Price Discovery in the Market of Ideas: When AI Decides Which Ideas Are Deployed
Starting from a Search
Recently, a small thing happened.
When I was searching for an article about artificial intelligence, translation, publishing, and economic growth, I found that search engines are no longer just returning a string of blue links. They have started to directly provide summaries, extract keywords, mark authors, and reorganize several concepts and judgments from a long article into an explanatory framework that can be instantly invoked.
Of course, this is not the first time. In the past few years, people have gotten used to such changes: Search engines are becoming question - answering machines, and question - answering machines are becoming explanatory machines. Users are no longer just searching for "information" but are asking "why" — Why can AI translate a book in just half an hour, but the publishing industry hasn't exploded? Why has the efficiency of individual local tasks skyrocketed, but the macro - growth rate hasn't taken off? Why does technology seem more and more like magic, but our lives are still being pulled by slow - moving variables such as education, healthcare, elderly care, family, organization, and institutions?
These questions are not simple. They are not factual queries but requests for explanations, requiring the answerer to understand technology, the market, organizations, institutions, and human reproduction within a common framework. More interestingly, today's large models have begun to extract such explanatory frameworks from long articles, compress them into a few concepts, and present them to users again.
On the surface, this seems like an "intellectual coronation": The algorithm seems to be hanging a medal on the chest of an article or an author, marking who proposed this concept, where that explanation comes from, and how this framework understands the current technological changes. Such a moment is certainly shocking because in the past, for an idea to gain such visibility, it often had to go through a long social process — being recognized by editors, reposted by the media, cited by peers, spread by students, and incorporated into textbooks. The ownership and influence of an idea were gradually formed in the slow memory of the community.
However, if we only focus on "AI crowning scholars," we are oversimplifying the problem. The algorithm is neither generous nor has a sense of honor. It doesn't pay tribute to human thinkers. It is just looking for a knowledge structure that can reduce its own answering cost, improve the quality of explanations, and meet user needs under specific objective functions and constraints. What really deserves our attention is not that an individual is noticed by AI, but that the pricing mechanism of the intellectual market has changed.
An AI citation is not an honor but a price. It tells us that in the new knowledge dissemination system, what kind of ideas are more likely to be discovered, what kind of explanations are more likely to be invoked, and what kind of concepts are more likely to be extracted from the vast texts and become cognitive tools shared by machines and humans. So, this is not an article about search engines, nor is it a hymn to the capabilities of large models. It aims to discuss an economic issue: What will happen to the intellectual market when the transaction cost of knowledge dissemination drops sharply?
The Long and Tedious Supply Chain in Knowledge Dissemination
Traditional knowledge dissemination is a long chain: Scholars write, editors select, the media spreads, readers read, peers cite, students memorize, and textbooks preserve. For an idea to be widely recognized from its inception, it has to go through many intermediaries, and each intermediary has its own judgment criteria and constraints — Editors are concerned about page layout and risks, the media about dissemination and timeliness, peers about norms and reputation, and readers about readability and usefulness.
This chain is not without value. On the contrary, many intermediaries undertake the functions of screening, editing, translating, and explaining. However, it is also a system with high transaction costs: Many valuable ideas fail to spread not because they lack explanatory power but because the discovery cost is too high. For an ordinary reader to understand "why AI hasn't made the publishing industry explode," they first need to realize that this is a question rather than simply believing that "increased efficiency will inevitably lead to an output explosion." Then they need to find relevant articles, be willing to read a tens - of - thousands - word economic essay, and finally be able to extract the concepts, mechanisms, and conclusions from it and connect them with their own experiences. The cost of this process is extremely high. So, the intellectual market in the old world often falls into a paradox: There is a lot of intellectual production, but little intellectual discovery. We think the scarce thing is ideas, but in fact, most of the time, the scarce thing is the mechanism to find ideas.
This is very similar to the product market. There is no shortage of producers or consumers in the world. The real difficulty is matching — Who needs what, who can provide what, and at what price a deal can be made. These pieces of information are scattered among countless people. The most important function of the price system is not just to make transactions happen but to compress the scattered knowledge into actionable signals. The same is true for the intellectual market: Authors know what they want to explain, and readers know what they are confused about, but there are huge costs of searching, matching, and understanding between them. Traditional editors, the media, and academic communities are the knowledge intermediaries in the old world. They reduce some costs but also create some frictions.
After the emergence of AI, this long chain has been compressed. The new chain becomes "scholars → corpus system → large model → hundreds of millions of users": The intermediate links have not completely disappeared, but their importance has been redistributed. The large model internalizes some of the costs of searching, screening, summarizing, translating, reorganizing, and explaining. It is not just a tool but more like a real - time intellectual clearing center — In the past, readers had to actively search for ideas; now, the moment users ask a question, the model will actively search for the ideas that can answer the question.
This is the revolution in transaction costs. The most important role of AI may not be to produce knowledge; at least in this sense, it is first to discover knowledge, reorganize the explanatory frameworks originally scattered in papers, monographs, speeches, newspapers, and long articles, and invoke them instantly when users ask questions. So, we shouldn't just ask whether AI can write articles. The more important question is: How does AI change the way articles are discovered, compressed, cited, and disseminated? Once this question is raised, we enter the realm of price theory.
Explanatory Power as an Asset
AI doesn't treat all texts equally. Some articles are very popular on social media, full of emotions, with neat sentences and loud judgments, but are difficult for large models to call stably; other articles may not cause an instant public outcry but are quickly absorbed by the model as an explanatory framework because of their clear concepts, well - defined mechanisms, and complete causal chains. Behind this is a simple and important truth: Large models prefer compressible ideas. Compressible doesn't mean simple or shallow; on the contrary, truly compressible ideas are often highly condensed, able to organize many phenomena with one concept and explain many seemingly scattered facts with one mechanism.
The "Baumol cost disease" is such a concept. It tells us why in some service sectors, labor productivity is difficult to increase as rapidly as in the manufacturing industry, but wages still rise with the overall economy. Thus, the costs of education, healthcare, nursing, and art performances become more and more prominent in the modern economy. The "translator's paradox" is also the same: AI can significantly improve the local efficiency of translation, but publishing won't simply explode because there is more than just translation from a foreign - language text to a real publication. There are also topics, copyrights, editing, proofreading, market judgment, distribution channels, readers' time, cultural needs, and institutional approvals. The bottleneck will only shift from one link to another, and an increase in local efficiency doesn't mean an explosion in overall output. The same goes for "human self - reproduction": Economic growth is not just about the increased efficiency of machines, software, and algorithms. A society also needs to reproduce people's bodies, knowledge, emotions, trust, families, organizations, and the next generation. AI can change many tasks but cannot eliminate the basic constraints of humans as social beings.
These concepts are valuable because they have high explanatory power — They are not emotions or slogans but transferable cognitive tools. From the perspective of information theory, they have a high compression rate, explaining large - scale real - world changes with less information; from the perspective of price theory, they are a scarce asset, reducing the cost for people to understand the world. This is also why AI prefers them: Not because AI respects depth or has an aesthetic sense, but because concepts with high explanatory power can reduce the organizational cost of the model's output. Facing users' complex questions, the model needs to find a path from the question to the mechanism and then to the conclusion, and concepts with a high compression rate just provide such a path.
Therefore, when AI invokes a certain concept, it is actually pricing this concept. This price is not a monetary price but a call price in the semantic market, manifested as being summarized, cited, reorganized, and repeatedly included in the answers. In the past, the price of an idea was mainly reflected through citations, reprints, sales, classroom dissemination, and public discussions; now, ideas have a new price — model invocation. This is a new price discovery mechanism.
However, price theory also reminds us that any increase in the price of an asset will stimulate an expansion of supply. If explanatory power becomes a hard currency, more and more people and machines will produce explanations. Then the question arises: When explanations become easier to produce, will explanations themselves still be scarce?
From Attention Economy to Explanatory - Power Economy
The keyword in the Internet era is attention. Whoever can capture people's attention gets dissemination — Titles should be sharp, emotions should be full, stances should be clear, and conflicts should be intense. Platform algorithms strengthen this tendency: They don't directly judge whether an idea has more explanatory power but only judge whether it can make users stay, click, forward, and comment. As a result, the intellectual market has been distorted by the price of attention. Many complex issues have been compressed into taking sides, many issues that require long - term thinking have been turned into hot topics, and many views with weak explanatory power but strong emotions have received far more dissemination rewards than their knowledge value. This is not a moral issue but a price issue: The attention market pays a high price for emotions, so producers naturally increase the supply of emotions; platforms give more weight to conflicts, so content producers naturally create more conflicts.
In the search - engine era, an idea often had to become a hot topic first to have a chance to be noticed. In the era of large models, another logic is emerging: When users ask questions to the model, they are often not looking for excitement but for explanations. They are asking why, how to understand, what it means, and what else they haven't seen. In such a scenario, click - through rate is no longer the only unit of pricing. What matters more is whether the answer has a structure, whether the explanation has logic, and whether the mechanism can string together the phenomena.
In the search era, what was contested was traffic; in the generative era, what is contested is the right to explain.
This doesn't mean that the attention economy will disappear. People are still attracted by emotions, and platforms still reward conflicts. Moreover, as we will see later, large models themselves are corporate products with commercial objective functions and may not naturally favor truth but may favor "sounding right." So, what is being said here is not a complete victory but an additional dimension: After the intervention of large models, explanatory power has for the first time begun to be directly priced by the machine system. In the past, if an idea didn't enter the media agenda, it was difficult to be seen by the public; now, as long as an idea can explain reality, it may be invoked when users ask questions. It doesn't have to become a hot topic first to be disseminated but can first become part of the answer and then gain visibility.
This is a significant change for knowledge producers. In the past, writing often needed to cater to dissemination; now, writing has been given an old - fashioned requirement again: to explain the world. Of course, this doesn't mean that good ideas will definitely win — The market is never a friction - free place, and price discovery is neither completed at once nor always correct. But the direction has changed. The intellectual market has at least added a mechanism to reward explanatory power, which is something worthy of serious attention.
When Intermediaries Become Producers
However, large models are not neutral referees. This is a piece of sobriety that must be maintained when understanding the intellectual market in the AI era. In the past, editors were intermediaries. They could decide what to publish but usually didn't directly produce the author's ideas; search engines were also intermediaries. They ranked links but didn't directly write complete answers; social media mainly determined distribution rather than directly generating content. Large models are different: They are simultaneously searchers, editors, summarizers, commentators, disseminators, and producers. They reorganize ideas while discovering them and generate explanations while citing them. For the first time in the intellectual market, there is a super - intermediary that is responsible for both price discovery and content production.
This will bring profound changes. First, large models have their own objective functions. They are not philosophers existing to pursue truth but corporate products in the real economic system, having to balance user satisfaction, computing cost, business model, brand reputation, legal risks, and regulatory boundaries. Second, they have their own risk constraints. They will avoid certain controversies, tend to use safe expressions, and take a compromising and evasive approach on many issues. The disputes surrounding copyright and training data remind us that large models are not truth - machines beyond the system but corporate entities within it. Third, they have their own semantic preferences. Different models have different training corpora, alignment methods, security strategies, and product positioning. Therefore, their absorption, compression, and expression of the same idea are also different.
Therefore, the price given by AI is not a pure price in a free market but more like a shadow price — a price formed under constraints. It reflects both scarcity and constraints, and contains both explanatory power and commercial interests, legal risks, regulatory boundaries, and the value judgments of algorithm designers. So, we can neither deny the price - discovery function of AI nor deify it: It is not the eye of God but a market participant with constraints. When we say that AI is re - pricing ideas, we must also ask: Who designed this market? Who owns the clearing center? Which ideas are more likely to be seen, and which ideas, even if they have explanatory power, will have their weights lowered? Which expressions will be encouraged, and which will be excluded? This is not only a question of price theory but also a question of institutional economics and political economics.
There is also an older economic issue hidden here: the distribution of property rights and rents. If large models are becoming the semantic clearing center of the intellectual market, then they are not only reducing transaction costs but also re - distributing the transaction surplus. In the past, publishers, universities, journals, peer reviews, and media institutions shared the rents in knowledge dissemination. They certainly had thresholds, biases, and rent - seeking, but at least they constituted a visible mechanism for reputation and income distribution. The change brought by large models is that they can absorb high - explanatory - power assets that humans have long deposited on the Internet into their answer systems to reduce output costs and improve product quality, but may not convert this invocation into actual rewards for knowledge producers: A concept may be used repeatedly, a mechanism may be continuously reorganized, and an author's explanatory framework may enter countless answers, while its origin, contribution, and income are quietly being diluted.
Thus, a new incentive gap has emerged in the intellectual market in the AI era — Explanatory power is priced, but the price may not be paid to the producers of explanatory power. This is not a simple copyright issue. Copyright is just the most superficial institutional form. The deeper problem is that if high - explanatory - power ideas can be invoked infinitely and reorganized without attribution, and producers cannot obtain monetary, reputational, or academic credit from it, then in the long run, the entire incentive structure of intellectual production will be rewritten. The market can discover value, but it doesn't mean that the market will automatically reward the creators of value. Once price signals cannot be converted into property - right income, it may induce a typical public - good supply dilemma — Explanation is non - competitive. The problem is not that it is used too much but that the people who create it do not receive rewards: All models need high - quality explanations, but there are not enough mechanisms to continuously reward those who create explanations.
The Concentration of the Right to Explain and the Expansion of Explanation Supply
A decrease in transaction costs does not necessarily lead to decentralization. This is a misunderstanding that many people have about technological progress: They think that since the Internet has reduced the cost of dissemination, the world will become more equal. This is not always the case. In many cases, a decrease in transaction costs actually strengthens economies of scale and brings stronger concentration. This is true for search engines, social media, e - commerce platforms, and probably large models as well. As more and more people understand the world through a few models, the right to explain begins to concentrate: How a model summarizes problems, selects frameworks, and organizes answers will affect the understanding of reality by hundreds of millions of people. In the past, media institutions competed for the right to set the public agenda; today, large models compete for the right to set the explanatory framework.
But this is only one side of the problem. On the other side, the supply of explanations is expanding rapidly. In the past, the cost of producing an explanatory framework that sounded rigorous, had a complete structure, and clear concepts was not low. A person needed to read, train, think, write, and also have a certain theoretical literacy. Today, large models can generate a large number of "seemingly reasonable" explanations in a few seconds. They have levels, concepts, mechanisms, and conclusions, and are even more orderly, fluent, and like a mature article than many human writers. So, a new phenomenon has emerged in the intellectual market: Explanations are starting to experience inflation.
This doesn't mean that explanations are no longer important. Precisely because explanations are important, AI produces them in large quantities. But when the production cost of a product drops suddenly, its relative price will change: What was once scarce gradually