Will AI Democratize Skills?
By María Lombardi
Special to The Times Kuwait
Recent AI advances have created widespread expectations of substantial productivity gains. Early studies, such as one showing that AI increased the productivity of customer-support agents by 15% on average (with less-experienced workers getting a much bigger boost), as well as emerging evidence of AI-driven productivity gains in macro data, have further elevated hopes for a boom in output per worker.
As with past innovations, a key question is how the gains from productivity growth will be distributed. Historically, technological advances have disrupted labor markets and often widened gaps in earnings and employment between individuals according to their educational attainment.
More than half of the overall changes in the US wage structure over the last four decades can be attributed to a relative decline in wages for blue-collar workers in manufacturing and clerical jobs whose routine tasks could be automated.
Many fear that AI will increase inequality. But the fact that, unlike previous technologies, these systems can perform complex cognitive tasks raises the possibility that AI could democratize skills. In other words, AI could enable workers with limited formal training to perform tasks that previously required extensive education, thereby narrowing the gap between workers.
In a recent study, my co-authors and I examined AI’s effect on individuals with a high-school diploma compared to those with a postsecondary degree. Participants completed a task based on a realistic hypothetical business scenario: they had to respond to an email from their boss by analyzing several sources of information. Half of them (randomly selected) had access to a virtual AI assistant embedded in the platform, while the other half did not.
We found that AI could reduce inequality between workers of different educational levels. While access to AI improved both groups’ performance, the effect was much greater for the less-educated participants. While the latter performed significantly worse than higher-educated participants in the absence of any technical support, access to AI closed 75% of the initial performance gap, with gains in both content and writing quality.
To understand the remaining gap, we analyzed participants’ interactions with the AI assistant and found qualitative differences in engagement. Higher-educated users were more likely to give structured prompts and provide specific instructions aimed at guiding the assistant through thinking over the task and generating the final output. Moreover, because many participants combined AI-generated text with their own contributions, quality still seemed to reflect education-driven productivity differences.
To determine whether using the AI assistant affected participants’ ability to articulate and justify their reasoning once it was no longer available, they were asked follow-up questions immediately after completing the task, with no access to the tool.
There was no evidence that prior AI use harmed subsequent performance for either group. In fact, some of the gains from the AI assistant carried over for the participants with no postsecondary education, a finding consistent with genuine engagement with the problem rather than pure task delegation.
To be sure, generative AI does not eliminate the role of human capital in performance, nor does it equalize fundamental abilities. Instead, we find that it relaxes execution constraints that are more binding for individuals with less formal education.
Whether AI ends up narrowing or widening inequality will depend less on the technology itself than on the companies and institutions shaping its adoption and use. Worryingly, recent evidence suggests that AI usage is already more common among more educated workers. Emerging corporate practices are reinforcing this trend, as major tech firms move beyond encouraging AI adoption to mandating it—and even factoring its use into performance reviews.
Since the push for AI adoption is mainly concentrated in highly skilled sectors like tech, it risks deepening existing disparities rather than broadening access to the workers who stand to benefit from it the most.
Compounding this problem is the finding that AI may be diminishing hiring for entry-level positions, which less-educated workers typically use to get their foot on the employment ladder—raising the possibility of that ladder being pulled up even as these tools improve and increasingly close the performance gap.
Businesses, schools, and governments can significantly expand opportunity by investing in AI training, broadening access to these tools, and devising policies that help less-educated workers use them productively, increasing the range of tasks they can perform well.
Policymakers must also ensure, through appropriate rules and incentives, that AI complements, rather than replaces, workers. If access, know-how, and organizational support remain concentrated among the already advantaged, the benefits of AI may be captured in ways that reproduce the inequalities associated with past technological change.
María Lombardi is Academic Dean and an associate professor at the School of Government at Universidad Torcuato Di Tella.
Copyright: Project Syndicate, 2026.
www.project-syndicate.org
