The New Economist published an essay by Professor Yuyu Chen (PKU):

陈玉宇：价格理论的复归（人工智能时代的一篇经济学散文）

The Resurrection of Price Theory (An Economics Essay for the Age of Artificial Intelligence)

New Economists WeChat Subscription Account

Well, first of all, price theory is never “gone” (studying it right now). But I agree with most of the views of the essay and I think Prof Chen’s points are solid and insightful. Markets as discovery machines is vital for uncover the true value of AI — assuming the institutional and distributional conditions for open markets will hold during precisely the transition that puts them under maximum stress.

Details below are some reading notes (translated and paraphrased) and comments:

Disclaimer: all mistakes are my own.

The “old foundation” fallacy

Treating AI as a cost-compression exercise on today’s task list repeats the historical error made about steam, electricity, and the internet. AI’s most important economic consequence is not the cheap replacement of existing jobs but the explosive expansion of what is feasible to produce — and this makes the price mechanism more essential, not obsolete.

Comment: but price mechanism is not omnipotent. We need a general “price mechanism” a good “allocation mechanism”.

As AI collapses the costs of cognition, prototyping, and matching, the binding constraint shifts from “can machines produce it?” to “how does society discover what is worth producing?” Scarcity doesn’t vanish; it mutates into scarcity of fit (is this right for me, now?), trust (will advice actually be acted on?), and direction (which of a million possible paths deserves real capital and attention?). Human roles migrate toward judgment, responsibility, and relationships.

The Hayekian update

Because preferences are heterogeneous, contextual, and often unknown even to consumers until products and prices appear, no central planner or algorithm can substitute for the distributed experiment of markets — prices, profits, and losses are society’s mechanism for discovering dispersed, local knowledge (an updated Hayek).

Note: Hayek’s original argument (1945). The case against central planning was never just “computers aren’t fast enough.” Hayek’s point in The Use of Knowledge in Society was that the information a planner would need doesn’t exist in any collectible form. It’s dispersed as local knowledge: the factory foreman knows this machine runs hot in summer; the shopkeeper knows her regulars stopped buying a product after a local rumor; you know that today, specifically, you’d pay extra for a quiet café because you slept badly. This knowledge is tacit, contextual, fleeting, and often unknown even to the person holding it until a choice forces it out. Prices solve this without anyone collecting it: when tin becomes scarce, the price rises, and millions of people economize on tin without knowing why it’s scarce.
I don’t really agree with Hayek and the updated Hayek argument. Discovery-by-price is only one discovery mechanism, and often not the best one.

The seesaw

Eventually, the application layer will dwarf the infrastructure layer: a hypothetical $2T-revenue AI company is only sustainable if downstream users create $20–40T of new value, just as electricity mattered less than the industries it powered. The institutional task is therefore not to plan the AI economy but to keep markets open — free entry, competition, antitrust, data portability — so that nameless future demands have a path to surface.

True

Some underdeveloped points:

  1. Distribution and re-distribution is waved about but it might be one of the central debate. We hand wave that industry transition/upgrade increase total welfare, but transition itself is painful. Yes, market and price would optimize itself towards more total welfare. But if AI’s gains accrue heavily to capital and a small cognitive elite while displacement is broad, the political preconditions for keeping markets open may collapse
  2. Speed asymmetry. Past transitions took decades, giving labor markets a generation to adjust. If AI displaces faster than new demand categories can institutionalize (training, credentialing, trust-building all take years), the “new jobs will come” logic can be true in the long run and still socially catastrophic in the interim.
  3. AI can distort the price mechanism itself. The essay treats prices as a neutral discovery device, but AI enables near-perfect personalized pricing, algorithmic collusion, manipulated preferences, and demand that is manufactured (engagement-optimized) rather than discovered. If the instrument of discovery is itself algorithmically gamed, “willingness to pay” becomes a noisier signal of genuine value. This deserves at least as much attention as monopoly.
  4. The 10%/5% cost-share assumption is doing enormous work. If foundation models commoditize, the cost share could be 1% (implying an implausibly huge downstream) or, if platforms vertically integrate, they might capture the application layer themselves. It could just end up being one big company.
  5. Whose preferences? The Hayekian framework takes preferences as sovereign data to be discovered. But AI systems increasingly shape preferences (recommendation loops, companionship products, addictive design). When the supplier of the good also engineers the desire for it, “the market discovered a real demand” and “the market created a dependency” become hard to distinguish. This is a problem classical price theory wasn’t built for.
  6. No demand-side macro. Two trillion in revenue requires purchasing power. If labor income shrinks before new income sources emerge, who buys the $40T of downstream services?