Translating a book only takes half an hour, so why hasn't there been an explosion in publishing?
01. From the Translator's Paradox to Real - world Frictions
Today, AI may only take dozens of minutes to translate an English work of over a million words into Chinese. Strangely enough, the translation and publication industry has not boomed as a result. There aren't suddenly a thousand times more translated works in bookstores, and readers don't have a thousand times more time for reading. This fact reminds us that economic growth is not determined by the fastest machines, but by the slowest human, institutional, and market factors working together.
The translator's paradox is an excellent entry - point in the AI era. It shows us that a sharp increase in the efficiency of a single - point task does not equal a corresponding increase in the quantity of the final product; an exponential improvement in a technological ability does not mean that the macro - economic growth rate will skyrocket. AI can rapidly reduce the cost of certain cognitive tasks, but it cannot simultaneously eliminate copyright, editing, responsibility, trust, distribution channels, consumer attention, and market risks.
In the previous article, I tried to explain that the real significance of AI is not reflected in the old industrial table and cannot be summed up by the phrase "reducing staff and increasing efficiency". If we only ask whether a department can hire fewer people, whether a process can be speeded up by a few minutes, or whether a position will be replaced by a machine, we are still imagining a technology that is supposed to rewrite the coordinates from the perspective of the old world.
The deeper significance of AI lies in its ability to expand the feasible set of human production and consumption. Activities that were previously impossible, unthinkable, too costly, or impossible to organize are now included within the realm of possibility for the first time. The replaced positions have names, but most of the newly created activities don't. The old ledger lists "translation", "copywriting", "customer service", "programmer", while we can't yet name the new ways of learning, companionship, medical care, aesthetics, and organization that will emerge in the new world.
This is the simplest wisdom of price theory: growth is never just an increase in quantity on a fixed menu, but an expansion of the feasible set, a rearrangement of relative prices, a reorganization of activity boundaries, and continuous trial - and - error in local knowledge.
However, admitting that AI has expanded the boundaries of the feasible set does not mean accepting technological utopianism. In recent years, an increasingly popular idea is that since the capabilities of models can increase exponentially, with parameters, computing power, reasoning, and automation all on the rise, the macro - economic growth rate will also rise steeply. Some even envision that AI will enable developed economies to jump from an annual growth rate of around 2% to 10%, 20%, or even more astonishing figures.
This is both an over - worship of technological capabilities and a neglect of economics. The real world is not a single production function. An economy is composed of property rights, organizations, trust, responsibility, physical bodies, families, education, health, institutions, and time. If one part speeds up, the whole system may not follow suit; if a task becomes free, the final product may not be free; if a certain ability explodes, the macro - aggregate may not necessarily explode.
Therefore, one of the most important questions in the AI era is no longer "Can the machine do it?", but "Can the entire social system absorb, reorganize, price, and spread it?" Why do the macro - data still seem calm today when the micro - reorganization has been so drastic? Why does the economy seem to be held back by some "gravity" while technology seems to be on an exponential curve? And why, in the next thirty years, the sectors that will truly expand may be neither traditional commodity production nor narrow - sense digital production, but the reproduction of human beings themselves?
02. Piercing the Technological Utopia: The "2% Gravity" of the Macro - Economy
Before starting to write, we need to clearly define the object of discussion. The so - called "moderate growth" or "2% gravity" mentioned below refers to the leading economies at the forefront of world technology. Today, the most typical example is the United States. The growth of frontier economies must be achieved by forging new paths in uncharted territory, rather than following the well - trodden paths of others. As for those economies still in the process of catching up, they can experience high - speed growth for a considerable period because their growth comes from convergence and diffusion, rather than frontier advancement. That is the topic of another article, so we won't discuss it here.
Back to the frontier. There was the famous "Solow Paradox" in the computer era: computers were everywhere, but they were nowhere to be found in productivity statistics. Today's AI is at a similar crossroads. GPU clusters, foundation models, valuations of startups, capital expenditures, chip demand, and data center construction are all booming, but the macro - total factor productivity has not risen to the same level.
Much of the so - called AI prosperity at present first manifests as capital deepening. Enterprises buy more chips, build more data centers, hire more engineers, and incur more cloud costs. This naturally triggers an investment boom and reshapes the capital market's imagination of future cash flows. However, capital deepening does not equal a fundamental leap in total factor productivity. Although piling up more capital can increase local output, a real productivity revolution must be reflected in the ability to stably produce more value with the same amount of labor, capital, and organization.
From the steam engine, electricity, and internal combustion engine to computers and the Internet, the diffusion of general - purpose technologies has never been instantaneous. The emergence of a technology is just the first step, and there is a much longer journey ahead: processes need to be rewritten, institutions need to be adapted, human capital needs to be retrained, legal responsibilities need to be re - defined, industry standards need to be gradually formed, and consumer habits need to be changed little by little. In the initial stage, a new technology may even lower the observed productivity because the whole society has to spend a large amount of resources on trial - and - error, migration, and reorganization.
This is the "2% gravity" of the macro - economy.
The "2%" here is not a mysterious constant, but a historical experience: frontier economies find it difficult to break out of the moderate - growth track significantly and continuously over a long period. Technological revolutions can create local waves, reshape industrial landscapes, create large enterprises and wealth, and accelerate productivity in some years, but they rarely enable the entire frontier economy to break free from the real - world constraints formed by education, health, organization, law, family, cities, and physical bodies for a long time.
AI may indeed make the future growth rate slightly higher than in the past. If the US can increase its growth rate from 2% to 3% in the next thirty years, or even reach 4% in some periods, it will already be a very significant historical change. However, imagining such a change as a twenty - fold increase in the growth rate is to mistake the model - ability curve for the macro - economic curve.
Why is this so? An important reason is that Baumol's cost disease will reappear in a new form in the AI era.
When AI rapidly reduces the marginal cost of cognitive tasks such as text processing, logic, coding, retrieval, translation, calculation, and image generation, the sectors that cannot be completely replaced by algorithms will become more expensive. In - depth medical care, psychological companionship, personality shaping in education, child development, elderly care, organizational leadership, public responsibility, complex negotiations, aesthetic judgment, and trust endorsement are not just information processing. They involve physical presence, emotional resonance, social commitment, and responsibilities that must ultimately be borne by humans.
In a highly automated economy, what becomes truly expensive is no longer computing power, but human beings themselves. The relative prices of easily automated processes tend to fall, while those of processes that are difficult to automate and whose demand does not decline tend to rise. As a result, the expenditure share of high - friction sectors in the economy is passively increased, becoming a damper on the macro - growth rate.
This is not a failure of technology; it is just a general equilibrium.
03. The Translator's Paradox: The Fastest Machine, the Slowest System
Looking at translation and publishing from the perspective of the production function, the problem becomes clearer.
On the technological side, AI translation is almost a miracle. Today, it may only take a machine dozens of minutes to translate an English work of over a million words into Chinese, while in the past, an excellent translator might take two or three years. Just looking at the "text conversion" task, the efficiency improvement is not 20%, not 200%, but a thousand - fold or ten - thousand - fold.
However, in reality, we don't see the number of translated and published books exploding at the same rate. The number of high - quality books translated from English, French, German, and Japanese in the Chinese market has not increased a thousand - fold just because of AI.
The reason is that the publication of a book is never just the "translation" task, but a Leontief - type production process with fixed proportions. The final product requires many complementary links to be in place simultaneously: topic selection, copyright, contracts, translation, proofreading, editing, review, layout design, distribution, marketing, channels, reader positioning, academic endorsement, legal responsibility, and the assumption of market risks. AI has eliminated or significantly reduced the cost of one link, but not the others.
If a production system has fixed proportions, its efficiency is not determined by the fastest - running link, but by the slowest bottleneck. The faster AI translates, the more obvious the bottleneck becomes.
First, there are frictions in property rights and compliance. Transnational copyright negotiations won't be automatically completed just because the machine translates faster. Publishers still need to find the rights holders, negotiate prices, sign contracts, and handle issues such as the scope of authorization, electronic copyright, derivative rights, regional restrictions, and legal responsibilities. A text can be translated in half an hour, but a copyright contract may take half a year to negotiate.
Second, there is the issue of trust and quality control. What readers buy is not just a bunch of Chinese sentences, but a reliable text. Who will guarantee the accuracy of the translation? Who will handle the conversion of concepts, contexts, terms, cultures, and the author's style? Who will bear the reputation loss after an error? AI can generate a first draft, but the final quality still depends on the endorsement of experts, editors, and translators. The stronger the machine, the more expensive the responsibility borne by humans.
Third, there is the discovery of market risks. Whether a book is worth translating is never determined by the machine. Whether readers want it, whether the market can accommodate it, whether the distribution channels are willing to promote it, whether the critics will discuss it, and whether schools, the media, and knowledge communities will recognize it are all highly uncertain discovery processes. AI can reduce production costs, but it cannot eliminate the uncertainty of demand.
Fourth, there are constraints on attention and time. Even if all the books in the world are translated in an instant, readers won't have a thousand times more time for reading. The bottleneck of knowledge products often lies not in the production end, but in the receiving end. Human attention, patience, understanding, and mental structure are the ultimate scarce resources.
This is the so - called "translator's paradox": translation as a technical task has been greatly improved, but translation and publishing as a social production process have not boomed synchronously.
This example shows that the impact of AI on the economy cannot be judged solely by the efficiency of single - point tasks. What economics concerns are the final product, system complementarity, and general equilibrium. The marginal cost of a task reaching zero does not mean the marginal cost of an industry reaches zero. Machines can run extremely fast, but the social system still has to go through institutions, organizations, and people.
This is true for material production, digital production, and even more so for medical care, education, culture, law, finance, scientific research, and government governance.
04. The Transition in the Next Thirty Years: From the Reorganization of Commodities to the Reproduction of Humans Themselves
In the short term, the main impact of AI is not an explosion in the macro - aggregate, but a drastic reorganization of production factors.
The workflow within enterprises will be rewritten. Jobs will disappear and transform. Many jobs that previously relied on rule - following, text processing, standardized analysis, and exam - type cognition will quickly lose value. At the same time, new activities, new processes, and new occupations don't have stable names yet. This stage will surely be full of frictional costs: employees need to be retrained, organizations need to be re - divided, management needs to re - define responsibilities, laws need to be re - drawn, and consumers need to learn to trust again.
So in the short term, we may see a strange combination: the micro - level has changed dramatically, while the macro - level remains relatively calm. Everyone in the enterprise feels that AI has changed their work, but the productivity growth rate in the statistics is still not very impressive. This is not a contradiction, but a typical state in the transition period - the technological dividends are quietly offset by the reorganization costs.
In the medium term, the truly profound changes will occur in the human ability structure.
In the past few decades, modern education and the labor market have rewarded a rather special set of abilities: exam - taking ability, rule - following ability, text - understanding ability, standardized calculation ability, in - organization promotion ability, and stable execution ability. And it is precisely this set of abilities that AI first impacts. Any ability that can be clearly described, regularized, covered by training data, and simulated by language models will experience a decline in relative price.
So, what will become more valuable?
Health, physical strength, appearance, expression, charisma, leadership, risk - taking, aesthetic judgment, empathy, psychological resilience, sense of responsibility, trustworthiness, on - site presence, organizational mobilization ability, and cross - border creativity. These abilities, which were often classified as "non - cognitive abilities" or "soft skills" in economics in the past, will become more and more "hard" in the AI era. Because when machines can take over more and more cognitive tasks, the differences between people will increasingly lie in their bodies, emotions, characters, styles, trust, and organization.
From this, we can make a larger judgment: in the next thirty years, the largest and most core production sector in the whole society may be "improving human beings themselves".
The so - called "improving human beings themselves" is not narrow - sense education, nor is it traditional human - capital investment. It is not about training people to be supplements to machines, but about shaping people into more complete, energetic, trustworthy, and creative beings.
If we break down the reproduction of human beings themselves, there are at least three levels. The first is biological reproduction, including health, physical strength, nutrition, lifespan, chronic - disease management, and child development. The second is psychological and personality reproduction, including resilience, self - control, sense of responsibility, trust - building ability, and emotional stability. The third is the reproduction of social abilities, including expression, aesthetics, leadership, charisma, cooperation, and organizational mobilization. The more AI reduces the price of standardized cognitive tasks, the higher the shadow prices of these three types of abilities will be.
Therefore, early childhood development, family companionship, sports training, nutrition, medical care, mental health, aesthetic education, expression training, social skills, personality shaping, risk education, leadership training, elderly care, chronic - disease management, intimate relationships, community life, and spiritual order will increasingly resemble the core production activities of the future economy rather than marginal consumption.
In the industrial era, the largest production sector was material production; in the information era, digital and cognitive processing expanded rapidly; in the AI era, after ordinary cognitive tasks are largely automated, the area with the highest marginal return will return to human beings themselves.
This may sound like a humanistic slogan, but in fact, it is just an inference of price theory. Resources, efforts, and talents will be attracted to where a certain ability becomes relatively scarce and expensive. AI makes some cognitive labor cheaper, so the non - automatable, non - compressible, and non - replicable attributes of humans become more valuable. The focus of economic growth may gradually shift from "producing more things" to "cultivating better people".
05. The Reconstruction of Statistical Calibers and the Return of the Cost Method
If this judgment holds, the traditional GDP accounting system will face increasing pressure.
The current national economic accounting system is a product of the industrial era. It is good at recording commodities, services, wages, investments, and market transactions, but poor at recording human development within families, non - monetized experiences, the improvement of personal abilities, the improvement of health quality, the accumulation of psychological resilience, and the formation of social trust. Many truly important production activities either occur outside market transactions or are underestimated as consumption expenditures.
For example, a family spending a lot of time accompanying their children in reading, sports, expression, socializing, and exploration is almost invisible in the traditional GDP. However, from the perspective of the future economy, this may be the production of the most scarce assets. If a society can systematically improve the health, psychological resilience, expression, and creativity of its children, the future production capacity it creates will far exceed many short - term commodity transactions that are fully recorded.
More subtly, as AI reduces the marginal cost of traditional digital commodities, text products, and some services, the economic scale under the old accounting system may even be underestimated or distorted. A society's real welfare may have increased significantly, but the growth of monetary transactions is quite limited; it may also invest more and more resources in human development, but it is recorded as consumption rather than investment.
Therefore, in the next thirty years, the national economic accounting system must expand its scope to include human abilities. The development of education, health, psychology, family, care, sports, aesthetics, and personality