Economists Breaking Ranks on AI’s Transformative Risk

Technologists often predict that their creations will have unprecedented economic impacts, especially when it comes to artificial intelligence (AI). Elon Musk has said that continued advances in AI would make human labor obsolete. Sam Altman, CEO of OpenAI, has said that AI will continue to shift economic power from labor to capital and create “phenomenal wealth.” Jensen Huang, CEO of semiconductor design firm Nvidia, has compared AI’s development and deployment to a “new industrial revolution.”

However, economists are not as bullish on the economic impacts of AI. Even if technologists create the powerful AI systems that they claim they soon will, the economic impacts of those systems are likely to be underwhelming, say many economists. According to Tyler Cowen, an economics professor at George Mason University, AI will “boost the annual US growth rate by one-quarter to one-half of a percentage point.” Nobel Prize-winning economist Paul Krugman has said that “history suggests that large economic effects from A.I. will take longer to materialize than many people currently seem to expect.” David Autor, professor of economics at the Massachusetts Institute of Technology, has said that the “industrialized world is awash in jobs, and it’s going to stay that way.”

Anton Korinek, an economics professor at the University of Virginia, is a certified member of the economics profession. However, he has broken rank with his colleagues. He uses the methods of his discipline to model what would happen if AI developed as many technologists say it might—which is to say that AI systems are likely to outperform humans at any task by the end of the decade. In the sterile lexicon of academic economics, he paints an alarming picture of a near future in which humans cease to play a role in the economy, and inequality soars.

In a recent paper, Korinek and one of his Ph.D. students, Donghyun Suh, write that “if the complexity of tasks that humans can perform is bounded and full automation is reached, then wages collapse.” In other words, if there is a limit to how complex the tasks humans can do are, and machines become advanced enough to fully automate all those tasks, then wages could drastically fall as human labor is no longer needed, leaving anyone who doesn’t share in the resources that accrue to AI systems and their owners to starve.

Economists have spent the last 200 years explaining to the uninitiated that the lump of labor fallacy—the idea that there is a fixed number of jobs and automation of any of those jobs will create permanent unemployment—is wrong, says Korinek. “In some ways, that’s our professional baggage,” he says. “We’ve spent so much time fighting a false narrative, that it’s difficult to pivot when the facts really do change, and see that this situation may indeed be different.”

“The scariest part about it is that the technological predictions that I use as inputs are what people like Sam Altman or [Anthropic CEO] Dario Amodei are freely preaching in public everywhere,” says Korinek. “I’m not a technology expert, and I don’t have any special insight or knowledge to these issues. I’m just taking what they’re saying seriously and asking: what would it actually mean for the economy, and for jobs, and for wages? It would be completely disruptive.”

Korinek’s father, a primary care physician, instilled in him a passion for understanding the brain. As a teen in Austria, Korinek became fascinated with neuroscience and learnt to program. At college, he took a course on neural networks—the brain-inspired AI systems that have since become the most popular and powerful in the industry. “It was intriguing, but at the same time, boring, because the technology just wasn’t very powerful using the tools that we had back then,” he recalls.

As Korinek was finishing college in the late 1990s, the Asian financial crisis hit. The 1997 crash, precipitated by a foreign exchange crisis in Thailand, followed many other episodes of acute financial distress in emerging markets. Korinek decided to do a Ph.D. on financial crises and how to prevent them. Fortunately for him—although less fortunately for everyone else—the global financial crisis struck in 2008, a year after he finished his thesis. His expertise was in demand, and his career took off.

As an academic economist, he watched from the sidelines as AI researchers made breakthrough after breakthrough; scanning newspapers, listening to radio programs, and reading books on AI. In the early 2010s, as AI systems began to outdo humans at tasks such as image recognition, he began to believe that artificial general intelligence (AGI)—the term used to describe a yet to be built AI system that could do any task a human could—might be developed alarmingly soon. But it took the birth of his first child in 2015 to get him off the sidelines. “There is really something happening, and it’s happening faster than I thought in the 1990s,” he recalls thinking. “This is really going to be relevant for my daughter’s life path.”

Using the methods of economics, Korinek has sought to understand how future AI systems might affect economic growth, wages, and employment, how the inequality created by AI could imperil democracy, and how policymakers should respond to economic issues posed by AI. For years, working on the economics of AGI was a fringe pursuit. Economics Ph.D. students would tell Korinek that they would like to work on the questions he was beginning to study, but they felt they needed to stick to more mainstream topics if they wanted to get a job. Despite having been awarded tenure in 2018 and thus being insulated from the pressures of the academic job market, Korinek felt discouraged at times. “Researchers are still also social creatures,” he says. “If you only face skepticism towards your work, it makes it much, much harder to push forward a research agenda, because you start questioning yourself.”

Up until OpenAI released the wildly popular chatbot, ChatGPT, in November 2022, Korinek says research on the economics of AGI was a fringe pursuit. Now, it’s “on an exponential growth path,” says Korinek.

Many economists still, quite reasonably, dismiss the possibility of AGI being developed in the coming decades. Skeptics point to the many ways in which current systems fall short of human abilities, the potential roadblocks to continued AI progress such as a shortfall in data to train larger models, and the history of technologists making overconfident predictions. A paper published in February by economists in Portugal and Germany sets aside AGI as “science fiction,” and thus argues that AI is unlikely to cause explosive economic growth.

“Two years ago, that was absolutely commonplace—that was the median reaction,” says Korinek. But more people are starting to entertain the possibility of AGI being developed, and those who continue to dismiss the possibility are providing solid arguments for their position rather than dismissing it out of hand. “People are, in other words, engaging very seriously in the debate,” he says approvingly.

Other economists are open to the possibility of AGI being developed in the foreseeable future, but argue that this still wouldn’t precipitate a collapse in employment. Often, they put the disagreement with those who do think the development of AGI might cause these things down to the other side’s economic illiteracy.

For example, in March, economist Noah Smith argued in a blog post that was picked up by the New York Times that well-paid jobs will be plentiful, even after AI can outperform humans at any task. This is, Smith forbearingly explains, because AI systems and humans can each specialize in the tasks where their relative advantage is greatest—an economic idea known as comparative advantage—and then trade with each other. When “most people hear the term “comparative advantage” for the first time, they immediately think of the wrong thing,” writes Smith.

But Korinek is hardly economically illiterate. “I claim to say I do understand comparative advantage,” he says with a grin. Korine