AI Must Enlarge the Economy, Not Divide It

By Mark Esposito and Aurélie Jean, Project Syndicate | July 31, 2026

To unlock AI’s full economic potential, we must steer it toward expansion. That requires replacing incentives that favor capital over labor with those that reward augmentation and investment in human capability.

CAMBRIDGE/PARIS—More than 200 economists and AI researchers, including 16 Nobel laureates, have signed the We Must Act Now statement, warning that AI could reshape the global economy on a scale larger than the Industrial Revolution—and far more quickly. We were among the signatories, but recognizing the stakes does not settle the question of how AI should be developed. One of the statement’s central claims is that AI should be steered toward complementing humans rather than imitating them. That distinction is more consequential than it may seem.

Complementing humans is not a matter of sentiment but a strategy for creating new markets. AI complements humans when it enlarges the set of things worth doing, broadens access by lowering costs, and enables people to do what was previously out of reach. It substitutes for them when it simply transfers certain tasks from workers to models. One path expands the economic pie; the other merely redistributes value. The purpose of AI is expansion, and policy should encourage it in that direction.

Much of the current AI debate rests on a familiar economic fallacy: the belief that there is a fixed amount of work to go around, that every task performed by a machine necessarily displaces a human worker, and the only question is how quickly technology destroys jobs. This zero-sum view is almost always wrong. The mechanical loom, the tractor, the personal computer, and much else were all expected to cause mass unemployment. Instead, by lowering production costs, they created more jobs than they eliminated.

Erik Brynjolfsson, one of the organizers of We Must Act Now, has called the AI-era version of this fallacy the “Turing Trap.” Designing AI to imitate humans, he argued, pushes technology toward substitution rather than augmentation, concentrates power in the hands of those who own the systems, and leaves ordinary workers with fewer opportunities to create value. While imitation divides the pie, augmentation makes it larger. Imitation may be the more obvious engineering target, but it often leads to worse economic outcomes for nearly everyone.

The process through which technology drives economic growth is well established, though it is often forgotten. When automation lowers the cost of producing something, demand typically increases, creating more demand for the human labor that machines cannot replace.

Consider the automated teller machine. As James Bessen has documented, ATMs reduced the number of bank tellers per branch from roughly 20 to 13. Because branches became cheaper to operate, banks opened many more of them. As a result, the total number of tellers continued to grow for decades, while the nature of the job shifted from counting cash to advising customers. In other words, a technology widely expected to eliminate a profession instead expanded it.

In a 2019 paper, Daron Acemoglu and Pascual Restrepo proposed a framework for understanding this dynamic. While automation shifts some tasks from workers to machines, they explained, those losses are offset by higher productivity and, more importantly, by the creation of entirely new tasks that require human skills. Many of the occupations that have fueled employment growth over the past several decades barely existed a generation earlier.

To be sure, there are no guarantees. Acemoglu and Restrepo also describe what they call “so-so automation”: technologies that displace workers without generating enough productivity gains or demand to create new opportunities, shrinking labor’s share of the economy.

AI has the potential to reshape the economy even more dramatically than previous waves of automation because, rather than merely automating individual tasks, it lowers the cost of cognition itself. When the cost of such a general-purpose input falls dramatically, markets do not simply consume more of the same services. They create new ones. Lowering the cost of legal, medical, pedagogical, and scientific expertise—domains where access was long constrained by the scarcity of skilled professionals—does not replace specialists. Rather, it enables them to reach vastly more people than was previously possible.

This is the real meaning of complementarity. A tool complements people when it increases the value of judgment, taste, accountability, relationships, and the other human capabilities AI does not provide. As raw cognitive capacity becomes abundant, those distinctly human inputs become more valuable, not less.

Of course, if these outcomes were inevitable, the statement we signed would be unnecessary. Whether AI replaces human work or complements it depends on the choices we make today. Replacing a worker is easy to price, finance, and explain, while building tools that enable people to accomplish things no one could before is harder, takes longer to pay off, and produces benefits that are more widely dispersed. Left to itself, capital tends to gravitate toward substitution because substitution is the simpler business case.

To unlock AI’s full economic potential, we must steer it toward expansion. That requires replacing incentives that favor capital over labor with those that reward augmentation and investment in human capability. Success should be measured not by how many workers AI replaces but by how much new demand and opportunity it creates. An AI system that merely automates existing work fails that test, even if it boosts short-term profits.

As we have argued before, the question is not whether AI will be powerful—it already is and will become more so—but whether we use that power to narrow the scope of human work or expand it into markets we cannot yet envision. Instead of automating tasks humans already perform well, we should empower them to accomplish previously unimaginable things and ensure that the resulting prosperity is widely shared.

Mark Esposito is Professor of Technology Policy at Northeastern University, Faculty Associate at Harvard University’s Berkman Klein Center for Internet & Society, and Chief Economist at micro1. He is co-author of Becoming AI Native (Routledge, 2026) and The Great Remobilization: Strategies and Designs for a Smarter Global Future (MIT Press, 2023).

Aurélie Jean is CEO and Co-Founder of Infra, an AI deep-tech startup in women’s preventive medicine.

Copyright Project Syndicate