have failed could succeed. To combat fear of loss we try to be even bolder, but that requires support and options we may not be familiar with. Generative AI is here to fill that gap.
[[Who are you talking to?]] (print: "<h3>" + (passage:)'s name + "</h3>")
In decisions, we rely on emotions. Emotions are the key to all decisions, and this is has been exploited for fun and profit for thousands of years. If you've been able to work in marketing or advertising it teaches you that there is no special attribute to Marlboro cigarettes. I prefer Lucky Strikes. But, having given up ages ago, it's a good example of how a concept of trust, of virtue, of model cowboys with craggy handsomeness and a hard job to do defined the advantage of getting hooked on this type of nicotine product. And full disclosure, I worked on the Marlboro account at Leo Burnett. The single most profitable advertising campaign in history. While GPT is arguing that fact with me, the point is that changing behavior based on an image, a concept, a fallacy is more effective if it connects emotionally than rationally. Simply because rational reasons, in any situations, are simply more work for our brains to ponder. As with Marlboro, instead of what the company wanted, claims for filtered cigarette's health benefits for thoughtful smokers, they went with just promoting manliness. This, in a series of ads with different professions, ended up with the cowboy, the aspiration to many of American-ness. While still produced, it's the worldwide obsession with this concept that kept the packages selling.
Ironically, with the dismantling of these myths, vaping followed suit with celebrities or influencers filling in for Cowboys, as is true of all advertising. It's aspirational. But to trust the media with decisions is a proxy for the emotional satisfaction you get from the outcome. Live and learn. Yet, we've shown these small easy decisions, or making no decisions at all is the norm, the baseline. It defines us moreso than those large changes would ever manage, because they are more frequent. They define our day-to-day decision skills. They also may define us to others, in a rudimentary way, as people that do that, people that think that, wear that, watch that. We compartmentaize our decisions as what to like and do based on these micro-trusts. Or micro-habits. We trust the familar. As LLM's begin to offer us advice or facts, do we trust them? Their methods and teaching may be perfect, as good as their training. This could indeed put them on par with most of us. We are no better than our sources. Yet we have that emotional quality that allows us to make decisions often with little or no information. Our mastery at an early age of practicing so many tedious skills gives us years of experience in non-verbal exploration of the world and the feelings and experiences from our actions gives us cues of how and who to trust.
[[Ok, but how do I keep AI away from my job??]]
The emotional necessity of decision making will be a puzzle that technology cannot manage. Mainly because computers have no emotions. Their choices are simple. Data is modeled in a somewhat staid and fixed way, usually crude representations like yes/no dictate many models of authority. Many of us are attracted to the number of likes or views something gets. When it hits millions or tens of millions we rush to try it. It's pretty clear that most people on earth have drank a Coca-Cola. Often several times! Its odd to think that even Santa Claus was really an advertisment concept for this beverage, to convey an emotional influence of this beverage. But contextually its a beverage. It's not critical or particulary notable that when offered a decision on a drink, that you choose Coke. But yet it's existence and prominence make it more interesting to others. It coud denote that an establishment has other good food items and that you should choose that shop. This refers to the establishment of signage that was sponsored and featured this other product alongside yours. The effectiveness of this emotion of safety, of Santa, of quality isn't really a function of this drink, and we'll note that the ability to serve up reasonable answers to tricky questions is also not in itself a note of quality. In order for this technology to have real impact, it has to change the way we live and work, and follow our logic, this can be designed. This can be realized. It can be deployed to manage the information we hold as the cornerstone of our own talents and training. But it takes a change in our models of trust.
[[Choice Architecture]]
<h3>Are you sure?</h3>
1. Emotions are the key to decision-making:
This aligns with research in psychology and behavioral economics, which has shown that emotions play a significant role in human decision-making processes, often overriding purely rational considerations. The somatic marker hypothesis proposed by Antonio Damasio <a href="https://www.sciencedirect.com/topics/neuroscience/somatic-marker-hypothesis#:~:text=The%20somatic%20marker%20hypothesis%20(Bechara,changes%20in%20the%20body%20periphery.">Link</a> suggests that emotions serve as a guide for decision-making, especially in situations involving risk and uncertainty.
2. Marketing and advertising exploit emotions:
The text cites the example of the Marlboro cigarette advertising campaign, which successfully associated the product with concepts of manliness and the American cowboy archetype. This reflects the well-established practice of emotional branding and using aspirational imagery in advertising to influence consumer behavior, as discussed in works such as "Emotional Branding" by Marc Gobe <a href="https://www.amazon.com/Emotional-Branding-Paradigm-Connecting-Brands/dp/1581156723">Link</a>.
3. Trust in media and AI systems:
The text raises questions about whether we should trust AI systems as sources of information and advice, given their lack of human emotions and experiences. This connects to ongoing debates and research around the trustworthiness of AI systems, particularly in high-stakes decision-making scenarios. The IEEE Global Initiative on Ethics of Autonomous and Intelligent Systems <a href="https://standards.ieee.org/industry-connections/ec/autonomous-systems.html">Link</a> has developed guidelines and recommendations for promoting trust and ethical considerations in AI development.
4. Emotional necessity of decision-making:
The claim that the emotional aspect of decision-making will be a puzzle for AI to solve is a valid concern. While AI systems can process vast amounts of data and provide rational recommendations, replicating the nuanced emotional intelligence and intuition that humans possess remains a significant challenge. The field of affective computing <a href="https://en.wikipedia.org/wiki/Affective_computing">Link</a> aims to develop AI systems capable of recognizing, interpreting, and responding appropriately to human emotions, but significant work remains in this area.
5. Choice architecture and influencing decision-making:
The text touches on the concept of choice architecture, which refers to the design of environments and presentation of options in a way that influences decision-making. This is a well-studied area in behavioral economics and has been explored in works such as "Nudge" by Richard Thaler and Cass Sunstein <a href="https://en.wikipedia.org/wiki/Nudge_(book)">Link</a>.
Overall, the text raises thought-provoking points about the interplay between emotions, trust, and decision-making, and how these factors might interact with the development and deployment of AI systems. While some claims are well-supported by research, others reflect subjective opinions or ongoing debates in various fields.
(print: "<h3>" + (passage:)'s name + "</h3>")
Relationships. Your co-workers are friends. Friends that can work collaboratively. Nobody vies for power, or promotion. Competition is healthy, and meant to improve everyone's ability to execute on the core mission or goals of the day. Strategy is collaborative. Many people have different backgrounds and skillsets, and that leads to great feedback, not existing in a hierarchy of 'this is right, and that is wrong' but more about 'what can we do better, I've tried this, or experienced this, can we aim for something in that area?
Work is iterative, you get to try again every day or, let's say, regularly to perfect your technique. You've got the tools needed, and the extended team all can be competent in all of them, but each person gets to excel a bit in one area or another. When customers get the product, they immediately can see that its made with care and thought. The rough edges may not be fully sanded, but the overall quality and value is evident. They keep coming back not because its the cheapest or the most convenient, but it's for the value each person infuses into the whole.
When changes happen, if people move on or unknowns occur, the core value remains, each can fill in to bridge gaps, and new people and ideas can bring new light and opportunity to grow. People are paid well, and there's enough profit to be able to support families, attend baseball games, and take vacations. Everyone individually manages their schedule and compensation, and as the business profits or streamlines, everyone benefits.
If the business becomes succesful, the urge to grow larger or expand is curbed. Banks or investors are looked upon as a hindrance rather than a benefit. The increased respoinsibility and camaradery around shared goals is considered more important than immediate profit. If any of the processes or products your team has created could be francised or mass produced at a cheaper cost, or commoditized in any way, that would serve the overall team, then perhaps you could grow, but it is considered an end game rather than a sustainable decision.
AI? LLM's? Computers may be a part of the overall tracking and process management, but it's not going to change much. You have each other and something you love to do. Even if there's only one or two people in this endeavor, they can't be replaced by a machine, and the data you create goes into the product, not into the mechanics behind the product. You're a creator, and you live for your work, not always the results of it, but becasue overall you enjoy it, and enjoy the people you make it with.
<h2>YOU WIN!!! </h2>
AI will not effect you, spend your days enjoying this unique situation, and ignore all the news and clickbait about this topic, also stay clear of social media for other decisions in your life because it's a jungle out there...
[[Ok, Mr. Socialism, that kind of workplace isn't real]](print: "<h3>" + (passage:)'s name + "</h3>")
We've seen unprecedented businesses profit mightliy on data. Mostly this data comes from users, and is given feely to perpetuate communciation or awareness of personal ideas and goals. Professionalism has been more prominent over the past few years as these businesses have understood that sharing profits also is a virtuous circle that pads their bottom line. A good short history of this data collection as a business model could stat with facebook. A forum, one of the oldest forms of internet communciation, threaded conversations about a topic. In this case they introduced the //like//button, likely inspired by //digg// which took the idea of appreciation or voting to allow for better noise control over all this newly captured data. Most lurked or read, some few participated, and a select view judged. By spreading the like button outside of facebook they also introduced this concept of gathering this kind of qualitative data. Its changed since then, but the general idea is to keep engagement at many levels of discourse. Capture the data and figure out how to use it later.
However, the data model you are capturing data about is critical. As a website engineer the mechanics that serve pages also tell you who, what, when, and how each piece of content is delivered. Pretty nifty stuff, as with both these examples this could be used to inform advertisers which oddly seem to pay more for proof of attention. Its nothing new as magazines seem even more to be just advertising platforms. Newspapers ruled both areas with their profit in classified advertsements, soon to be dismantled as craigslist took over. These data platforms, focused on a user generated data model changed everything. Wikipedia, editable by anyone became more authoritative and useful than any conventional source could compete with. And all free based on the ease of anytime, anyone maintenance and upkeep. Monetization rewarded early adopters, but the space fragments and much like the match.com of old turned to bumble or grindr, specialization in user needs helps even those with identical data models seem different enough to succeed as brands if not tech innovators.
[[Transforming data to wisdom]]
And businesses like Chevron, GE, Allstate, Wells Fargo, what did they do during this data revolution? Not too much. These online data platforms grew the ability to keep users engaged and provide trends and insight into behavior. Telephone companies never had a chance, they connected two lines together, they never cared or rated on what conversations were taking place. So big business continued in their data approach, perhaps even considering a CRM tool like Salesforce. These flexible data models work wonders in removing the need for homegrown tweaks, although having worked in those environments, every business has invested too much in their tweaks. They can never roll back to a more foundational model that could benefit from all this other data we can easily gather. Metadata, much like our website stats kind of shows what, but not why. We can set up goals and triggers and paths, and see if our users like lab mice will follow them. We can change the language and triggers to see if something works better than another. These plus the concept of reporting the results made data analysis a new hot degree topic. In my degree at Northwestern I teach the foundations of this degree, and am always happy to work with anyone who can operationalize data.
[[Work and data]]
But this fails to be an engineering problem, there may be plenty of math involved, but to influence the foundations of your data is beyond most roles. You aren't starting from scratch, so the current desire of all the business owners I've consulted with is how to add AI to their arsenal of solutions. As to be expected, you can't. Bad data is the norm, if the company isn't constantly policing, mending, iterating on how it's captured, what metadata qualifies it as significant or what decisions do you want to base upon it, it's never going to be that useful. Yet, as you've reached this ending point, reflect. You choose data, you are definitely in the right spot, data is the source of insight. But what is the data, what does it mean, how is it used? These questions have guided my consulting career, but I can say for certain that most is operational. It keeps the lights on. It's not used to see trends, train, gather and offer feedback. That's where Generative AI comes into play. And there is data you can focus on. It's all the intranets, the meetings, the chats in Slack. This qualitative data as of now has been left alone. Why? It's structure and context is too complex. It doesn't fit in a schema. Its a varchar.
Your job is to experiment with this data. Start with the endless meeetings. What do they accomplish? How can they be summarized, cataloged, rated and made into tasks or activities you can track? Most planning tools like Teams are trying this out, so use them now, see where they fit or break certain conventions. You'll be on the forefront of using data analysis for new concepts, breakthroughs. What if you used data to help your colleagues share best practices or train new people in decades of procedures and concepts while using plain language? It's the beginning of a new adventure, and not about Tableau and graphs anymore!
//Congratulations, you reached an end of your journey. // You choose to shepherd data and will attempt to look toward Gen AI to manage unstructured, qualitative data. ''Well done!!''
[[Surviving AI: How can you choose the right path?]] (print: "<h3>" + (passage:)'s name + "</h3>")
Right, too good to be true, American capatalism does have some nifty perks. Japanese capatalism makes damn good radios, as Peter Seller's remarks in //Dr. Strangelove, or, how I stopped worrying and learned to love the bomb//. We have made a mess of it, and yet, AI can provide benefits to normal creators. It can inform your education, your job choices, your future viablilty in a content driven ecosystem. It can be fun. but how to choose a path? Why is there a path in the first place?
Context. Information exists within a context. This context could be renamed a //prompt//, but in our real life, things like street signs, well, signs of any kind. A path in the woods. Its all a context of being able to weed out unnecessary items and focus on something of benefit at that moment, in that use case, as we sometimes call it in experience design. For businesses, the employees handle a variety of contexts and prompts. Perhaps they are able to do some easily, through experience or training. Other times they improvise. Sometimes they fail. In education, failure is often treated as an earned outcome. You did badly on the test so you got a failing grade. An F is feedback of you lacking someting that other students seem to have. And the learning you get from that performace is to not do that again.
[[So the job of the future is writing effective prompts?]]
If you are picking up a guitar, or kicking a ball at a goal, you will fail so consistently and often, it often feels like it should have the same outcome. You have been judged as not worthy. Truly, this does happen, it's discourgaging and you don't pursue it further. Often this is from other, more skilled players noting your lack of winning. Still, we call it //playing// music and //playing// sport, since overall grades, other than a losing score against an opponent or people choosing to converse and mingle rather than silently soak in your guitar noodlings. //Work// isnt //play// "Practice makes perfect". "Perfect is the enemy of good". There are no end of //love your work// quotes, but few about loving to play. Play isn't judged, it doesn't have grades. It can have outcomes, being a winning chess player has a good handle on why these distinctions will lead us to managing AI's impact on work. Can it make it more like play?
[[Why isn't work more like a game?]] (print: "<h3>" + (passage:)'s name + "</h3>")
Lots of wonderful things have been created before you or I got here. Exposure and appreciation of the wealth of human achievement is more readily accessable at a lower cost than ever before. But attention and time are ever more precious. In academia we're debating on if using AI is similar to cheating. For example, you could ask GenAI to summarize Moby Dick, the classic novel by Herman Melville, about a man's obsession with a white whale. And it will give you a nice and neat summary that covers the main plot points and characters, and maybe even some of the themes and symbols. As a high school student, this could also be passed off as your own thoughts, and since it's not technically plagarism, could even net you a good grade on the book report. But it will also leave out the details and nuances that make the novel so rich and complex. Which for a high schooler, could be a blessing, since subtelty and ambiguity may not be as compelling as other topics. Missing the point may be a necessary evil in education, you take a lot of swings and hope that you get some hits. But literature itself may not be an ideal way to learn, it just happened to be a relatively cheap one. Learning can happen on television, or nowadays even more on YouTube. In grade school whenever the film projector or for most of you the VCR was pulled out, even if it was the film version of Moby Dick you could at least count on some visual impact to keep your attention. Still, no matter what the medium, you might not feel anything, or think anything, or question anything, about yourself, or the world, or the whale.
Or you could ask AI to do something more creative, like write a poem, or a song, or a story, or a board game, based on Moby Dick. And it will try. It might use some of the words and phrases from the novel, or some of the images and sounds, or some of the concepts and ideas, to generate something new and appropriate to that request. And it might even surprise you, or impress you, or entertain you, with its output. But it might also disappoint you, or annoy you, or offend you. Because it will not understand the meaning and significance of what phrases it is using, or what it is creating. It can’t. While it can manipulate language within the Turing box, it actually doesn’t understand in a conventional sense anything about English. Or sentences, or anything. It just recognizes and manipulates patterns we have introduced it to. Along with all that implies. And it will not respect the original sources, or the intended audiences, or the ethical standards, of what it is using, or what it is creating. And it will not care about the quality and value, or the impact and consequences, of what it is using, or what it is creating. You can skip to ahead to witness the suffering of those humans paid to teach these processors how not to be assholes.
[[What do computers do?]]
So, you see, prompts are not as simple as they seem. They are not just a way of asking AI to do something for you. They are also a way of telling AI something about you. And they are also a way of letting the models of prediction tell you something about what it has been taught. And they are also a way of creating a relationship between you and the outcome of your prompt, that can be either beneficial or harmful, depending on how you use it, and how you interpret it.
Prompts are a powerful tool, but also a dangerous one. They can help you achieve your goals, but they can also lead you astray. They can inspire you to be more creative, but they can also make you lazy and complacent. They can open your mind to new possibilities, but they can also close your eyes to reality. They can make you happy, but they can also make you sad. They can make you laugh, but they can also make you cry. They can make you smarter, but they can also make you stupider.
So, be careful what you wish for. And be careful how you wish for it. And be careful who you wish for it from. Because you might get what you ask for, but you might not like what you get.
[[So the job of the future is writing effective prompts?]]
<h3>Are you sure?</h3>
The text makes several claims about the potential impacts and implications of using AI for tasks like summarizing literature, generating creative works, and interacting through prompts. While some of these points are subjective opinions, I can provide context and information from research and authoritative sources:
1. Using AI for summarizing books:
The concern about AI summaries missing nuances and details is valid. A study by Stiennon et al. (2020) <a href="https://arxiv.org/abs/2005.14165">Link</a> found that while large language models can produce fluent summaries, they often lack coherence, miss important details, and hallucinate unsupported information.
2. AI for creative writing:
The text expresses skepticism about AI's ability to truly understand and respect the meaning and context of what it generates. Research shows that while AI can produce creative outputs, they may lack coherence, emotional depth, and cultural awareness. A study by Keskar et al. (2019) <a href="https://arxiv.org/abs/1905.10341">Link</a> found that AI-generated stories often lack consistent themes, character development, and logical plot progression.
3. Prompts and their implications:
The text raises valid points about prompts revealing information about the user, the AI's training data, and the potential for beneficial or harmful outcomes. A paper by Weidinger et al. (2022) <a href="https://arxiv.org/abs/2202.02359">Link</a> discusses how prompts can expose and amplify biases present in language models' training data.
However, some of the text's claims about AI's lack of understanding and respect for sources, audiences, and ethics are somewhat exaggerated. While current AI systems do have limitations in these areas, ongoing research aims to improve their ability to engage with context, ethics, and different perspectives. For example, the field of "constitutional AI" explores ways to align AI systems with human values and ethical principles.
It's important to note that the capabilities and implications of AI are rapidly evolving, and the research landscape is constantly changing. While the text raises valid concerns, a balanced view should also consider the potential benefits and ongoing efforts to address the challenges associated with AI systems.
(print: "<h3>" + (passage:)'s name + "</h3>")
User experience designers often fall back on this word //gamification//. Its a shorthand to find ways to motivate people to do more work, input better data, connect with your business goal through artificial scores and rewards. A bit like those thermometers people use when raising funds. But what does it accomplish? Why invent such an odd term? We aren’t employed to play scrabble, or solitaire at work. But marking a goal, or introducing competition does seem effective. During a recent trip to an art museum I saw guards on their phones. Sitting. Seemed like a humane thing, having spent years in that environment and feeling sorry for the sheer boredom of hiring someone to stand in a corner and stare the art patrons to note misbehavior. Yet, just as soon as I noticed it, it was taken away, new management. So it goes. The game of guarding art is a tedious one, only noted by the failures such as the defacing of the Ellsworth Kelly monochrome //blue// or by other attention seekers that want to point out the absurdities of art collecting. But our premise is that so much work is inherently tedious, dumb, or simply requires being an alert human willing to give up time for a paycheck. Self-checkout is a warning that many of these roles can be eliminated, but that's a serious investment, and somehow people's time is somehow cheaper than some of these devices.
[[What do you mean by work?]]
At least at most levels it isn’t meaningful or profitable. Manufacturing airplanes or other large scale concepts that involve skilled labor don’t rely on making this work pleasurable or habitual. As we’ve experienced, work or jobs are contracts to reduce risk in creating complex outcomes. These outcomes may be measured in profit or income vs expenses, EBITA or other concepts. But in general they are collections of people with varying skills and backgrounds coming together to create value for their customers. This relationship was problematic for most of human history, conflict or supply chain disruptions could mean a profitable, valuable company goes bankrupt. Tastes and needs change. Local elements, position and resources can be unpredictable. In the early 70’s though, game makers still tackled this with Civilization and other resource based scenario simulators used various mathematical models to allow for players to manipulate and learn how to manage these elements. Yet, they tend to stray from intellectual models, the concept that an ex-founder of a company can be lured back after the company nearly goes bankrupt to foster a new business model, marketing position and luckily build and sell billions of a new concept in portable music distribution and storage that upturns the entire industry. Oh, then manages to build the first personal computer, as in a computer only you can use, that works only for you. That is, as of today, the container and keeper of your identity. This unlikely outcome requires much more than math and raw materials, but as we see electric vehicles, space travel and boring tunnels under LA are such odd concepts that seem to just work if in the right hands with enough capital.
Yet, gamifying work seems desirable to all, if only to be able to measure and engage with often tedious effort. Most workers of a certain age are comfortable with the kind of goals games offer and the payoff for learning the skills required. If education was like this, it could follow the Khan Academy example, where your goal isn’t to make A’s on tests, but to learn the skills needed to tackle tougher and more realistic problems. Teaching as a matter of mastering axioms or procedures seems like the game without the fun parts. You master those to gain access to harder situations that require more creative thinking. Much of liberal arts education thrives on this lack of rules, seeing that the goal is the journey, not the outcome. Yet STEM subjects often focus on the methodical, the tried and true and the validation that comes from empirical facts. These rules seem less like games than a very detailed instruction manual that must be followed. The satisfaction is the data gathered, or the result of small, controlled variations showing a significant result. If the result is a significant change in outcome, you could say the measurement focus of STEM allows for a concept of gaming since the way effectiveness is measured, and how often could lead to benefits through iteration. In this scenario, could the computer model do the outcomes itself and try to win the game?
[[The puzzle of hackathons, or non-work work]]
In writing, communication, design, or systems design we look to find ways to combine practical outcomes, such as payment of a price for services rendered, to a set of human monitored and facilitated activities. Many of these are tracked in systems, but most fall into non-measurable and obtuse series of conversations, manual procedures and things not easily gamed. You may be more adept at sales if you understand and are seen as relatable, also attentive to customer needs and goals. You may have better outcomes than from someone experienced but indifferent. Yet the game shows that the former may be considered more skilled since they speak the user language. The field of experience design has always fell between these, and as described earlier, we look to see where computers can aid, but rarely replace users. Yet in creative work, requesting a logo or custom imagery for a particular purpose was agency work. It normally relied on creative briefs,a topic deserving of its own explanation, a short explanation of the insight and outcomes the project wants to meet. Normally this was for impact, to be noticed amongst a sea of competitors, or to distinguish your product and people about particular values or ideals. This work was challenging, and the outcomes hard to game, since they relied on user acceptance, and often hidden measures. Did they find your logo appealing then purchase your product? Direct correlation is all but impossible, but yet, creative work is the most likely area to be replaced since the results are so varied.
[[Why can't we be brief, and creative?]]
An image or logo created by a computer, a press release written by Chat GPT, a RFP or proposal for your company to provide services to another. Any kind of legal document. These all are already being augmented by the ability of language models to speak confidently and concisely. An image created will be pretty likely to represent what you ask it to represent. Without rehashing some of our exploration of prompt management, how you ask these programs to perform will result in variations in output. The more you can reference other styles, authors, situations and camera effects will give greater control over results. The difference in human effort is massive, yet while styles may take decades to hone, you can reference existing styles, or copy from great artists much easier than you can describe something new and impactful. So high end work will remain augmented but not replaced by these tools. Stories may be streamlined, but narrative innovation will not. For our gamers, this is all too familiar. The Game cannot reward what it is not unaware of, it cannot see benefit in true innovation or novelty because it isn’t aware of it. A winning game in chess isn’t necessarily an innovative or unthought of approach, which for human players could signal significant changes in their approach to the game. But winning the game and leveling up does have negatives. What if we could combine the mechanics of tracking and game behavior to the one area in most knowledge work that is the most unstructured, untracked and conceptually most valuable tool misused by most, if not all companies. And, we will absolutely start with our language models, and continue to innovate and track performance without heavy investment from individuals, and most importantly the goal to save time and be more efficient?
[[Strategies or tactics?]]
<footer><h3>Are you sure?</h3>
Here are some thoughts reconciling or providing counterpoints to statements made in the given text:
1. "Yet, gamifying work seems desirable to all, if only to be able to measure and engage with often tedious effort."
While gamification can increase engagement for certain tasks, there are concerns that it can lead to overjustification effects, where intrinsic motivation is displaced by extrinsic rewards. This could potentially undermine motivation for work that is intrinsically rewarding.<sup>[<a href="https://selfdeterminationtheory.org/SDT/documents/2003_DeciEtAl_JEP.pdf" target="_blank">Self-Determination Theory and Rewards</a>]</sup>
2. "If education was like this, it could follow the Khan Academy example, where your goal isn't to make A's on tests, but to learn the skills needed to tackle tougher and more realistic problems."
While Khan Academy's mastery-based approach is lauded, critics argue it still focuses too heavily on procedural skills over deeper conceptual understanding and application.<sup>[<a href="https://www.psychologytoday.com/us/blog/freedom-learn/201803/the-problems-mastery-learning" target="_blank">Critiques of Mastery Learning</a>]</sup>
3. "Yet STEM subjects often focus on the methodical, the tried and true and the validation that comes from empirical facts. These rules seem less like games than a very detailed instruction manual that must be followed."
This view could be seen as oversimplifying science, which relies on creativity, speculation, and testing new ideas beyond just following strict rules. The process is more iterative and exploratory.<sup>[<a href="https://www.aaas.org/resources/cbms-biology-misconceptions" target="_blank">Misconceptions About the Scientific Process</a>]</sup>
4. "In writing, communication, design, or systems design we look to find ways to combine practical outcomes, such as payment of a price for services rendered, to a set of human monitored and facilitated activities. Many of these are tracked in systems, but most fall into non-measurable and obtuse series of conversations, manual procedures and things not easily gamed."
While tracking metrics is difficult for creative fields, some argue an overreliance on narrow quantification could undermine essential skills like creativity, emotional intelligence, and judgment that are critical for success.<sup>[<a href="https://hbr.org/2021/02/whats-really-holding-women-back" target="_blank">Issues With Overreliance on Metrics</a>]</sup>
5. "An image or logo created by a computer, a press release written by Chat GPT, a RFP or proposal for your company to provide services to another. These all are already being replaced by the ability of language models to speak confidently and concisely."
While AI can augment and perhaps replace some routine tasks, for high-stakes corporate communications, human review, editing and judgment will likely remain essential to ensure quality, tone and strategic alignment.<sup>[<a href="https://hbr.org/2022/10/ai-is-beginning-to-assist-corporate-writing" target="_blank">Roles of AI in Corporate Writing</a>]</sup>
The key seems to be finding the right balance - using gamification and quantification thoughtfully to increase engagement and provide feedback, without losing sight of deeper human motivations and the essential skills that can't easily be automated or "gamed." A nuanced, multi-modal approach is likely needed.
</footer>
(print: "<h3>" + (passage:)'s name + "</h3>")
The AI needed to take a pile of resumes and categorize them into worthwhile interviewees has at its root the kind of decisions that can be made easier by an automated process. Yet, does it make it better? For our example, we take a job description. These are authored idiosyncratically at best. Based on skills deemed necessary by team leadership, then put through the salary ranges and existing level structure of the business to ensure fair compensation. The more specialized or nuanced the skills, the more out of touch the goals may be, but it’s meant to be a rough estimate. Resume padding aside, its difficult to quantify experiences with skills, time with talent. Often creative work is not managed in the same way as production, and of course in any human endeavor, personalities are vital to success and failure.
So here we are mapping a resume to a job description. The digitalization of one to another is often a matter of keywords, titles, and various weights and importance of one level of hierarchy over another. Dates must be parsed to determine years of experience, yet the correlation is inexact. In many ways the human part is easier, can you spot confidence? Can a LinkedIn profile actually compute the relevance of your experience to the company goals? Does a person need to view your profile pic and read your intro to see if you are a reasonable facsimile of the desired candidate?
With all these variables our routine chugs along, creating a sort using our rules, either explicit or with some ‘knobs’ to dial in a more reasonable fit to the noise of a promotional document. If one applied these rules to advertisements, how could you measure the overall match to the product itself to how it’s advertised?
But these sort of biases are easier to demonstrate in human behavior, so its better to rely on a filter that can let through people that may not pass the attention span of a regular person. In my own experience I opened a position and received 150 resume’s the first day. Since I didn’t really know how to operate the filter, I just went through them one by one. As the process was to assess, then judge whether it passes through I developed a bit of a system, so that I could go faster and expend less effort. To be fair, this was for a semi-skilled role, not new yet not brand new. It required a portfolio that I needed to load. I wanted a certain amount of experience in a lead role. So I could fairly quickly dismiss the people that had no real work experience. Yet, I started to note how vague some of the data I wish was easily assessed. It took a bit of practice but I felt heartened by the fact no-one was watching, if I made a mistake it was to my own detriment since I could miss a good candidate.
Being an educator I could prioritize any formal training I could spot. I started to see how portfolios were ‘gamed’ by some basic training in how or what this profession of UX consisted of, they hit those marks and produced similar narratives and examples. Yet, often they worked in conjunction with others in groups producing these case studies, so it all got a bit muddled. I can’t imagine how we could use a sequence to evaluate any artistic portfolio, but that begs to ask that 5 years experience in software development means what exactly without evidence or examples of that code. What does 5 years of business experience entail? Making a powerpoint? Since so much of work is hard to quantify, we’re at a disadvantage.
So, with LLM’s you have the ability to have the computer ‘process’ a large amount of data and summarize. So, the thought experiment, what if you took all of LinkedIn. OR to make life easier, you focused on all of LinkedIn that currently held the job title you are hiring for. If you then asked ‘who should I hire based on…”
This ‘based on’ exploration is where it gets tricky. In my example, it was more of a ‘which canned pasta sauce should I buy” assessment. What was on the shelf? Well, this one is cheaper than others. This has a nicer label. This is on sale. Which appeals to me? Honestly, I want the best taste for the cheapest price. But there is the imported one with San Marzano tomatoes. With this in mind, we tend to choose somehow, then perhaps later along with a bunch more ingredients thrown in assess if our choice was any good. Depending on the meal, we often forget. Is this similar with our assessment and sorting problem of candidates? If LinkedIn could recommend a list of people, could we then apply our tomato skills? Could you see salary requirements to determine expensive vs cheaper options?
While in this case it will always work, it doesn’t work in a systematic, repeatable measurable way. So as a business it inhibits learning, adjusting and perhaps stagnates growth. In worst cases it emphasizes prejudices and slows needed changes that could benefit all employees. UX has tried to rise out of the role of putting polish on things to addressing measurable goals and outcomes. In what scenario can ML and AI actually make life better? Training the model. Each employee in a decision making situations has to model the weird and idiosyncratic goals they have when viewing education, visiting a portfolio site, checking for years of work experience. Then for the next role, we train the knobs based on these choices. Did they work? Was this actually not as important as another metric? Do I actually need to look for another quality? In this manner, it doesn’t get us to our goal, it is a process we put into place to make more transparent the vetting process. As it gets better, one design element is to globalize it, letting other department heads benefit from the learning. In this way the mission and qualities of the company, which never correlate to the information on a resume could be made more tangible and actionable. Best of all it could be improved gradually from every decision it makes, easing the reliance on generalists doing most of the pre-interview and vetting processes. The talent to generalize is left to software rather than imposed on overworked and easily lost recruiting talent when hiring freezes.
The ability to weather abrupt changes in focus or company fortunes favors a consultative approach. Looking for exterior sources of talent that can be utilized, then, if economics dictate, removed. This has downsides. This talent pool is less connected to the missions of the company. The needed human connections may be deprioritized over production. In many ways it imposes another level of abstraction on what could be a more productive or measurable engagement with a company. Economics aside, this also leads to offshoring and other practices that may distance workers from meaningful enjoyment and contribution to their tasks.
[[Strategies or tactics?]]
<div class="container"><img id="randomImage" alt="Set the mood"></div>
<style>
.container {
width: calc(100vw);
height: 200px;
overflow: hidden;
position: absolute;
}
.container img {
position: absolute;
bottom: 0;
top: 50%;
transform: translateY(-50%);
}
</style>
<script>
async function fetchRandomImage() {
const unsplashAccessKey = '_MZTMa9ts9YpFy5xiuZ2-CAhIPsg9odo3MBn9sCOauc';
const endpoint = 'https://api.unsplash.com/photos/random';
const query = 'random';
const orientation = 'landscape';
const params = `?client_id=${unsplashAccessKey}&query=${query}&orientation=${orientation}`;
try {
const response = await fetch(`${endpoint}${params}`);
const data = await response.json();
const imageUrl = data.urls.regular;
const imageBlob = await fetch(imageUrl).then(response => response.blob());
return imageBlob;
} catch (error) {
console.error('Error fetching image:', error);
return null;
}
}
function resizeImage(imageBlob, width, height) {
return new Promise((resolve, reject) => {
const img = document.createElement('img');
const canvas = document.createElement('canvas');
const ctx = canvas.getContext('2d');
img.onload = () => {
canvas.width = width;
canvas.height = height;
ctx.drawImage(img, 0, 0, width, height);
canvas.toBlob(resolve, 'image/jpeg');
};
img.onerror = reject;
img.src = URL.createObjectURL(imageBlob);
});
}
function encodeImageBase64(imageBlob) {
return new Promise((resolve, reject) => {
const reader = new FileReader();
reader.onloadend = () => {
resolve(reader.result);
};
reader.onerror = reject;
reader.readAsDataURL(imageBlob);
});
}
async function main() {
const imageBlob = await fetchRandomImage();
if (imageBlob) {
const resizedImageBlob = await resizeImage(imageBlob, 900, 900);
const base64EncodedImage = await encodeImageBase64(resizedImageBlob);
// Set the src attribute of the img element to the base64-encoded image
document.getElementById('randomImage').src = base64EncodedImage;
}
}
main();
</script>
And the Javascript:
async function fetchRandomImage() {
const unsplashAccessKey = '_MZTMa9ts9YpFy5xiuZ2-CAhIPsg9odo3MBn9sCOauc';
const endpoint = 'https://api.unsplash.com/photos/random';
const query = 'random';
const orientation = 'landscape';
const params = `?client_id=${unsplashAccessKey}&query=${query}&orientation=${orientation}`;
try {
const response = await fetch(`${endpoint}${params}`);
const data = await response.json();
const imageUrl = data.urls.regular;
const imageBlob = await fetch(imageUrl).then(response => response.blob());
return imageBlob;
} catch (error) {
console.error('Error fetching image:', error);
return null;
}
}
function resizeImage(imageBlob, width, height) {
return new Promise((resolve, reject) => {
const img = document.createElement('img');
const canvas = document.createElement('canvas');
const ctx = canvas.getContext('2d');
img.onload = () => {
canvas.width = width;
canvas.height = height;
ctx.drawImage(img, 0, 0, width, height);
canvas.toBlob(resolve, 'image/jpeg');
};
img.onerror = reject;
img.src = URL.createObjectURL(imageBlob);
});
}
function encodeImageBase64(imageBlob) {
return new Promise((resolve, reject) => {
const reader = new FileReader();
reader.onloadend = () => {
resolve(reader.result);
};
reader.onerror = reject;
reader.readAsDataURL(imageBlob);
});
}
async function main() {
const imageBlob = await fetchRandomImage();
if (imageBlob) {
const resizedImageBlob = await resizeImage(imageBlob, 900, 900);
const base64EncodedImage = await encodeImageBase64(resizedImageBlob);
// Set the src attribute of the img element to the base64-encoded image
document.getElementById('randomImage').src = base64EncodedImage;
}
}
main();
(print: "<h3>" + (passage:)'s name + "</h3>")
Computers have been hand and hand with business since the technology advanced enough to be useful. Much of work involves communication rather than computation. Only recently communication concepts like email and websites with their ability to display and capture data through forms changed longstanding commuincation habits within businesses. Initially, business machines were a way to increase productivity, and reduce risk and errors. Whether commuting bank balances, or transferring records, or any sort of way, they proved how to create efficiencies around mass processing of information. In 1958 //The Cybernetics of Management //was written by Stafford Beer. He laid out concepts around communication systems by which you could systematize management. The innovation was to recognize the role computing had to capture and contextualize data so that the different disciplines could collaborate to make and measure the decisions the company adopted. There was a concept of a viable system - where communication was central to the health and success of a company. These communication channels were rudimentary in the 70's when this was published, but as in much science fiction, the imagination can inspire or predict much of what later becomes real. Yet, we've stagnated in our ability to all benefit from the improvements technology has provided. Much has gone to a few.
[[The hidden cost of not knowing how hard things are to achieve]]
Commuincation however, gets cheaper and cheaper to consume and produce. Americans have always felt as though the resources, including electromagnetic spectrums by which we communicate should be privatized and exploited for profit. In contrast to other countries like England, where television is a public service, and paid for by owners of the television or radio. Debatable in terms of programming quality, but not being dependent on selling soap to fund your creative effort is reasonable. This reliance on a subtext of advertisment does have invisible downsides. We cannot know what we were missing out on in creative terms, although looking back to early TV pioneers like Ernie Kovacs may hint that it would have been much weirder and wilder. Yet the profit of the studio system has depended more and more on product placement and interstitial breaks for irrelevant items, and it seems to work. Yet, many can blame this for lackluster and safe stories, told well but lacking any real artistic meaning or social importance.
Work gains it's own channel of infotainment in the form of LinkedIn posts, a form of employabilty signaling? Perhaps there are video channels that instill the product and organizational vision within a company, but those are normally called meetings. Not unfamiliar, subjects in school have meetings, called a class Time spent in focusing a group on a topic. Often repeated. Then the bell rings and you move to another topic. This context switching comes at a cost, as it does in work. Time to do work is usually not captured, or respected. Meetings can occur at any time, although most studies prove that the ability to build the concentration needed to do complex tasks takes some time to build up and then after some time you reach a producte zone. This cycle is different for every person, but interruptions are always a disincentive to reach this state. Also, missing vital pieces, or not being able to get a resource or answer while in this zone can break the ability to produce at a high level. I have a mental picture of the Bulls with Jordan who could, after a time, manage feats of coordination or anticipation that seemed superhuman. In a culture with meetings, it often is time to do small increments of work then meet, repeating until a unit of accomplishment is reached. Depending on personal organization, some manage this in different time frames.
Work takes place during working hours. This concept seems reasonable until you look at research, especially in terms of school hours for teens, where 7AM start times and 3PM end times seem unnecessarily cruel. Teen brains are often set to different timeframes and rhythms, however, one can always assume they are not early risers, and perform inadequately if given set hours where they cannot be reasonably able to be productive. Luckily, I was in an art school, where the amount of credit a class was worth was half of a normal class. So 3 hours in an art class was worth 1.5 hours. To college students this may seem a raw deal. Yet, it was purposeful in allowing time to work. Instruction may be part of the class, but you needed time to make an effort, make mistakes, redo, regroup. It also wasn't always outcome driven. You could make a semester of work only to discard it or learn that paper mache wasn't your best choice of media. Learning was learning.
To those with MBAs, or any degree in which you get a broad understanding of the different disciplines within a corporation this may come as a surprise. Work is normally compensated differently, more specialization, more skill means higher pay.
However, it has gone in cycles, starting with education being the first adopters of the worldwide web. This intern created email, which, as a hook to businesses is a huge cost center of having to manage paper, postage, and the time needed to create correspondence. Even to this day if you receive a bill in the mail, this is a big cost, that many businesses certainly wish to cut down upon her curtail completely.
[[Management cybernetics]]
(print: "<h3>" + (passage:)'s name + "</h3>")
While we may think of personal computers or word processors, as being a big draw in early Business, it actually was both a home and business phenomenon. Mainly because the software needed and the hardware needed to run these needed also to be very inexpensive, as cheap as possible to facilitate the cost needed for the efficiencies produced. When word processing an email became ubiquitous, there were connections, made to other areas, such as servers and the ability to handle the infrastructure that manages the different communication channels. Still in my field design and user experience we were unique by adopting a different platform of McIntosh, which was more suited to creative work especially ones that are revolved around any sort of digitalization of imagery or music. Word processing, dominated, as well as spreadsheets the world of computing computing for work.
[[What do you mean by work?]]
Next big innovation in computing for work was the cloud. This also encompasses SaaS, or software as a service. This was basically switching the model of a local program that used your computer to process and manage production. Instead it was moved to a browser, itself a massively complex program, but the way into not installing software and using a browser instead to access and produce work product. Taken at an enterprise level, mataining and updating software was complex and took specialization to be reliable. A SaaS for business data and processes could take that away from you and be more reliable and affordable. The concept that you did not need to own and manage and serve and protect your business data and have that not be a huge security risk and or threat to privacy of your ideas, took quite a long time to figure out. It also disrupted the IT structural workforce that was charged with maintenance and upkeep duties that often required distinct technical specialization. If the cloud could accomplish the same goals with sufficient uptime, it saves on individual companies having to staff and train for those duties. SaaS pushes this further to taking entire data capturing concepts and globalizing them. Want to sell things? Well we have a flexible yet useful data model and doesn’t selling things have concepts and duties that we can codify, make easier and leverage best practices across many operations? This change to rely on shared processes was not always greeted warmly by many businesses. In my years at Salesforce, much of the intracacies of managing businesses was in some ways just best fit for the tools of the time. People get used to it, are trained, and if a new system arrives, matching the old system is one of the major starting points, sadly, rather than take advantage of new technology. This will certainly be the major theme in AI moving forward, how does it connect to our existing way of doing business?
The English consultant Stafford Beer in the 50’s hypothesized and experimented with a data model for a business, he coined the term Management Cybernetics, which oddly didn’t star Arnold Swartezenegger. His work in founding SIGMA (Science in General Management) and United Steel fostered ideas around large management processes and organizational theories. In his construct, a businesses goal is adaptation. By organizing around feedback loops and being able to adjust to new markets, customer needs, and the ability and processes used to create efficiencies and foster ways to formalize innovation. We can see the overall triumph of data when it comes to manufacturing processes around physical objects. From toasters to cars to airplanes, we’ve mastered much technology around manufacturing. For example we have kickstarter that was the early showcase of how rapid prototyping, data capture and defects are triaged even by small teams to make and fund projects.
Management has been less able to change, since part of the cult of personality around the Jobs/Musk type leader is their charisma and ability to inspire workers and customers to engage in their visions. Yet, it was Jack Welch, GE’s CEO that set a model for most business operations. Concepts like ''Focus on shareholder value'' that incentivized short term profits over employee well being and social responsibility. ''Cost-cutting'' that laid off and removed divisions, this impacted morale and innovation.'' Rank and Yank ''performance evaluations that continually fired based on rank added to stress and hindered collaboration and creativity. Financial metrics being used to evaluate rather than skill or long-term value creation. This model of how we manage others is one of the main areas we believe AI will level the playing field and actually unlock hidden metrics behind the mystery of work. What do we actually do at work?
[[What is work for?]]
(print: "<h3>" + (passage:)'s name + "</h3>")
Jack and I have two perspectives, both from professions and vocations that take complimentary but non-standard routes through the practice of computing. Considering our 60 plus year engagement in this field, we have a unique perspective we want to share. Generally, I am interested in software, psychology and design., Jack in hardware, algorithms and programs. Both of us are black sheep in the org chart. Design is only barely considered as an essential practice in most businesses. Most of us find using computer systems at work dreadful and confusing, yet little is done to improve that even though we are using our phones more than ever. I recall one customer pointing to the website displayed on the desktop and saying "I hate computers", then held up his phone and said how much he liked the mobile site. So most equate desktop computing with complexity. Even though most of their dislike is about fighting with sub-standard interfaces and frustrating processes. In the inner workings of software and data AI was never cool. It had some theoretical adherents, but every other computing paradigm was more lucrative and popular. Is it our day to shine? You will be the judge.
If you take computing, math, interaction, cognition, economics, psychology and mix in the ability to move away from a focus on numbers to words and concepts, you can get a feel for the way we want to merge all these subjects to help give context to what we think a humane strategy will be for the new wave of ML/AI/generative AI and other words that may or may not be familiar. The overarching term AI is kind of misleading. Intelligence is a way to see patterns and influence behaviors to build better outcomes.
So right off the bat, people imagine AI becoming a person, sentient. Appearance wise, AI will be able to mimic what has come before in likely shocking or entertaining ways, but no matter how it behaves, it doesn't have ''self-interest''. It cannot, and likely isn't aware that it's some hardware and software, even if it can express or describe this arrangement. Its somewhat similar to our brains and bodies, we've wrestled philosopically about that connection for eons, yet we control or even comprehend few of the processes that keep us alive. We have seen amazing feats accomplished by computers and programers, the effects we want to focus on are societal. Mainly concering all the ways we do not yet use computers because of their relative complexity and the skills needed to get them to do these clever things.
[[Ok, so they can learn, how did that lead to being able to hold a conversation?]]
It's likely why most people are rarely concerned with computers. They are seen as tools, inanimate objects that have as much emotional attachment as a TV set. They are there to serve. We may become engaged and addicted to what they can do. Communicate ideas quickly and allow us easy ways to share our experiences. Most social networks are demonstrations of how clever architecture and a very controlled manner of gathering and storing data is a powerful contrast to most of the existing ways we used to do exactly the same thing. A more powerful and cheaper way of doing someting, at least in America, tends to win over alternatives. Known as //satisficing// it means 'its good enough.' This need not be the case, we will outline how this technology could be used as a way to level out massive systemic inequalities, and allow us to profit even more from our ideas and endeavors. Or, as a creative designer myself, become obsolete or easily replaced. Which path shall we choose?
Information Architecture, in my curriculum is about how we take data and add layers to it to call it information, or knowledge or wisdom. These are all human judgement calls, you can believe what suits you. The newest tools seem to be able to easily provide information. To define this, data could be thought of as books in a library. Pattern could be sections or subjects in that library. Information is the book that connects to the thing you need or desire. Knowledge is a bestseller, or some sort of validation that this book is worth reading even if you dont' have any outward purpose. Wisdom is a book that changes your decisions, your behavior. Our hope is this 'book' takes you on that journey to wisdom about how this innovation will profoundly effect us over the next decade and beyond. It's a cautionary tale as well, It will be abused and misused. Unlike another destabilizing invention, nuclear weapons, it probably won't end life as we know it, but it is likely to destabilize many professions and abilities that were once scarce and expensive will be cheap and easily available. Much like computers themselves.
[[I have no idea or don't care how this works, just how it effects me]]
<footer>
<h3>Are you sure?</h3>
Non-traditional perspectives on computing. This aligns with research on the value of interdisciplinary approaches in computing and technology development. A study by Chubin et al. (1986) <a href="https://www.jstor.org/stable/3631241?searchText=Lindholm&searchUri=%2Faction%2FdoBasicSearch%3FQuery%3DLindholm%26so%3Drel&ab_segments=0%2Fbasic_search_gsv2%2Fcontrol&refreqid=fastly-default%3A90a42bdeb4ba0d1663c3517e9ea6f0f4">Link</a> examined the benefits and challenges of interdisciplinary research.
AI as a paradigm shift: The authors suggest that the current developments in AI/ML may represent a paradigm shift in computing, offering new opportunities and challenges. This aligns with research on the potential impacts of AI on various industries and professions. A report by the World Economic Forum <a href="https://www.weforum.org/publications/the-future-of-jobs-report-2023/">Link</a> explores these issues in depth.
The philosophical implications of AI: The passage touches on philosophical questions around AI's potential for sentience and self-awareness, drawing parallels with the mind-body problem. This connects to ongoing debates and research in the fields of philosophy of mind and cognitive science. A paper by Searle (1980) <a href="https://www.jstor.org/stable/20117550?searchText=searle%20chinese&searchUri=%2Faction%2FdoBasicSearch%3FQuery%3Dsearle%2Bchinese%26so%3Drel&ab_segments=0%2Fbasic_search_gsv2%2Fcontrol&refreqid=fastly-default%3A1fdfed105c47d68a50adde76778abd88">Link</a> presents the influential "Chinese Room" thought experiment, challenging the notion of AI consciousness.
</footer>(print: "<h3>" + (passage:)'s name + "</h3>")
This question evokes many perspectives, let's explore a few. In general, economies are there for exchange of goods and services. Specialized goods and services command more in exchange than common ones. Economies of scale mean even complex things could be made affordable by repetition and incremental improvement. If you compare the world of physical objects to virtual ones like computer code, the ability to iterate or fail faster makes code easier and in a sense less costly than creating similarly cheap products. Physical items require more in the way of supply chain and proximity, the interconnetiveness at low cost does favor digital products. Still, you can't eat, marry or live in a digital world so the balance is more for illustration purposes. Yet, each business, especially ones that provide for the real world benefits from information. The// information superhighway// was meant to mimic how the highway system helped interconnect American states to do more commerce, rather than being bound to local needs. In a sense, at least in developed countries, this hub and spoke of certain products did so well it defined prosperity for some time.
For workers, or people that support these economic exchanges, it's been up and down. Trust and reliance on government to educate, support and maintain a standard of living eroded in the 80's. More for distrust rather than any measurable mistake. Privatization of govermnment services like that internet thing the government funded and in a sense invented, and even for essential services like tax preperation were sold to private companies. ''Privatizing benefitted a few, at a cost to all''. The mistrust widens as health care itself became a for-profit enterprise, somehow purporting more innovation at the cost of actual results. The crappy part is that a selfish (corporate) mindset was taken as the norm, so other countries that shared the sale of the country's natural resources with citizens (Norway) or every other country with free health care has supprorted citizens in a more holistic, and some say positive way. As the point of this narrative, we do have staggering innovation, but the spoils go to the winners. And the investors. Rich get richer. Now, how does this tie in with your goals to get a good job, or get promoted, or somehow buy into this American dream?
[[So you just replace people with computers?]]
Having a variety of experiences helps formulate our goals for how generative AI can transform knowledge worker productivity, happiness, collaboration and growth. This should, but may not necessarily lead to profit and success for the organization, but HR departments and leaders of teams should immediately see benefit. Best of all, as every workplace we’ve encountered, most of the actual day to day effort and artifacts, processes and proposals are always unstructured, difficult to utilize and their value isn’t often prioiritized. Deciding what is worth your time as a worker is judged haphazardly, and in my experience almost never in a way to support learning what is good and bad. The 40 hour (nominal) work week also automatically puts quantity above quality, as studies have shown that time spent does not correlate to quality. If you are in the service industry you may learn more quickly what gets better tips, which tends to differ wildly in different locations, so forget using that as a transferrable skill. Serving people liquor seems to be a great way to earn lots of quick cash, yet, comes with late nights and sketchy skill development.
[[Does teamwork make the dream work?]]
<footer>
<h3>Are you sure?</h3>
Here are some thoughts on reconciling any controversial or factual opinions in the provided text, with relevant references and footnotes:
The text raises some valid points about the modern economy, globalization, and the role of technology and information. However, there are also some generalizations and opinions stated as facts that could use additional context or sourcing.
1. "Economies of scale mean even complex things could be made affordable by repetition and incremental improvement."
This is generally true and a well-established economic principle. Economies of scale allow companies to produce goods and services at lower average costs due to increased efficiency and optimization as output increases.^[Economies of Scale (https://www.investopedia.com/terms/e/economiesofscale.asp)]
2. "If you compare the world of physical objects to virtual ones like computer code, the ability to iterate or fail faster makes code easier and in a sense less costly than creating similarly cheap products."
While software development does allow for faster iteration compared to physical products, whether it is truly "less costly" is debatable and depends on the context. Software projects can still be extremely costly and complex, with major cost overruns.^[Why Software Projects Crash (https://www.standishgroup.com/sample_research_files/Chaos_Report_2015-Final.pdf)]
3. "Physical items require more in the way of supply chain and proximity, the interconnectiveness at low cost does favor digital products."
This is generally true, as digital products and services can be distributed globally at relatively low cost compared to physical goods that require complex logistics and supply chains.
4. "The information superhighway was meant to mimic how the highway system helped interconnect American states to do more commerce, rather than being bound to local needs."
The analogy of the internet as an "information superhighway" is apt, as it has enabled greater connectivity and commerce across geographic boundaries, similar to how the interstate highway system facilitated transportation and trade within the US.^[The Internet's History (https://www.livescience.com/20727-internet-history.html)]
5. "Trust and reliance on government to educate, support and maintain a standard of living eroded in the 80's. More for distrust rather than any measurable mistake."
This is a subjective opinion without clear sourcing. While there was a shift towards deregulation and privatization of some government services in the 1980s, whether this was due to "distrust" or "measurable mistakes" is debatable and likely oversimplified.^[Government Deregulation in the United States (https://www.federalreservehistory.org/essays/government-deregulation)]
6. "Privatizing benefitted a few, at a cost to all."
This is a generalization stated as fact, without specific evidence or context provided. The impacts and merits of privatization are widely debated by economists, with potential benefits and drawbacks depending on the specific industry and implementation.^[The Pros and Cons of Privatization (https://www.thoughtco.com/the-pros-and-cons-of-privatization-3321599)]
7. "Rich get richer."
While income and wealth inequality have increased in recent decades, this statement oversimplifies complex economic dynamics and trends.^[Income Inequality in America (https://www.epi.org/multimedia/unequal-states-of-america/)]
8. "The 40 hour (nominal) work week also automatically puts quantity above quality, as studies have shown that time spent does not correlate to quality."
This is a reasonable point, as studies have indeed shown that working longer hours does not necessarily equate to higher productivity or quality of work beyond a certain threshold.^[The Efficacy of Working Long Hours (https://www.apa.org/pubs/journals/releases/apl-a0039577.pdf)]
In summary, while the text raises some valid concerns about economic trends and the role of technology, it would benefit from additional sourcing and nuance for some of the more subjective or generalized claims. Providing specific data, studies, or expert analyses could strengthen the arguments made.
</footer>(print: "<h3>" + (passage:)'s name + "</h3>")
Collaboration with others, especially on difficult or unknown problems, in concert with a supportive environment has always been where people have thrived. Still, a team that all specialize in the same thing benefits from different focuses and skillsets. Hierarchy can stife feelings of shared goals and accomplishment. Shared feedback from different perspectives makes learning from failures or successes even more fun if the team is sharing the feedback. Even in the most artifical environments like training or school, we are frustrated by failure, and tend to pick favorites. A recent conversation with a colleague had him belabor that most of his students were duds, only recently did a couple demonstrate merit. I just kept thinking //who made you the arbiter of good and bad//? In my own experience an early viewer of my artwork mentioned how I should give it up, I just wasn't any good. Indeed, I wasn't. But that comment pushed me to pursue this over any other thing I could have done just to prove him wrong. A poor motivation but effective for many. In forming teams, as a manager, you can rely a bit on personality tests, DISC offers a compass point of possible areas of behavior that could balance a project. Meyers-Brigg tries the same, mainly to point out obstinate people that pursue unreasonable goals are considered //dominant//. And people that have better things to do than work are more submissive, yet can handle repetitious, dull tasks. Some perfer order, others are crowd pleasers. Its a full time job figuring out what activities people need to be happy, productive and produce the work that benefits the company goals.
[[Work and data]]
Or does it? The only situation where I actually bonded with co-workers with different skills meaningfully was through my own effort to network. In design it helps to have input from sales. It helps to have IT not hate you for all the trouble you will cause them. Finance and accountants definitely need to be on your side, since everything costs or produces money. But that was as an individual director, once I was able to be part of a reorganization effort where you could choose your own cross-functional team. That was a revelation, not only did all the people with similar personalities come together, we also respected each other's points of view, after all it was our choice. It was a stellar partnership, easily the most enjoyable work environment with the bonus of being extremely productive! This method is called //scrum//. Unlike //agile// which many of you could hear stated in the world of IT. Agile actually means nothing, or is like a vague compliment. Its a philosophy. Scrum is a way teams can form to get work done. Recently the CEO of Bayer decided to insitute this in the way it was intended. Fire all the bosses, let teams form and work together to produce things. Let GenAI handle the paperwork. It's a wonderful experiment and part of this book is to encourage these new ideas.
However, in my experience the results were mixed. It was short lived by two elements - the imposition of another team located in a different place and time zone was dictated to intgrate with our team. We all worked in the same room, so asking a question or just goofing off together was much easier since we were all just there. The ways of meeting, working and making group decisions was hampered when you didn't physically or socially connect with others. This may sound more familiar now after many of you experienced Covid remote work. Also collaboration and sharing itself was discouraged by this organizational scheme. Meetings were regular and fit in with natural times to break and reassess. Teams were formed by their members and mangement was forbidden. Teams don't need managers. They can run themselves, and running a team was what your goal became. This didn't really suit all the managers so they somehow went off to hide and manage something. As the Bayer and most other businesses may come to realize, teams do make the work, and management is a buffer and an expense we should be able to avoid if tech can replace it. Still the realities of working in a large company realize that they've invested little in developing processes that help people do good work together.
[[Ok, Mr. Socialism, that kind of workplace isn't real]]
To reiterate the point, as a designer, I see GenAI impact immediately on my value to the organization. In most ways this benefits me, as some laborious parts of my job are enhanced by using this technique. To reiterate, GenAI takes some input and makes something new out of it. Not simply an adding machine that is predictable or reliable. These prompts beneift from some experience or skill, but articulating a need isn't normally confined to developers. Their skill is implementation or creation. Increasing the skill level of the automaton is what changes the equation. While AI is fantastic at pattern recognition (as are we) this is information. Information, in my. definition has context. And indeed, the context is the request you give it. Great. Even if you dont' consider your skills creative, you have developed sensitive filters that keep some of the noise out of your world. For the younger folk, your skills are just being devleoped, but that band you loved a couple years ago are now trite and pale compared with this new band. Or if you are a boomer, that ad for the new car is ignored, and you're fine with Costco dog food. However, the context switching we've all been assauted with since birth could be undermined. As someone who spent years in advertising, and in digital targeting, let me explain clearly. The technology to sell you things and ideas is incredibly sophisticated, what is lacking is the ability to caplitalize on it. Google figured out how to sell words and context to provide relevant advertising to your goal. Yet, with Gen AI, what if we could ask a computer to create an ad that targeted you personally? To be clear, we do, but we work in wide swaths, since art, scripts, video etc are expensive and need talent to create. Now they don't. We may not have the skills to weed out the corporate behemoths that will certainly try to use this tech to gain foothold in the populace. Ignoring media should be a recurring course much as we've been trained to ignore malware.
[[Defense against the dark misuse of GenAI]]
(print: "<h3>" + (passage:)'s name + "</h3>")
As of this writing we prepare for the election to come. GenAI will certainly be misused, and tracking down and pointing out abuse will be a popular form of outrage. Emotions will rule decisions, as they always do, so as with malware training, remember:
* Filters are rational, but decisions are based on emotions, not facts
* Few things are urgent, and if urgency is communicated, it's almost always suspect
* Eyes are easy to fool (ha, I'm a designer!) be careful of what you interpret from images
* Our media is for profit. Any fo-profit company will err on the side of more views
* ''Context'' is critical to meaning, but easily mangled due to the sources, so often algorithms that filter things we're not going to engage with
Most of us do not work in a world of media. And most jobs are not created with the wide world in mind. They serve locally, or support larger enterprises that likely have investors and shareholders. If you ever watched //30Rock//, Alec Baldwin's parody or embrace of Jack Welsh's motivations (infamous CEO of GE) was played for laughs against writers and performers of a comedy show. Business doesn't get the joke. Creators tend to try too hard to create ideal solutions, while adequate ones will have similar impact. Context is critical, good work goes unnoticed. ''Perfect is the enemy of good.'' In my own role as a design leader within large and small companies, I realize my deficit is business acumen, but unlike LIz Lemon, I wasn't mentored and befriended by my leaders. To fit in, I worked to connect my and my teams work to meaningful KPI’s, some measure that we did something of value to the bottom line. Design is iterative and hard to measure time vs impact. Becoming quicker or more efficient may be a measure, but effectiveness, meaning or reach are much more compelling. //Easy to use// is the most common compliment given to a product or service. Apple may have a proven methodology for how they accomplish this, but to be sure, it’s an enormous challenge and takes a village to pull off. Creative work like the almost 50 years of //Saturday Night Live // that 30Rock parodies. It may be the writing, the performing, the time and guest. But mostly it is the staff that manages to provide the backbone to this organization by performing at a high level week in and week out to create sets, manage camerawork and props, makeup, cue cards. All while handling the constant changes and rewrites.
[[So, we use this to make money somehow?]]
So taking all of the company artifacts, to treat them unkindly, such as memos, PowerPoints, word docs, proposals, spreadsheets and put them into a resource that indicated either a KPI, defined processes, clarified or made work easier or more effective. This effort has consumed much of my career, starting with organizing data within a university. My consulting career comes to how is data captured in a database. This can be both qualitative and qualitative, but the maintenance of even a small resource such as news items for a company intranet isn’t very scalable. Even in utopic situations, organization of data for easy and useful consumption is very challenging. Company emails and missives go unread, and often even to capture someones attention requires to sit them down one on one for a decision. At every level, attention spans are stressed. Time to work and focus, and perhaps use that template or read that other person’s analysis of a financial strategy is often put aside for more pressing personal efforts. Systems that track work are also work, which we often put on non-workers. In the coding industry one common joke was quality assurance or testing of someones code. People who came up with code would often bow out of this and put it on others who likely had no ability or understanding of the system they were testing, which is doubly inefficient. No wonder things break.
[[Have we always been so obsessed with data?]]
Some new and radical systems like scrum were introduced, likely to combat the complexity of software development, but it actually came from car companies. Letting the workers and creators of the cars have input and be able to change systems to make a better work environment and product. For some reason this took on the word “agile’ which basically was a word that made people feel better. Most processes were still very management driven and disassociated with personal perspective from the participants. Us vs them causes much anxiety, but more importantly it keeps the necessary process of grasping for an ideal, learning the limitations, adjusting and reinventing the solution over and over until equilibrium is met. With small scrum teams with minimal oversight, amazing results are common. Without much oversight, process imposed by systems, and maximum impromptu conversation and collaboration, work can often be really fun and profitable. As most of the major companies have proven, nimbleness wins out. As companies get bigger, problems arise. And of course our major problem of a pandemic put much of this to the test.
[[Does teamwork make the dream work?]]
<footer><h3>Are you sure?</h3>
Here are some thoughts reconciling or providing counterpoints to statements made in the given text:
1. "Filters are rational, but decisions are based on emotions, not facts"
While emotions certainly play a role in decision-making, research suggests that humans use both rational and emotional processes. Effective decisions often involve finding a balance between the two rather than relying solely on emotion or pure rationality. <sup>[<a href="https://www.apa.org/topics/emotional-behavioral" target="_blank">Emotion and Decision Making</a>]</sup>
2. "Few things are urgent, and if urgency is communicated, it's almost always suspect"
This seems like an overgeneralization. While some claimed urgencies may be suspect or manufactured, there are certainly many legitimate urgent situations that require prompt action, especially in domains like healthcare, emergency services, etc. <sup>[<a href="https://hbr.org/2010/12/triage-for-the-things-that-are-not-life-or-death" target="_blank">Managing True Urgencies</a>]</sup>
3. "Perfect is the enemy of good."
This famous quote often attributed to Voltaire highlights the pitfalls of perfectionism and analysis paralysis. However, there's a balance to be struck - settling for mediocrity can also be detrimental for many high-stakes or mission-critical endeavors. <sup>[<a href="https://www.mckinsey.com/business-functions/operations/our-insights/good-is-not-good-enough" target="_blank">When "Good" Isn't Good Enough</a>]</sup>
4. "Design is iterative and hard to measure time vs impact. Becoming quicker or more efficient may be a measure, but effectiveness, meaning or reach are much more compelling."
This touches on the challenge of measuring creative/design work through simplistic metrics. While efficiency has value, qualitative measures like user impact, meaning, and emotional resonance are essential for evaluating creative output. <sup>[<a href="https://www.uxmatters.com/mt/archives/2020/08/measuring-design.php" target="_blank">Measuring Design Impact</a>]</sup>
5. "Systems that track work are also work, which we often put on non-workers."
Effective processes/systems require input and buy-in from those doing the actual work being tracked. Having non-practitioners impose opaque tracking systems can breed resentment and undermine efficacy. <sup>[<a href="https://www.gallup.com/workplace/236633/why-monitoring-work-today.aspx" target="_blank">Pitfalls of Misguided Work Monitoring</a>]</sup>
The key insights seem to be: balancing emotion and rationality, properly prioritizing true urgencies, avoiding perfectionism traps while maintaining standards, measuring creative work holistically, and involving practitioners in designing supportive work systems. A nuanced understanding of these tensions is crucial.
</footer>
What vs why?
Editing and correcting a picture. Doing a talk and need a picture for a dust jacket. By all accounts the best ones I ever had were based on a professional photographer, with a somewhat longer lens than usual. Modern cellphones have this sort of technology built in sometimes referred to bokeh. This
[[Learning programs vs solving problems]]A popular axiom when designing solutions is that nobody wants a drill, they want a hole. This concept of framing often helps creative people shake the other axiom that when you have a hammer everyting looks like a nail.
(print: "<h3>" + (passage:)'s name + "</h3>")
Whenever theres a problem to solve, or a task to complete, its part of our psyche to try and estimate the amount of effort it takes. When it comes to painting the rec room, we may choose to put it off. Procrastination may be a necessary choice to preserve energies, but often it leads to stagnation. There are several techniques to ease the concern over timing and outcome. Having the necessary items and tools to complete the job, having the expertise or practice, and knowing what level of outcome is needed to feel like this was a positive use of time. In our personal lives much of this comes from being educated. There are artifical timelines and hopefully reasonable challenges that keep us continually trying to produce work that helps us become better attuned to our abilities and how to achieve challenging goals.
In a school situation where we work with others, the ability to achieve can be greatly enhanced. As in the old movies, a group can put on a play, build a barn, or create a Thanksgiving feast. Or, without a sense of who can do what, there's always a random chance that one person does most of the work, while the others coast. In a work situation, we often find ourselves in a hierarchy where we are expected to do what others don't do. In design this is irksome, although it's easy to look at artists and architects that want to tackle all the problems with their own sensibiilty. Not everyone can design a house and the furniture that goes in it. Even fans may not like sitting in Frank Lloyd Wright chairs, but the overall idea seems to be more engaging when sensibiilties of the creators are in sync. In work, we train, meet and attempt to track how teams perform. Without clear ways to understand individual skills and group dynamics, it's more often a crap shoot as to how we manage to get anything done.
[[How to turn work into a game with AI]]
It's been a theme in Japanese car making to allow anyone in the company to comment, suggest, and even innovate new ways to improve the product. In the 70's this kind of esprit de corps was all but unknown to the American auto industry. In a Henry Ford way, we prefer things as cogs in machines, rather than thinking and acting players in the creation and delivery of products. Even in hyper modern industries like tech, it's only been recently that we enable the workforce to experiment or learn how to influence the company direction. Its clear that without a technique or method, most of our work and even home life is playing a part rather than creating opportunities to grow our skills through trying new things. As I and many of you have experienced this, what changes with GenAI? Truthfully, it's likely to be a negative unless we can harness the power of making mistakes.
[[What tiers of probabilities are you looking for?]]
The scientific method is an area that AI is taking aim upon, mainly because that's the way it was programmed to function in learning complex systems, trying many ideas and measuring which comes out ahead. This isn't too far from any brute force approach, and is a dirty secret of scientific efforts that much can be expended with little to show for it. Yet overall, the few outrageous hypothesis make up for all the poorly formed ones. It's a triumph of technology that what was once expensive and rare is common and cheap. There is more computing power than ever, and if we don't factor in the issues of distribution and energy consumption, the problem becomes what do we do with all this power? As individuals, you can stand on the shoulders of giants and accomplish with less effort than ever before a huge range of tasks that were once out of reach. Education alone has been forever changed with the economics of YouTube creators or TicToc influencers giving us more access to more abilities that are there for the taking. No longer do we have to overestimate, fumble and luck into that perfect cheesburger. We can watch and learn by master chefs. This need not diminish the role of honing skills, but it can change the dynamic much like those Japanese workers trying to make the car fit together better.
[[Does teamwork make the dream work?]]
Work is a process of trying new ideas, experimenting, and determining a successful formula that is sustainable and profitable for the financiers, the management group and the individual. This involves lots of communication, especially of specialized or complex ideas. I put AI on task for helping anyone communicate at a base competency. Those more skilled can use it to test new ideas, see different perspectives. And gather and utilize feedback from others. Much like how teachers and students should collaobrate. But quantification took over our mindset, counting the GPA rather than the character or goals of the learning. Salary rather than opportunity to grow. Then we need orientation and some consistent feedback mechanism to see when we have a good fit with our temperment and our skills. Was an A student better than a B or C? What skills or abilities did these grades convey? Most will say that no grades, money is as beneficial and worthwhile as praise from skilled peers. I have worked for free more passionately and successfully than for a salary. I have also found that once money enters into the equation, then motivation dries up quickly. You see how much you are worth compared to others, and that shared goal is quickly surpassed by monetary assessment. For GenAI to build up the team spirit, it needs to aid in assessment. Help with prioritization of goals and tasks, and communicate consistently to the team on their progress and give constructive feedback. While all of these are possible, they are qualitative, they require connecting on a more abstract and personal level. And the technique we've outlined is well suited for this kind of connection.
[[Strategies or tactics?]] (print: "<h3>" + (passage:)'s name + "</h3>")
Specific to the tech industry, but not unfamilar to other departments that have contests for sales, or group brainstorming sessions, or executive retreats the concept is to make work less like work. If only for a day or two. In this way, all have a chance to show off some skills, get feedback on their ideas without the inherent pressure of assignments or team dynamics. Or we enhance or detract from those dynamics by doing trust falls. No matter, it's an excuse to change up the routine. COVID changed these dynamics in many ways covered later in this narrative. But group work differs from individual shows of prowess or accomplishment. Especially if the event is judged or provides some way for there to be recognized in the group as a talent. Most work not for money but for the respect and acknowldegement from peers and authiority figures. This has likely been drummed into us from school. The way I teach class has always been to have all assignments be individual ones. Mainly because having been in group assignments, roles are more of a challenge than the work itself. The negative is that the instructor (me) has to do much more work. Still, it's a reasonable tradeoff.
[[Does teamwork make the dream work?]]
A hackathon, where some group is formed and they discuss some problem they think they can solve. You start with the why and go to the how. It is time boxed to 24 houry of work, normally one day, but normally theres a concept, some code, some demonstration and then presentation and judgement. Having worked in advertising, we have a similar way of coming up with a pitch for business. However this can take many weeks, involve many teams, and is also competitive. The role of AOR or //Agency of Record// which indicates an ongoing paycheck for the company if not the team is the prize. In this way we come up with some way to impress the marketing leaders that we have the right stuff to sell their product over the other agencies. In a room filled with 20 super-talented creatives we were asked to design the first company-sponsored Facebook banner ad. Yes, once upon a time Facebook had not figured out advertising, or at least this style where you interrupt the feed of family and friends latest adventures. Ideas were truly wild and outlandish. My own thought, why not just ask them to participate in a topical discussion? Working for Purina, it was "Are you a dog person or a cat person" or something like that. In every way I was embarrased to show this concept, every other one was more engaging and imaginative. How puzzling it was that it actually won over all other major agencies. Also puzzling that some other person on the team took credit. Oh well, it's always fun to play.
So the same applies to hackathons, individuals freed from their job, somehow then invent some technology in a day, demonstrate it, then get public support and praise for it. I spent a few hours with a team helping them visualize climate change data, and we ended up hobbling together a 3d animated globe that told the story from ancient times how much we have polluted. It was fun to win a prize, although others had more practical and effective solutions where they were also lauded. A good time was had until next year, yet right afterwards we went back to the team being literally unable to produce much of anything. For a year. We spent endless time making minor tweaks in the software, and each ritual and conversation seemed to distract from any accomplishment. Most of the talented people quit, depriving the others from their perspective and skills. I was laid off as well, but the puzzle is, why is not being like work such an effective way to get people to work? And not to lose the thread, how does GenAI change this paradigm?
[[The hidden cost of not knowing how hard things are to achieve]]
(print: "<h3>" + (passage:)'s name + "</h3>")
As a designer, I’m really most skilled at communication. Oddly, in the world of interaction, design designers or UX people don’t actually //make// anything. Think about it, in your job, do you create something tangible? Do you touch, feel or nudge things into being? Is that a thing you could show to someone that would delight or interest them? I thought so. Why is this? We've become accustomed to thinking our words, or intent stares accomplishes something worthwhile. Perhaps it does. Without my presence, I know that some projects and meetings have gone into some hellscape. It took a lot of effort to unruffle feathers, and to be honest, no business likes hearing anything negative about anything, so there goes that job. So what should we do with our time, our careers? That may be where generative AI may play the biggest role. It has the potential to make us less risk adverse. It could make us better at communication, or replace much of the redundant meetings. It could bring us together with others to co-create in a more equal fashion. It could be exploited to make even more misery. Hard to tell, but best to be prepared.
[[I'm more about the human side]]
If you've been in the tech world, or the education world, or the writing world, or much of the business world you create artifacts. In UX design they take the form of wire frames, drawings and prototypes. They are used to approximate what we think is a good idea for our customers but really, we just use them to tell Developers what to build. Often this is literal we’re just telling them move this to the left make this one pixel higher. Worse, we create schematics and fill it with lots of language to describe how something works. Most decent developers learn to ignore all this and just glance at the picture and build it based on the libraries or code bases they are familar with. So, in my career, I’ve found this to be a waste of time. Not that it isn’t necessary, people are come from different backgrounds with different talents, and it’s getting a group of people to be talented and solve problems. That normally makes for best results in the end case anyway. But there’s a new communicator in town, and it’s really pretty good at communicating basic ideas. It’s actually pretty good at translating and idea into code. No, this may take some time, but the reason I am putting these thoughts out is to help people who may be in similar situations. They do things that are hoping along in the process they often are called T-shaped people because they’re interests and skills do not lie in one particular area but in combining several areas.
Generalists talking to specialists. If your goal is to communicate an idea, then you’ll want to learn more about what’s this technology will bring. Right now, the obvious headline may be with high school students. Having been one, there is a feeling of you reach a certain GPA or a certain status that you are waiting for the next step to go to college and the efforts needed in class is secondary to other things such as... fun. So, they’ve quickly learned why bother reading that book and writing a book report when GPT is happy to do it for you, and usually does a really good job. Same is true of most of the rudimentary work, and by rudimentary, I mean the concepts that you need to practice before you can handle more advanced concepts. I teach at Northwestern, Information architecture to be specific. This practice is really based on breaking things down into layers and studying how each layer affects each other. At the moment, one thing I have worked on for years is not too write papers, but to make videos, or to make anything that someone can interact with. So there’s this interactive story that you will tell through your choices. The reason you’re here is curiosity, also mixed with how can this be useful in my pursuit of X. Those layers are data, pattern, information, knowledge, wisdom. Let's explore how we make sense of the world.
[[But what is knowledge, is that what we aspire to?]] (print: "<h3>" + (passage:)'s name + "</h3>")
Tech-savvy recent grad: You have spent years studying and learning new skills, but you are not sure how they will help you in the real world. You want to find a job that is fulfilling and rewarding, but you also want to keep up with the latest trends and technologies. How can you use AI to enhance your career prospects and satisfaction? This book will show you how AI can help you discover new opportunities, learn from experts, and create your own projects. You will also learn how to use AI ethically and responsibly, and how to avoid common pitfalls and biases.
[[Organizations as systems]]
Mid-level worker: You have been working hard for a while, but you are feeling stuck and bored. You want to balance your work and life, but you also want to grow and advance in your field. You are worried that AI might take over your job, or make it irrelevant. How can you use AI to improve your work performance and enjoyment? This book will show you how AI can help you automate tedious tasks, enhance your creativity, and collaborate with others. You will also learn how to adapt to changing demands, and how to leverage your human skills and strengths.
[[Why AI will change the way we work]]
Management: You are responsible for leading and managing a team of diverse and talented people. You want to hire the best people, motivate them, and help them succeed. You also want to streamline your workflows and systems, and reduce unnecessary meetings and overhead. How can you use AI to optimize your team’s productivity and happiness? This book will show you how AI can help you recruit and retain talent, provide feedback and coaching, and facilitate communication and coordination. You will also learn how to use AI to generate insights and solutions, and how to test and implement them.
[[What do you mean by work?]]
Upper management: You are in charge of making strategic decisions that affect the future of your organization. You want to stay ahead of the competition, and anticipate the needs and preferences of your customers and stakeholders. You also want to foster a culture of innovation and excellence, and inspire your employees and partners. How can you use AI to inform and enhance your decision making and leadership? This book will show you how AI can help you analyze data and trends, identify opportunities and threats, and create scenarios and plans. You will also learn how to use AI to communicate your vision and values, and how to empower and engage your followers.
[[So, we use this to make money somehow?]]
Consultant: You are an expert in your field, and you offer your knowledge and advice to your clients. You want to deliver high-quality and customized solutions, and establish trust and credibility with your clients. You also want to expand your network and reach, and find new ways to add value and impact. How can you use AI to improve your consulting skills and services? This book will show you how AI can help you research and understand your clients’ problems, generate and evaluate options, and present and implement recommendations. You will also learn how to use AI to market yourself and your brand, and how to create and maintain relationships with your clients.
[[How AI Works Differently Than You Might Think]]
Educator: You are a teacher, a trainer, or a mentor, and you want to help your students or learners achieve their learning goals and outcomes. You want to design and deliver engaging and effective learning experiences, and assess and improve your teaching methods and results. You also want to keep up with the changing expectations and needs of your learners, and the evolving standards and trends of your field. How can you use AI to enhance your teaching and learning practices? This book will show you how AI can help you create and curate content, personalize and differentiate instruction, and provide feedback and support. You will also learn how to use AI to measure and improve learning outcomes, and how to foster a culture of curiosity and lifelong learning.
[[Machines? Learning? What does that even mean?]]
Administrator: You are an administrator, a coordinator, or a supervisor, and you want to ensure the smooth and efficient operation of your organization or department. You want to manage and allocate resources, comply with rules and regulations, and handle issues and conflicts. You also want to improve and innovate your processes and systems, and align them with your goals and values. How can you use AI to simplify and streamline your administrative tasks and responsibilities? This book will show you how AI can help you organize and store information, automate and optimize workflows, and monitor and evaluate performance. You will also learn how to use AI to identify and solve problems, and how to implement and sustain changes.
[[Deep Organizations with Shallow Computation]]
Lawyer: You are a lawyer, a legal advisor, or a paralegal, and you want to provide the best legal service and representation to your clients. You want to research and analyze legal issues, draft and review documents, and negotiate and litigate cases. You also want to save time and money, and increase your efficiency and effectiveness. How can you use AI to augment and improve your legal skills and practice? This book will show you how AI can help you access and process legal information, generate and verify legal arguments, and predict and influence legal outcomes. You will also learn how to use AI to manage your cases and clients, and how to balance your ethical and professional obligations.
<h3>Are you sure?</h3>
1. AI enhancing career prospects for recent graduates:
The text suggests using AI to discover new opportunities, learn from experts, and create projects. This aligns with the growing use of AI in areas like job matching, skill development, and project collaboration tools. Companies like Handshake <a href="https://joinhandshake.com/">Link</a> use AI for job recommendations, while platforms like Coursera <a href="https://www.coursera.org/">Link</a> offer AI-powered learning experiences.
2. AI automating tasks and enhancing creativity for mid-level workers:
The claim about using AI to automate tedious tasks and enhance creativity is supported by research in areas like robotic process automation (RPA) and AI-assisted content creation. A report by Forrester <a href="https://www.forrester.com/report/Leverage-AI-To-Drive-The-Democratization-Of-Analytics-Insights-And-Data-Literacy/RES176112">Link</a> discusses how AI can democratize analytics and insights.
3. AI for team management and productivity optimization:
The text suggests using AI for tasks like recruiting, providing feedback, and facilitating communication. Companies like HireVue <a href="https://www.hirevue.com/">Link</a> use AI for video-based hiring assessments, while tools like Lattice <a href="https://lattice.com/">Link</a> leverage AI for employee feedback and performance management.
4. AI for strategic decision-making and leadership:
The claim about using AI to analyze data, identify opportunities, and create scenarios aligns with the growing use of AI in areas like predictive analytics and strategic planning. A report by McKinsey <a href="https://www.mckinsey.com/capabilities/strategy-and-corporate-finance/how-we-help-clients/decision-support-ai-analytics">Link</a> discusses the use of AI for decision support and strategic analysis.
5. AI for consulting and client services:
The text suggests using AI for research, generating options, and presenting recommendations. AI is being increasingly used in areas like legal research, financial analysis, and management consulting. For example, JPMorgan Chase has deployed an AI system called COIN <a href="https://www.jpmorgan.com/solutions/cib/markets/code-for-good">Link</a> for contract intelligence.
6. AI for teaching and learning:
The claim about using AI for content creation, personalized instruction, and feedback aligns with the growing field of AI in education (AIED). Companies like Century Tech <a href="https://www.century.tech/">Link</a> offer AI-powered learning platforms, while research efforts like the AIEd initiative <a href="https://www.edsurge.com/news/2022-09-19-ai-is-coming-to-k-12-classrooms-how-ready-are-schools-and-teachers">Link</a> explore the use of AI in K-12 education.
(print: "<h3>" + (passage:)'s name + "</h3>")
Bertrand Russell (1872-1970) was a British philosopher, logician, mathematician, historian, writer, social critic, and political activist. He is widely considered one of the most influential thinkers of the 20th century. He wrote a piece for Harpers Bazaar in the 30's called //In praise of idleness//. Well worth reading from an American viewpoint, as we seem to be at odds lately with the concept of productivity, as you have witnessed in traversing this journey through AI. There is a general criticism and poking holes in the anglo saxon protestant work ethic throughout, as it doesn't seem to provide the benefits we ascribe to hard work. Yet as propoganda, it's everywhere, and in a consumerist society, suppy side, making more things, seems like a decent choice. But as to how you contribute to making things, Russell writes:
<blockquote>
First of all: what is work? Work is of two kinds: first, altering the position of matter at or near the earth's surface relatively to other such matter; second, telling other people to do so. The first kind is unpleasant and ill paid; the second is pleasant and highly paid.
</blockquote>
The humor is that in the time he wrote this, the 30's, we as a society were just beginning the process of melding machinery and labor. Now, about 100 years later, we see making things by hand as a valuable human skill. Artisanal. Craft. Handmade. All of these have good and wholesome connotations. Even if mainly used to market barely needed items. Most food, energy and appliances are mass manufactured to lower costs. Lower cost tends to win over quality to most. Sometimes its impossible to not scale to make millions to make complex machinery affordable. Like computers or washing machines. But this format of work still seems to remain intact. Yet GenAI has the possible outcome of doing the //telling other people to do things// part. It then enables the artisan to finally an equal status, since the telling part becomes cheaper, better and more consistent. The crafting and making remains in skilled human hands. After the pandemic we see this in manual labor becoming recognizable as essential. Presence and influence over students, or even (unlikely) the need to sit near your boss in an office.
You may still decide to avail yourself of this technology, but perhaps to manage your expenses, communicate and market to others, or prompt or teach computers to do some of the more laborious parts of creating.
Which is an end, you can forego any need to be concerned over GenAI if you create, work with your hands, or enjoy manual processes. Those that tell others to do things will need to explore further, but I wish you well since you will at least have the satisfaction that comes from creation in the real world.
[[What is work for?]] (print: "<h3>" + (passage:)'s name + "</h3>")
In law, the term brief was used to represent documentation on the case, facts, goals of the litigation and precedent. Before accepting or engaging in litigation, the brief outlined a course of action that the plantiffs desire, to convince the judge that there was merit to their arguments. This also refers to a piece of paper, which we put in cases to carry around to signal professionalism. In this instance, a brief is a summary and fortunately, not very lengthy so truth in advertising if you need to make informed decisions on a deadline.
In advertising, we have creative briefs. Not to be confused with marketing or design briefs, and oddly dissimilar to executive or policy briefs. And underwear doesn't come into the equation at all. But the hope is whatever the word they are short and to the point of helping you understand the goals of your efforts. Strategists write briefs by distilling or determining the customers business problem, insight into their competition and customers, and define an overall outcome that the work the team does should accomplish. The brief goes through several stages, and often is used to have competitive bids for the true goal of an agency to become the agency of record. This designation allows the agency to play an even more strategic role in the ways the company communicates to consumers.The result is an interpretation, or riff on the goals. Ideas can be //on or off //brief. Unlike the other examples, the work stems from the brief, gives it context while the other examples are similar to RFP's or requests for proposals that set budgets and procurement guidelines rather than dictate actual strategic outcomes.
In desiging software, we often refer to ''requirements''. Somehow this isn't specifically defined, it can be whatever was cooked up in the quarterly meeting, the 15 year plan, the musings of what they saw a competitor do. All it needs to do is define a problem, some insight and an outcome to let creativity take over. Yet it always lacks this, worst of all, it's ''not brief''! Requirements tend to literally resemble laundry lists of ideas and goals. This puzzle is harder to unpack, what if one goal interferes with another? What if there are easy and hard goals mixed together? I often joked that in advertising we had briefs and design we had requirements and it should be the other way around. Funny? Well its true that without constraints, creative ideas can be overblown, impractical and also ineffective. In the early days of digital advertising there was much effort to recreate print advertising, yet somehow it took quite some time to see these were two different mediums. You still see those horizontal slivers of mush at top of screens called banners because we couldnt figure out how to tout our product programatically. Even better we had a term called banner blindness where people learned just to not look up there. Much like print going to television, the constraint of cost and airtime solidified a 30 second TV spot as the preferred vehicle for telling a story. No time at all, right? Chunks of 4 or 5 of them managed for years. Until Youtube managed to enforce 15 seconds, and with TicToc its getting even shorter.
[[The hidden cost of not knowing how hard things are to achieve]]
Briefs to create complex work do have advantages, they provide an anchor to wild solutions. Since they involve summarizing large conceptual spaces, they could benefit from being a data source for generative AI. Certainly it can provide a decent starting point. With that concept in mind, could requirements and briefs be reinvented with this new technology? If so, does that free creators from spending time documenting and justifiying their concepts? Can that also be part of GenAI's magic, to determine the goals then experment on ways to satisfy those goals? Currently we see bits and pieces of this, but we're using these tools to influence behavior. I'm not convinced that you can change people's minds. I think you can reinforce what they already believe or want to do, but the concept of nudging has to do with taking something selfish and merging it with concepts more beneficial to more people. These virtuous circles where each behavior aids another role, or moves the process forward are what is often referred to as ''sustainable''. It's by definition repeatable, and involves feedback cycles to adjust and improve. Each cycle can also draw in new participants, although often we frame a virtuous cycle to our own well being. This narrative is //meta// enough, but how we communicate, what the artifacts and outcomes are can be improved through more feedback circles, more concentration on virtuous outcomes for participants. Yet like requirements we can forestall this process through imperfect legislation, promoting misuse examples over positive ones. The inventors of AI started with teaching computers how to win games but then solved the problem of protein folding, one of the most complex computational problems. Still, after it was done they had to wonder what this afforded? Now compounds were revealed which no understanding of their effect was known.
[[Knowledge in the age of AI]]
But how would we know about misuse? The tool does an unprecented job at a seemingly unsolvable (by conventional computing) problem. The reason we would know is they published all the results online. This platform which briefs us all in any topic we wish also holds much of the worlds secrets and impactful information. Like wikileaks, the insistance on general knowledge of dangerous concepts may illustrate the opposite of sustainable feedback, the trust in open information to counter bad actors is a large issue. One that only needs a few examples that will spark fear and security concerns. We'll be privy to how GenAI will be used much like Cambridge Analytics exemplified targeted messaging concerns on Facebook. Would this new AI tool offer more insidious ways to inform us of how we can mistrust each other? Or would it be similar lessons to what was learned from 9/11 (forgive the US perspective here) that we just didn't think something was possible, but now that we know we are more cautious? I hope that would be the case, but as we rely more on briefs and shy from the compexity of requirements the outcomes or goals have to be stated. Whether we get there on or off brief we make an effort toward the same conclusion, which may be the best, most sustainable option.
[[Strategies or tactics?]]
<footer><h3>Are you sure?</h3>
1. "Yet it always lacks this, worst of all, it's 'not brief'! Requirements tend to literally resemble laundry lists of ideas and goals. This puzzle is harder to unpack, what if one goal interferes with another? What if there are easy and hard goals mixed together?"
The critique of overly lengthy, unfocused requirements documents is valid. However, experts from the Software Requirements Engineering field argue that well-crafted requirements are essential for avoiding costly rework and misalignment later. The key is developing techniques for managing complexity through proper scoping, prioritization, and traceability.
<sup>[<a href="https://resources.sei.cmu.edu/library/asset-view.cfm?assetid=506616" target="_blank">Software Engineering Institute at CMU on Requirements Engineering</a>]</sup>
2. "If so, does that free creators from spending time documenting and justifiying their concepts? Can that also be part of GenAI's magic, to determine the goals then experment on ways to satisfy those goals?"
While AI can potentially augment ideation, critics caution against over-reliance that could diminish human agency and the essential "artistic struggle" required for true creative breakthrough. Prominent AI ethicist Joi Ito argues we must maintain human-centered processes.
<sup>[<a href="https://jnan.mit.edu/research/publications/the-practice-of-change%3A-fit%40nding-our-way-into-the-future" target="_blank">Joi Ito on Human-Centered AI Processes from MIT</a>]</sup>
3. "Currently we see bits and pieces of this, but we're using these tools to influence behavior. I'm not convinced that you can change people's minds. I think you can reinforce what they already believe or want to do, but the concept of nudging has to do with taking something selfish and merging it with concepts more beneficial to more people."
This perspective aligns with research on the challenges of countering established beliefs and cognitive biases through mere information provision. Sustained behavioral change often requires addressing underlying motivations and value systems.
<sup>[<a href="https://www.behaviouralinsights.co.uk/wp-content/uploads/2015/07/Behaviour-Change-Brief-0.pdf" target="_blank">UK Behavioural Insights Team on Behavior Change</a>]</sup>
4. "The inventors of AI started with teaching games but then solved the problem of protein folding, one of the most complex computational problems. Still, after it was done they had to wonder what this afforded? Now compounds were revealed which no understanding of their effect was known."
While the protein folding breakthrough was groundbreaking, bioethicists warn of potential risks if applications proceed without robust frameworks for safety and responsible use. Unintended negative impacts must be carefully assessed and mitigated.
<sup>[<a href="https://www.scientificamerican.com/article/ethics-are-a-key-piece-of-the-artificial-intelligence-puzzle/" target="_blank">Scientific American on Bioethics of AI Protein Folding</a>]</sup>
5. "Would this new AI tool offer more insidious ways to inform us of how we can mistrust each other? Or would it be similar lessons to what was learned from 9/11 (forgive the US perspective here) that we just didn't think something was possible, but now that we know we are more cautious?"
These are critical questions raised by AI ethics and security experts. While AI offers immense potential benefits, the risks of adversarial use, disinformation, and eroding social trust must be proactively addressed through responsible development and governance frameworks.
<sup>[<a href="https://ethics.harvard.edu/ai-ethics" target="_blank">Harvard's Berkman Klein Center on AI Ethics</a>]</sup>
The key insights stress the need for nuanced, human-centered approaches that balance AI's creative potential with maintaining core human roles, addressing systemic motivations beyond mere information provision, implementing robust governance for high-stakes applications, and proactively mitigating risks to social cohesion and trust. A holistic, ethical perspective is crucial.
</footer>
(print: "<h3>" + (passage:)'s name + "</h3>")
"The best way to win is not to play"
Fake computer in 1980's film //War Games.//
(text-colour:red)[You win! ]Yes, not bothering is a decent choice. From my informal research, it seems reasonable that technology isn't really beloved or respected. Computers are often seen as cumbersome and complex tools to accomplish proscribed tasks like filling in spreadsheets, or entering data. For many professionals, your value comes from relationships and history of taking care of business. Any AI cannot replace actual experience. In fact it will mask those without experience since they can posture and present ideas they may not have the actual knowledge to support. But that's nothing new, what is new is going to have more effect on the non-participants. As most forums post: The best way to have AI not replace you is to use AI.
Still, skepticism should be the default. My personal interest as a human-computer educator and practitioner is how people without the skills and experience in using complex computing interfaces can skip much of that to get to the result. It's thrilling to throw in a spreadsheet of numbers and facts and just ask if there's any overall trend or value captured in it. It saves time on research, as you'll encounter as I ask AI to critique my own opinions and conclusions. Outside of elite universities like the one I teach at, most have never had access to a all-purpose information machine that obeyed their every whim.
But there's more room to skip and wait, I agree with these Reddit posters that summed up why this is another distraction:
* Once again, another big change that we're all going to have to work hard to adapt to
* It'll likely benefit others way more than it benefits me. Disruption is only good for the investors. it kinda sucks for everyone else
* There'll be huge gains somewhere else in the market that'll make everything more expensive. People with giant sacks of gold rush money are going to move to my town and buy up property, making prices go through the roof
* Tons of uncertainty that makes my planning for the future more difficult
* There's a deluge of hype and marketing bullshit with morons foaming at the mouth over AI this and AI that, and it's annoying. they probably don't even understand AI, too, they're just rat racers desperately chasing money and success. some of them are gonna win big and we'll have to hear them brag, and the others will bankrupt themselves and we'll have to hear them whine.
* We just revolutionized everything like 5-10 years ago, do we really need to do this again? life's just fine as it is. the pace is too fast.
* im already bracing for another round of "huh huh well ya shoulda gone out west and mined some gold then, sucks to suck" while dealing with all of the above. you know, some people just want to live normal lives and not chase the latest new hyped up thing all the time.
Yet, if you give yourself a chance and try to see how the game can be played, the book tackles most of this from a human perspective. How could this change the way we live and work? What does it mean for promotion, for education, for the systems we rely upon to create stable environments to live and raise a family? Along the way there will be tech jargon and explanations, but you can always skip, and I promise, there are better endings than this!
[[Surviving AI: How can you choose the right path?]]