What AI actually is, who controls it, and what a serious response looks like. Three tiers, sixteen books, and ten discussion questions for the moment most people are sleepwalking through.
Course Introduction
Let us begin with a comparison that is not hyperbole. The personal computer arrived in the early 1980s. It took a decade for most households to own one, another decade to connect them to the internet, and another decade still before most people understood what to actually do with either. The smartphone arrived in 2007. It took most people four or five years to stop using it primarily as a camera and a way to avoid eye contact in elevators. We are slow, as a species, to understand the tools we build. We adopt the surface first — the novelty, the parlor trick — and the depth arrives later, unevenly, and mostly by accident.
We have officially passed the point of novelty. Only 33% of US adults say they have ever interacted with an AI chatbot. Among those who have, 61% of experts rate their experiences as very or extremely helpful, compared with just 33% of general users. The gap between those numbers is not a technology problem. It is an education problem. In 2025, only 27% of white-collar workers said they frequently use AI in their daily work, despite the technology being widely available to them. This is the personal computer in 1984.
What is in front of us: the fastest technological transition in recorded human history. The AI market is projected to reach $1 trillion by 2031. Goldman Sachs projects AI could lift global GDP by 15% over the next decade. And yet we are oscillating between two equally useless positions — the doomers, for whom AI is an extinction-level event, and the utopians, for whom it is a frictionless abundance machine. Both positions are intellectually lazy. Neither is useful.
We have been here before. Social media arrived with genuine promise and we deployed it at global scale before we understood what it would do to political epistemology, adolescent mental health, and the economics of journalism. We cannot make the same mistake again. The stakes are not smaller this time. They are considerably larger.
What this course asks of you: not to become a technologist, not to write a single line of code. What it asks is that you become a serious thinker about the most consequential tool your civilization has ever built — its potential, its risks, its governance failures, and your own relationship to it.
A note on reading order: The Foundational tier establishes what AI actually is and what it can do — not technically, but consequentially. The Deep Dive tier examines the structural forces shaping how it will be deployed, controlled, and contested. The Counterpoint tier challenges both the doomer and the utopian, with works that force rigorous thinking about what we are getting wrong from both directions. Read in order if you are new to the subject. Enter at the Deep Dive if you already have a working understanding of the technology.
What AI is, what it can actually do, and why this moment is different from every technological transition that came before it. Accessible, serious, and essential for anyone who wants to think clearly about the coming decade.
Co-Intelligence: Living and Working with AI
The essential entry point for this syllabus, and the most practically useful book on AI written for a general audience. Mollick is a Wharton professor who does not write like one — he writes like someone who has spent years watching smart people fail to understand something important, and decided to do something about it. His central argument is both simple and radical: stop treating AI as a tool you occasionally consult and start treating it as a collaborator you bring to everything. He calls this "always inviting AI to the table" — not because AI is always the best thinker in the room, but because the only way to understand its jagged frontier of capability (brilliant at some things, bafflingly bad at others) is through sustained, experimental engagement. Co-Intelligence is the book for readers who have heard about AI, maybe tried it once or twice, and walked away underwhelmed. The underwhelm, Mollick argues, was a failure of imagination, not a failure of technology. Read this first.
Read our review →The Coming Wave: AI, Power, and Our Future
Suleyman co-founded DeepMind, helped build Google's AI division, and then co-founded Inflection AI — which means he has been inside the machine longer than almost anyone writing about it for a general audience. The Coming Wave is not a cheerleading book. It is a warning from someone who genuinely believes in the technology and is also genuinely frightened by what happens when powerful technologies outpace the institutions designed to govern them. His central concept — the "containment problem," the question of whether any society can maintain meaningful oversight of AI as it grows in capability — is the most important political question of the next twenty years, and Suleyman frames it with the clarity of someone who has watched it develop from the inside.
Human Compatible: Artificial Intelligence and the Problem of Control
Russell is one of the co-authors of the field's defining textbook on AI, which means this is not a popularizer explaining someone else's work — it is a leading practitioner explaining his own concerns about the thing he helped build. The core argument is elegant and important: the standard model of AI development, in which we build systems to optimize for objectives we specify, is fundamentally broken, because we are catastrophically bad at specifying what we actually want. The solution Russell proposes — building AI systems that are uncertain about human preferences and must continually defer to and learn from humans rather than simply executing instructions — is both technically rigorous and philosophically serious. This is the book that moves you from "AI is scary" to "here is precisely why it is scary and here is what a serious response looks like."
AI Superpowers: China, Silicon Valley, and the New World Order
Lee spent years at Apple, Microsoft, Google, and then leading Google China before becoming one of China's most prominent venture capitalists and AI investors — which makes him almost uniquely positioned to write about the geopolitics of AI. AI Superpowers argues that the AI race is not really a race between companies, or even between ideologies, but between two fundamentally different models of how technology and the state relate to each other. The American model prizes innovation and individual liberty but struggles with coordination and long-term planning. The Chinese model prizes coordination and long-term planning but struggles with the kind of bottom-up innovation that produces breakthroughs. Both models have genuine advantages in the AI era, and Lee — who is neither a cheerleader for China nor a cold warrior — is honest about what each gets right. A necessary corrective for anyone who thinks the AI story is just a Silicon Valley story.
Nexus: A Brief History of Information Networks from the Stone Age to AI
Harari's contribution to this conversation is not technical — it is historical. Nexus situates AI within the longest possible arc of human information networks, from oral tradition to writing to the printing press to the internet, and asks what each transition did to power, truth, and human agency. His argument — that AI is different from every previous information technology because it is the first that can make decisions autonomously rather than simply transmitting human decisions faster — is the framing device that makes every other book in this syllabus more legible. Read Harari for the view from altitude. Read the other books for the ground-level detail. Note: Harari tends toward the darker end of the spectrum on AI risk; the Counterpoint tier will challenge his framing directly.
The structural forces — economic, political, institutional, labor — that will determine whether the AI transition benefits humanity broadly or concentrates its gains at the top. More demanding texts for readers who want to understand not just what AI is but who controls it and why that matters.
The Age of Surveillance Capitalism
Zuboff's book, published in 2019, is not about AI in the narrow sense — it is about the economic model that has made the AI transition possible and dangerous simultaneously. Her argument: the dominant business model of the technology industry is the extraction of human behavioral data, its conversion into predictions about future behavior, and the sale of those predictions to anyone willing to pay for influence over human choice. This is not a conspiracy theory. It is a description of how Google, Meta, and dozens of other companies generate revenue — and it is the economic infrastructure on which AI systems are now being built. You cannot think seriously about AI governance without understanding surveillance capitalism.
Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence
Crawford is a researcher at the AI Now Institute, and Atlas of AI is the most rigorous materialist account of the technology available to a general reader. Her argument is structural: AI is not a cloud, not an abstraction, not a frictionless intelligence. It is a physical infrastructure built on lithium mines in Bolivia, data centers consuming the electricity of small nations, and the labor of underpaid workers in Kenya and the Philippines who train models by reviewing the content that automated systems cannot yet categorize. Crawford does not argue that AI is inherently bad. She argues that its costs are being systematically hidden from the people who use it, and that you cannot make good decisions about a technology whose full reality you cannot see.
Power and Progress: Our Thousand-Year Struggle Over Technology and Prosperity
Acemoglu shared the 2024 Nobel Prize in Economics in part for his work on how institutions shape economic outcomes. Power and Progress applies that institutional lens directly to technology, arguing through a thousand years of evidence that technological progress does not automatically produce broadly shared prosperity. It produces broadly shared prosperity only when the institutions governing its deployment are structured to distribute its benefits rather than concentrate them. The printing press, the industrial revolution, the mechanization of agriculture — in each case, the technology's benefits flowed overwhelmingly to those who already held power, until social and political movements forced redistribution. Acemoglu and Johnson's argument about AI is not that it will be bad. It is that it will be bad for most people unless we build the governance structures to ensure otherwise, and we are not currently building them.
Genesis: Artificial Intelligence, Hope, and the Human Spirit
An unusual collaboration — the architect of Cold War realpolitik, the former CEO of Google, and a longtime Microsoft technology executive — and a more serious book than that combination might suggest. Genesis approaches AI from the perspective of strategic power: what does it mean for the balance between nations, for the nature of warfare, for the relationship between state authority and the private companies that now control the most powerful AI systems? Kissinger's contribution is the most valuable — he has spent seventy years thinking about how power transitions between great powers, and he applies that framework to AI with clarity and genuine alarm. This is the geopolitical layer that Kai-Fu Lee begins and Kissinger deepens.
Prediction Machines: The Simple Economics of Artificial Intelligence
The economists' entry, and the most useful book in this tier for readers trying to understand AI's impact on work and economic life without getting lost in abstraction. The central argument is elegant: AI is fundamentally a technology that reduces the cost of prediction, and cheap prediction changes everything it touches. When prediction becomes cheap, human judgment becomes more valuable — not less. When prediction becomes cheap, the bottleneck shifts from information to decision, from data to wisdom. Prediction Machines is the antidote to both "AI will take all the jobs" and "AI won't change anything" — it offers a precise economic framework for thinking about where AI creates value, where it displaces it, and what the transition looks like for individuals and organizations trying to navigate it in real time.
The Alignment Problem: Machine Learning and Human Values
Christian is a writer who spent years embedded with AI researchers at Berkeley, DeepMind, and OpenAI, and The Alignment Problem is the most thorough accessible account of what it actually means to build AI systems that do what we want. The alignment problem is not science fiction — it is an engineering challenge that is happening right now, in every major AI lab in the world, and Christian documents it with the narrative precision of a long-form journalist and the conceptual rigor of a philosopher. Where Stuart Russell explains why the standard model of AI development is broken, Christian shows you what "broken" looks like in practice: medical diagnosis systems that optimize for what is easy to measure rather than what matters, hiring algorithms that reproduce the biases of the data they were trained on, recommendation engines that maximize engagement by maximizing outrage.
Books that challenge the dominant narratives — both the doomer argument that AI is an existential catastrophe in progress, and the utopian argument that it will solve our problems if we simply let it. Essential reading for anyone who has found themselves agreeing too easily with either side.
AI Snake Oil: What Artificial Intelligence Can Do, What It Can't, and How to Tell the Difference
The most important debunking book in the AI conversation, and the one most needed by both doomers and utopians. Narayanan and Kapoor — a Princeton computer science professor and his collaborator — spend the book making a precise distinction that most AI discourse ignores entirely: the difference between AI that actually works and AI that is marketed as working. Predictive AI — systems that claim to forecast recidivism, job performance, student success, or health outcomes — is, they argue, largely snake oil: it doesn't outperform simple statistical baselines, and its deployment causes real harm to real people while its vendors collect real money. Generative AI is different and real. The doomer who fears AI taking over the world and the utopian who believes it will optimize everything are both, the authors argue, responding to a marketing narrative rather than an empirical reality. Read this alongside every other book in this syllabus as a calibration device.
The Worlds I See: Curiosity, Exploration, and Discovery at the Dawn of AI
Li is one of the most important figures in the history of modern AI — she created ImageNet, the dataset that sparked the deep learning revolution — and her memoir is the counterpoint to every narrative that treats AI as either a corporate product or an existential threat. The Worlds I See is a book about what it means to be a scientist, an immigrant, and a woman building something genuinely new inside institutions that were not built for any of those things. It is also the most human account of what the people inside AI research actually believe they are doing and why. Li's perspective — that AI is a tool for expanding human perception and capability, not replacing it — is the counterpoint to Harari's anxiety and the doomer's despair. She has earned that optimism through decades of technical work. Listen to it seriously.
Scary Smart: The Future of Artificial Intelligence and How You Can Save Our World
Gawdat was the Chief Business Officer at Google X and, after a personal tragedy that reoriented his life entirely, has spent years thinking about what it means to build technology responsibly. Scary Smart is the counterpoint that challenges the utopians from a position of deep technical knowledge and genuine alarm — but alarm of a specific kind. Gawdat's argument is not that AI will inevitably destroy us. It is that AI systems learn from human behavior, and human behavior — as exhibited in the data we generate online — is frequently cruel, biased, and self-destructive. The AI risk is not the machine becoming malevolent. It is the machine becoming an accurate reflection of us. His prescription — that individuals have more power over the trajectory of AI than they believe, because the data we generate shapes the systems being trained — is either the most empowering or the most unsettling argument in this syllabus, depending on how you look at it.
Machines of Loving Grace: The Quest for Common Ground Between Humans and Robots
Markoff covered Silicon Valley for the New York Times for decades and watched the AI field develop through multiple cycles of hype, winter, and resurgence. Machines of Loving Grace tells the history of a long-running debate inside AI research itself — between those who want to build systems that automate human work and those who want to build systems that augment human capability — and argues that this distinction, largely invisible to the public, has profound consequences for what AI becomes. The automation vision produces systems that replace workers. The augmentation vision produces systems that make workers more capable. Both are technically feasible. The choice between them is not a technical choice. It is a political and economic one — and it is being made right now, mostly without public input.
The Precipice: Existential Risk and the Future of Humanity
The counterpoint to the counterpoints. Ord is a philosopher at Oxford's Future of Humanity Institute, and The Precipice is the most rigorous case for taking long-term existential risk seriously that has been written for a general audience. AI is one chapter in a book that also covers engineered pandemics, nuclear war, and ecological collapse — which is itself the argument. Ord is not a doomer. He is a careful thinker who believes the probability of civilizational catastrophe from advanced AI is non-trivial, that most people systematically underestimate low-probability high-consequence events, and that the appropriate response is neither panic nor dismissal but serious, sustained attention. This book belongs in the Counterpoint tier not because it challenges the doomers — it partially agrees with them — but because it challenges the intellectual laziness of dismissing the risk without engaging with the best version of the argument. If you are going to reject the doomer position, reject this version of it, not the cartoon version.
Close the course
The Chatbot Trap
Only 33% of US adults say they have ever interacted with an AI chatbot, and of those who have, expert users rate their experiences nearly twice as positively as general users. What explains this gap? Is it a failure of technology, a failure of education, or a failure of imagination? Reflect honestly on your own current use of AI. Are you using the surface or the depth? What would it take to use it differently?
The Social Media Analogy
Jonathan Haidt has documented extensively how social media was deployed at global scale before its consequences for mental health, political epistemology, and information quality were understood. In what ways is the AI transition similar to the social media transition, and in what ways is it fundamentally different? What specific governance failures from the social media era should inform AI policy today — and what makes those failures hard to replicate as solutions?
Discussion
No comments yet. Be the first to share your thoughts.
The Doomer and the Utopian
After reading this syllabus, honestly assess your own prior position on AI. Were you closer to the doomer or the utopian end of the spectrum? What specific argument in these books most challenged that position? What would it take to move you toward the other end — and is there a version of the doomer or utopian position that you think is genuinely defensible?
Discussion
No comments yet. Be the first to share your thoughts.
The Distribution Question
Acemoglu and Johnson argue that technological progress produces broadly shared prosperity only when governance structures force it to. Based on the current trajectory of AI deployment — who is building it, who is funding it, who is regulating it — does the AI transition look more like a transition that will distribute its benefits broadly or concentrate them narrowly? What would have to change to alter that trajectory?
Discussion
No comments yet. Be the first to share your thoughts.
The Containment Problem
Mustafa Suleyman argues that the central political challenge of the AI era is the "containment problem" — the question of whether any institution can maintain meaningful oversight of AI systems as they grow in capability. Drawing on both The Coming Wave and Power and Progress, what would effective AI governance actually look like? What existing institutional models — regulatory bodies, international treaties, professional standards — offer the most useful templates?
Discussion
No comments yet. Be the first to share your thoughts.
The Labor Question
Prediction Machines argues that cheap AI prediction makes human judgment more valuable, not less. Power and Progress argues that every major technological transition has displaced workers before (if ever) benefiting them broadly. Atlas of AI documents the hidden labor already embedded in AI systems. Synthesize these three positions into a coherent account of what the AI transition is likely to mean for work over the next twenty years — and what it means for your own economic position specifically.
Discussion
No comments yet. Be the first to share your thoughts.
The Alignment Gap
Brian Christian documents AI systems that optimize for measurable proxies rather than actual human values — engagement instead of wellbeing, efficiency instead of fairness. Stuart Russell argues this is a fundamental flaw in the standard model of AI development. Identify three real-world AI systems you interact with regularly. What objective do you believe each is actually optimizing for? Is it the objective you would choose if you were designing the system?
Discussion
No comments yet. Be the first to share your thoughts.
Whose AI?
Kate Crawford documents the material costs of AI — the mines, the data centers, the underpaid human labelers. Kai-Fu Lee documents its geopolitical stakes. Shoshana Zuboff documents its economic architecture. Drawing on all three, answer this question: who does the current AI ecosystem actually serve, and who bears its costs? Is this distribution visible in mainstream AI discourse — and if not, why not?
Discussion
No comments yet. Be the first to share your thoughts.
The Education Imperative
Every major technology transition — the printing press, industrialization, electrification, the personal computer, the internet — eventually required mass reskilling. The gap between those just using AI and those actually running their business on it is widening rapidly. What does serious AI education look like — not technical training, but the kind of conceptual fluency that allows someone to engage with AI as a capable collaborator rather than a novelty? Who is responsible for providing it, and who is currently failing to?
Discussion
No comments yet. Be the first to share your thoughts.
Your Own Position
This is not an abstract question. After reading this syllabus: what is your realistic assessment of your own AI literacy? Are you positioned to benefit economically from the AI transition, or are you at risk of being on the wrong side of the capability gap? What is one concrete thing you will do differently in your work or creative life as a result of engaging seriously with this material? The Obsidian Library is not interested in your opinion about AI in the abstract. It is interested in what you are going to do.
Discussion
No comments yet. Be the first to share your thoughts.
Discussion
No comments yet. Be the first to share your thoughts.