Exploring the Effects of Generative AI on Inequality

What effect is generative artificial intelligence likely to have on wage inequality?

While no one can say for sure yet, it’s possible generative AI might reduce the contemporary societal problem of income inequality. That’s one of the conclusions reached by MIT Sloan School Associate Professor Nathan Wilmers in a new analysis.

Wilmer’s new working paper, “Generative AI and the Future of Inequality,” was recently published online as part of a new collection called An MIT Exploration of Generative AI. This collection consists of 25 papers on the societal impact of generative AI that were authored by members of the MIT faculty, with support from seed grants from MIT. Wilmers, who is the Sarofim Family Career Development Associate Professor and an Associate Professor of Work and Organization Studies at MIT Sloan, is a member of the core faculty of the MIT Institute for Work and Employment Research (IWER). He has done extensive past research identifying factors that drive labor market inequality.

In his new essay, Wilmers observes that generative AI could reduce the premium in wages received by college-educated knowledge workers—a contributor to inequality— because the technology may disrupt higher-paid white-collar jobs more than low-wage jobs. What’s more, he notes, early research suggests that, within a given job, large language models (LLM) benefit lower-skilled workers more than top performers—a result likely to compress wage scales over time.  “LLMs appear to be a skill leveler, delivering the biggest performance boosts to the previously worst performers,” Wilmers writes.

Should this occur, Wilmers points out that generative AI might over time undermine the rationale for the competitive meritocratic systems that sort students into colleges and jobs and increase income inequality along the way. “The meritocratic regime is only cost-effective if the productivity differences between slightly different candidates [for a job or for admission to a school] are large,” Wilmers writes—something that may become less true in the age of generative AI.

In addition, Wilmers makes the case that protections against displacement by generative AI use are the kind of workplace “collective goods” that motivate workers to seek union representation—suggesting that the technology could result in increased interest in unionization, including among white-collar workers.  And unions, Wilmers notes, not only help provide employment protection but also tend to reduce income disparities among their members through standardization of pay scales. As a result, he argues, “white-collar unionization that is primarily stimulated by generative AI…could strengthen inequality-reducing effects” of the technology by bringing about more standardized pay for white-collar workers.

On the other hand, Wilmers points out that there’s no guarantee that generative AI will reduce income inequality. Indeed, it could have the opposite effect. Wilmers cites a number of possible reasons. College-educated workers could end up competing with non-college-educated workers for blue-collar jobs. Some firms may prove far more adept than others at harnessing generative AI for competitive advantage, resulting in significant disparities between companies, and thus in the wages of their respective workforces. What’s more, some research suggests that managers who experience more precarity in their own work lives—something that technological changes like generative AI can engender—are in turn more likely to impose precarity on the people who work for them.  Finally, generative AI could result in more intensive workplace surveillance of workers that enables companies to tie wages closely to individual worker productivity.

While it’s impossible to predict precisely what effects generative AI will have on labor market inequality, Wilmers believes there’s a benefit to thinking through a range of possibilities. Identifying a number of the possible impacts large language AI models could have, he writes, his can help  “guide policymakers to broaden the consideration set of ways that LLMs can effect inequality.”