Maybe AI Isn’t Coming for Everyone’s Job After All

New studies suggest that, like most technology, AI makes people more productive, more valuable employees.

Good luck tracking the zig-zags of academic research on artificial intelligence and the future of work. A University of Pennsylvania/OpenAI study earlier this year found that 80 percent of jobs could see some of their tasks automated through generative AI and around 20 percent could have half or more of their workload fully automated. The same study rang other alarm bells by pointing out how white-collar professionals, rather than lower-skilled workers, were most at risk of AI-driven automation.

Several weeks ago, a new study published by the National Bureau of Economic Research of the actual impacts of AI on European workers zagged the other direction. Looking at data from 2011 to 2019, researchers from the University of Pittsburg, Oxford, and several European central banks found that greater exposure to AI increased employment shares, defined as the percentage of total employment that a particular group accounts for in the overall workforce, and had no significant effect on wages.

The authors plot growth in employment against exposure to AI, using two previously developed measures of exposure. Jobs with low exposure to AI, such as food preparation assistant, are at the bottom end of this distribution, while jobs with high exposure to AI, like insurance actuary, sit at the top. They find that moving from the 25th to the 75th percentile of this distribution—moving away from food preparation assistants and towards actuaries—is associated with 2.6 to 4.3 percent increases in employment shares, depending on the measure used to track exposure. These increases are modest but statistically significant, regardless of which measure the researchers used.

As the authors explain, these findings are consistent with “skill-biased technological change” theory, which holds that technological change and automation increase demand for high-skilled labor. The important lesson from the study is that skill development and human capital accumulation at the top of the skill distribution “continue to be the way to accommodate new technologies without employment losses.” In other words, the best insurance against unemployment is higher levels of education, training, and skills—not, say, a “pause” on AI development. As an added benefit, the authors find these gains don’t come at the price of lower wages—at any skill level.

These findings, if they hold up, would be very good news. While the authors don’t find that AI has more than a “mildly positive” impact on employment levels, neither do they forecast an employment Armageddon for workers. As with technological change more broadly, this study suggest AI will increase aggregate economic demand and stimulate more growth and jobs over time. In June, tech entrepreneur and venture capitalist Marc Andreessen put it this way:

Technology empowers people to be more productive . . . This in turn causes economic growth and job growth, while motivating the creation of new jobs and new industries. If a market economy is allowed to function normally and if technology is allowed to be introduced freely, this is a perpetual upward cycle that never ends.

One potential limitation of the study from Europe is that it doesn’t fully account for the recent leaps in generative AI, which the authors acknowledge is still in its early stages. It is worth noting, however, that several AI breakthroughs during the study period in fields like robotics, supervised and unsupervised learning, natural language processing, image recognition, and performing non-routine tasks (such as giving medical advice or writing code) included nascent forms of generative AI technology. The difference, therefore, is a question of the degree of technological penetration, not the type.

It is still possible that AI will have far-reaching effects that create considerable labor market turbulence as workers adapt to changes in skill demands and new jobs. Indeed, we can also expect many more studies estimating the long-term economic and employment effects of AI as the technology morphs into new, unforeseen configurations and applications. While we wait for that future to unfold, the best advice remains the same for students and workers: develop and maintain broad, flexible knowledge and skills that support learning and adaptation. It matters less what particular capacities and knowledge one has right now than whether one is ready to learn and adapt in the future.