Careers & University / Big Questions / AI and work
From the research interviews we hold · updated 09 Nov 2026
The research here disagrees: some findings show AI narrowing gaps by helping less experienced workers, while others show automation and robots widening gaps and hurting lower-skilled jobs. Whether AI makes work more unequal depends heavily on which technology, which workers, and which local economy you look at.
Word for word, from interviews with the researchers. Each opens the moment they say it.
Generative AI produces an equalizing effect by providing greater time savings for less experienced workers than for more experienced workers.Donghyun Suh, Bank of Korea — Rapid Adoption of Artificial Intelligence and Its Productivity Effects: The Case of Korea · 7:39Paper
Automation is unequalizing both between capital owners and firms and labor and within labor, harming lower skill and middle skill workers.Daron Acemoglu, Massachusetts Institute of Technology — Daron Acemoglu on Robots and Jobs · 3:52
In local economies with more robots being adopted, employment and wages fall, and production blue-collar occupations disappear as industrial robots take over their tasks.Daron Acemoglu, Massachusetts Institute of Technology — Daron Acemoglu on Robots and Jobs · 1:11
Automation can push workers, especially skilled ones, out of the labor market, as these workers typically perform highly repetitive tasks.Rosa Aisa, University of Zaragoza — How Tech and Demographics Reshape Work · 3:34
Uncertainties likely have large differential effects, causing some types of workers to lose relative to others, even though overall employment figures do not show this.Andrea Caggese, Universidad Pompeu Fabra — How Economic Uncertainty Reshapes Hiring and Firing in Firms · 13:55Paper
The key challenge for workers in 2026 is understanding both the capabilities and limitations of AI, and prior expertise enables one to evaluate AI-generated answers by recognizing when a small portion of the response is inaccurate.Mark Shapiro, University of Miami — Professional Credibility, Early-Career Performance, and Workplace Reputation in Business · 12:04Paper
The question sounds simple, but the research splits depending on what kind of AI you mean and who you measure. On the equalizing side, [1] finds that generative AI saves more time for less experienced workers than for experienced ones, which would compress gaps rather than widen them. That is a hopeful signal: if a junior worker can do more of what a senior worker does, the ladder gets shorter. But [3] argues the opposite for automation broadly, saying it hurts lower- and middle-skill workers and shifts power toward whoever owns the machines. [7] backs this up at the local level: places that adopted more robots saw employment and wages fall and blue-collar production jobs disappear. [9] adds a twist, finding that automation can push skilled workers out too, because even highly trained people often do repetitive tasks that machines do well. So the same word, AI, covers tools with very different effects.
Why do researchers disagree? First, they study different technologies. Generative AI, the chatbot kind, assists people at desks; industrial robots replace people on factory floors. [6] notes that generative AI already touches over half of workers while physical AI reaches only about 11%, so the two are at different stages. Second, they measure different things. [21] points out that overall employment figures can look fine while some groups lose relative to others, so averages hide the inequality. Third, context matters: [23] stresses that workers who understand both what AI can and cannot do are better placed to catch its errors, which suggests the advantage may go to those with prior expertise, not beginners. That cuts against the equalizing story in [1].
There is also a design question. [26] warns that AI systems are often calibrated to be harsher than humans, and [13] notes that AI can repeat or amplify bias from unrepresentative data. If that happens, the technology may entrench inequality even when it saves time. Meanwhile [14] suggests AI agents can be built to supplement rather than replace people, which is a choice, not a fixed outcome. Put together, the evidence does not say AI inevitably makes work more unequal. It says the direction depends on whether the tool assists or replaces, who gets training, and whether the gains are shared. For someone choosing a path, that means the specific job and the specific technology matter more than the headline about AI.