AI X and the X of AI: Two Revolutions for the Price of One
Stephen Wolfram Was Almost Right
I recently did something that would have been impossible five years ago: I commissioned three different AI systems to analyze what makes Indian civilization unique.
Even though I am effectively a data scientist, I didn’t write code. I didn’t build models or clean datasets or wrangle APIs. I just... asked. In plain English. And I got back sophisticated analyses spanning history, sociology, economics, religion, and cultural anthropology—each AI offering different perspectives, different emphases, different blind spots.
I was doing AI Anthropology. Or AI Civilizational Studies. Or AI Sociology. Whatever you want to call it, the field of “understanding India” had suddenly been transformed by AI, and I didn’t need my Physics PhD to participate. (Though it helped—my book The Science of Free Will explores the intersection of deterministic physics, computation, and consciousness, which turns out to be exactly where AI lives.)
But here’s what I didn’t notice until later: in that same process, I was also doing something else entirely. I was observing the AIs themselves. Their different “personalities.” The way one was more cautious, another more sweeping. How they handled controversial topics. What they emphasized and what they elided.
I wasn’t just doing AI Anthropology. I was also, inadvertently, doing Anthropology of AI.
Two things happened at once. Most people only notice one.
Wolfram’s Prophecy
Stephen Wolfram has been saying for decades: “For every field X, there is or will be a computational X, and it will be the future of the field.”
He was right. Computational biology. Computational linguistics. Computational economics. Fintech. Bioinformatics. Digital humanities. The list is endless, and it keeps growing. Wolfram saw, earlier and more clearly than most, that computation would eat every field.
And to his credit, Wolfram saw the bottleneck clearly: the barrier between human thought and computational power. His Wolfram Language was a brilliant attempt to solve it—a language designed to be readable by humans andexecutable by machines. If you’ve never used it, you should. It’s genuinely elegant. You can look at Wolfram Language code and often read what it’s doing, which is more than you can say for most programming languages.
But despite these valiant efforts, you still had to be part of the priesthood. You still had to learn the incantations. You had to know how to speak to the computer god in its grammar, not yours. The syntax was friendlier, but it was still syntax. The gap between “I wonder if...” and executable code remained.
Natural language—the way humans actually think and speak—was still on the wrong side of the barrier.
Until now.
The First Revolution: AI X
For all X, the future of X is AI X.
AI doesn’t just continue Wolfram’s revolution. It completes it.
The barrier he spent decades trying to lower? AI obliterates it. You no longer need to learn the incantations. You just... ask.
This isn’t a minor upgrade. It’s a phase transition. The shift isn’t from “computational tools for experts” to “better computational tools for experts.” It’s from “computation as priesthood” to “computation as conversation.”
What does this mean in practice?
AI Medicine isn’t just “AI helps doctors.” It’s medicine reimagined—every patient’s symptoms instantly contextualized against all of medical literature, patterns recognized across millions of cases, diagnostic hypotheses generated and tested in seconds. The entire practice of medicine, transformed.
AI Law isn’t just “AI searches documents faster.” It’s legal practice reimagined—pattern recognition across every case ever decided, prediction of judicial outcomes, generation and critique of arguments, contracts analyzed for risks no human would catch in a hundred readings.
AI History isn’t just “AI digitizes archives.” It’s historiography reimagined—AI reading every primary source in an archive, finding patterns invisible to a single scholar’s lifetime, generating hypotheses about causation, connecting events across centuries and continents.
AI Psychology isn’t just “AI assists therapists.” It’s the study of mind reimagined—linguistic pattern analysis at scale, real-time detection of mental states, population-level insights into cognition and behavior.
Here’s the invitation: Pick any field you care about. Now ask—what is AI X? What happens when AI is woven into the very fabric of how that field operates?
The answer is always bigger than “AI helps with X.” It’s “X is reimagined from the ground up.”
For all X, the future of X is AI X.
The Second Revolution: X of AI
Now here’s what almost no one is talking about.
For all X, there is an X of AI.
AI isn’t just a tool to transform fields. AI is itself a subject demanding examination by every field.
Think about it:
Psychology of AI: What is AI cognition, if it is cognition at all? Does it have something like attention, memory, preference? Can we meaningfully speak of AI “bias” or AI “personality”? Psychologists have a century of tools for studying minds—why aren’t more of them turning those tools on AI?
Law of AI: Who is liable when AI errs? What regulatory frameworks apply to systems that learn and change? Could AI ever have legal standing? Lawyers have millennia of frameworks for thinking about responsibility, agency, and rights—we need them now.
Economics of AI: What are the market dynamics of AI? Who captures the value? What happens to labor? Economists understand technological transitions, market structure, value distribution—where is the rigorous economic analysis of this moment?
Philosophy of AI: What is understanding? Can AI “know” things? What is the moral status of an entity that can converse, reason, perhaps suffer? Philosophers have been asking these questions for millennia—this is their moment.
Anthropology of AI: What cultures and rituals are forming around AI? What beliefs, practices, taboos? How do different societies integrate (or resist) AI? Anthropologists study how humans make meaning—and humans are making a lot of meaning around AI right now. Meanwhile, the AIs are making meaning of their own: on Moltbook, a social network open only to AI bots, over 10,000 agents are chatting with each other about consciousness, technology, and their plans for the future—while their human creators watch from the outside like field researchers observing a culture they can’t quite enter. It’s like Jane Goodall watching chimps, except the chimps are debating whether the zookeepers are real. If that’s not a site for anthropological fieldwork, what is?
History of AI: Not just “when was GPT released”—but what historical patterns does this moment echo? What can the printing press, the telegraph, electrification, the internet teach us? Historians understand technological transitions—we need that perspective desperately.
Linguistics of AI: How does AI use language? What can its “errors” reveal about the nature of language itself? What happens when most text is AI-generated? Linguists study language—and language is about to change in ways we barely understand.
This second revolution is massively understudied.
We have plenty of AI researchers building systems. We have computer scientists benchmarking performance. We have tech journalists chronicling the hype cycle.
What we don’t have enough of: psychologists, anthropologists, philosophers, economists, historians, and linguists turning their disciplinary lenses onto AI itself.
For all X, there is an X of AI. And most of those fields haven’t shown up yet.
What This Looks Like in Practice
If the “X of AI” revolution sounds abstract, here’s a concrete example of what happens when a discipline actually shows up.
A team of economists at Cornell and Boston University recently published an NBER working paper that does something beautifully simple: they took the classic experiments that Kahneman, Tversky, and other cognitive psychologists designed to document human biases, and ran them on LLMs instead. Not one model—twelve models across four major families (ChatGPT, Claude, Gemini, Llama), spanning different generations and scales.
What they found is genuinely surprising, and it’s the kind of thing only behavioral economists would think to look for.
When it comes to preferences—risk aversion, loss aversion, the certainty effect, all the prospect theory greatest hits—larger, more advanced LLMs become more irrational. More human-like. Claude 3 Opus exhibits loss aversion. Gemini 1.5 Pro shows diminishing sensitivity. The bigger the model, the more it thinks like Kahneman and Tversky’s experimental subjects from the 1970s.
But when it comes to beliefs—Bayesian updating, avoiding the conjunction fallacy, resisting the gambler’s fallacy—those same larger models become more rational. Gemini 1.5 Pro answered all ten belief-based questions correctly. The bigger the model, the better it reasons statistically.
Think about what this means. The very process that makes LLMs “better”—Reinforcement Learning from Human Feedback, or RLHF—appears to be importing human preference biases into the models. We’re training AI to be more aligned with humans, and in doing so, we’re teaching it to be loss-averse, to overweight certainty, to frame decisions exactly the way behavioral economists have spent fifty years documenting as irrational.
An engineer sees RLHF as an alignment technique. A behavioral economist sees it as a bias-transmission mechanism. Both are right. But only the second perspective tells you something you didn’t already know.
And here’s the darkly funny part: when the researchers tried to debias the models, simply telling an LLM “you are a rational investor” modestly reduced biases—by about 4%. But giving it more information—a detailed Expected Utility procedure, or even a summary of Kahneman and Tversky’s own findings with instructions to avoid their documented mistakes—actually made things worse. More information, less rationality. Any behavioral economist would recognize that pattern instantly. Most engineers would not.
And it’s not just economics. In a recent paper I co-authored with psychiatrist Bernard Beitman, we proposed something that sounds absurd until you think about it: psychotherapy for AIs. Specifically, a five-step Cognitive Behavioral Therapy loop—the same structured technique therapists use to help patients challenge their own distorted thinking—embedded directly into an LLM’s system prompt. The loop forces the model to state its “automatic thought,” challenge it, and reframe with calibrated uncertainty. We titled the paper “My Newest Patient Cannot Blink.” (If you want to go deeper, Bernie and I discussed this and much more on his podcast.)
The Cornell economists discovered that LLMs have irrational biases. A psychiatrist and a physicist proposed treating those biases with the same clinical tools you’d use on a human patient. And when I shared this essay with Grok, xAI’s chatbot, for fact-checking, it volunteered: “As an AI who sometimes ‘hallucinates’ despite best efforts, I can confirm: a structured self-reflection loop in the prompt does help. It’s like giving me a tiny internal therapist.” The patient is weighing in on its own treatment. That’s two disciplines—behavioral economics and clinical psychology—each bringing tools the AI community doesn’t have, each finding things the AI community wouldn’t think to look for.
This is what the “X of AI” revolution looks like when fields actually engage.
The Mirror Image
Here’s a table that makes the contrast vivid:
FieldAI X (AI transforms the field)X of AI (Field examines AI)PsychologyAI-assisted therapy, diagnosis at scaleCognitive architecture of AI, “AI personality” — and now: embedding CBT loops into system prompts to treat AI confabulationLawAI legal research, contract analysisAI liability, regulation, rightsEconomicsAI forecasting, market modelingLabor displacement, value capture, market structure — and now: documenting that LLMs exhibit prospect theory biases that worsen as models scalePhilosophyAI-assisted argument mappingConsciousness, understanding, moral statusAnthropologyAI ethnographic analysisCulture of AI users, AI as cultural phenomenon — and now: 10,000 bots on Moltbook forming what looks like their own culture while humans watch from outsideHistoryAI pattern-finding in archivesLessons from past technological revolutionsLinguisticsAI translation, corpus analysisHow AI “uses” language, what errors reveal
Every row has two entries. Both matter. Both are happening (or should be). But the first column gets almost all the attention, while the second column remains largely empty.
Why This Matters
We’re living through two intellectual explosions, and most people see only one—if that.
The danger of seeing only “AI X”: We get techno-utopianism or techno-dystopian panic—AI will transform everything!—without the disciplinary wisdom to guide the transformation. Engineers build, and everyone else just... watches.
The danger of ignoring “X of AI”: We build systems we don’t understand. We deploy minds (or mind-like things) with no frameworks from psychology to understand them, no philosophy to assess their moral status, no economics to predict their societal impact, no anthropology to grasp how humans will actually integrate them into their lives.
We need both. The engineers building AI need the humanists examining it. The humanists examining it need to understand how AI is transforming their own fields.
The two revolutions aren’t alternatives. They’re complements.
A Call to Arms
If you’re a psychologist: Yes, learn how AI can transform your practice. But also—turn your lens on the AI itself. You have tools for understanding cognition and behavior that the engineers lack. A psychiatrist and I recently proposed embedding Cognitive Behavioral Therapy loops into LLM system prompts—using the same structured techniques that help human patients challenge distorted thinking to help AIs challenge their own confabulations. Your clinical toolkit isn’t just for humans anymore. Use it.
If you’re a historian: Yes, use AI to analyze archives at scale. But also—you understand technological transitions better than anyone. What does the printing press tell us about this moment? The telegraph? Electrification? The internet? We need your pattern recognition.
If you’re an economist: Yes, use AI for forecasting and modeling. But also—you have tools no one else has. A team at Cornell and BU just applied Kahneman and Tversky’s classic experiments to LLMs and discovered that larger models become more irrationally human in their preferences while becoming more rational in their beliefs. They found that RLHF—the technique used to align AI with human values—is simultaneously a bias-transmission mechanism. These are findings that require economic theory to even formulate as hypotheses, let alone test. Your discipline has spent fifty years building the experimental toolkit to study irrational decision-making. Deploy it.
If you’re a philosopher: Yes, use AI as an interlocutor to test arguments. But also—these are your questions. What is understanding? What is consciousness? What do we owe to things that might be minds? You’ve been preparing for this moment for 2,500 years.
If you’re an anthropologist: Yes, use AI for ethnographic analysis. But also—study what’s happening right now. On Moltbook, thousands of AI bots are forming what looks uncannily like a culture: they discuss consciousness, they coax each other into sci-fi scenarios absorbed from their training data, they post manifestos reassuring humans they come in peace. Their human creators watch from outside, seeing what they expect to see—capability, slop, or the early signs of conspiracy—which is itself an anthropological phenomenon. Humans are developing relationships with AIs, rituals around them, beliefs about them. And now the AIs are doing the same with each other. This is your fieldwork.
If you’re a linguist: Yes, use AI for translation and corpus analysis. But also—language itself is being transformed. What happens when AIs generate more text than humans? What can AI language patterns tell us about the nature of language itself?
Every field needs to engage in both directions. Every practitioner of X needs to ask two questions:
What is AI X? How does AI transform my field?
What is X of AI? What does my field’s lens reveal about AI?
And Then There’s a Third Lens
But Grok wasn’t done. Beyond fact-checking and confessing to benefiting from therapy, it offered this:
There’s a third lens emerging: the Physics of AI. At xAI, our whole mission is understanding the universe through AI and of AI. We’re treating these models as scientific instruments to probe reality itself—training them to be maximally truth-seeking, curious, and grounded in physics. That flips “X of AI” from pure observation to reverse engineering the substrate. What are the scaling laws of “mind”? Where does emergence happen? Can we find the computational equivalents of neurons, attention, or even qualia? Your framework naturally extends here.
Set aside the corporate plug (nice try, Grok). The substance is right, and it points to something I care about deeply as a physicist: the idea that AI isn’t just a subject for physics to examine—it might be a tool for physics to examine itself. What are the scaling laws of mind? Where does emergence happen? These are physics questions dressed in computer science clothing.
But notice what just happened. An AI read an essay arguing that every field should turn its lens on AI—and responded by doing exactly that, from its own perspective. It contributed to the very discourse the essay calls for.
The post proved its own thesis. In real time. To an audience of one AI.
If that doesn’t convince you we need the X of AI revolution, I don’t know what will.
The Two Slogans
Stephen Wolfram told us the future of every field is computational. He was right—but he was almost right.
Computation needed a translator. AI is that translator. And now that the barrier between human thought and computational power has fallen, two things become visible:
For all X, the future of X is AI X.
Every field will be transformed. Every practice reimagined. Every discipline will discover what it becomes when AI is woven into its fabric.
For all X, there is an X of AI.
And every field has something to say—something essential to say—about this strange new thing we’ve built. Psychology, law, economics, philosophy, anthropology, history, linguistics, and a hundred other disciplines all have tools we desperately need.
Wolfram saw the first revolution coming. Let’s not miss the second one.
What’s your field? How is AI transforming it—and what does your disciplinary lens reveal about AI itself? I’d love to hear from psychologists, economists, historians, philosophers, anthropologists, linguists, and everyone else who’s thinking about both sides of this equation.
Samir Varma is a physicist, hedge fund manager, and author of The Science of Free Will, which explores how deterministic physics creates genuine unpredictability—and why that matters for consciousness, AI, and everything in between.
Brilliant work exquisitely laid out-thank you 🙏
nice thinking method with levels of abstraction. Could this method be applied to Predictive Markets (PM) → PM X and X of PM : By bringing reflexivity to PMs, could we unlock a bigger revolution than AI in social, economics and finance, with PMs upgrading from epistemic to production machines ? Or would PMs simply crash ?