Does AI help beginners catch up to experts, or let experts pull even further ahead?

This explores when AI narrows the distance between more- and less-skilled people, and when it pulls them further apart. The short answer from the corpus: on a single task, AI usually narrows the gap. Over a career, it can widen it, depending on whether people still get to practice. The strongest evidence for narrowing is striking. In a randomized experiment with over a thousand adults, AI cut the advantage of higher education on a business problem-solving task by about three-quarters, and less-educated participants kept part of that gain after the AI was taken away Can AI narrow the education performance gap?. At Procter & Gamble, individuals working with AI produced solutions as good as two-person teams working without it. Their proposals were also more balanced across functions, so AI partly filled in expertise a person didn't have Can generative AI replace the benefits of having a human teammate?.

One reason for this leveling effect comes from an unexpected part of the collection. Models trained on many imperfect experts tend to settle on the consensus answer, which averages out individual mistakes Can models trained on many imperfect experts outperform everyone?. If that's what AI hands you, it makes sense that it lifts people who are below the consensus the most, and adds less for people already above it. AI raises the floor. It doesn't obviously raise the ceiling.

The catch is what happens next. Wu et al. found that workers did much better with AI on content tasks, but when they later did similar tasks on their own, they hadn't improved at all Does AI assistance help workers learn lasting skills?. Better performance in the moment is not the same as building skill. That matters most early in a career. Interviews with South Korean software engineers describe AI absorbing the entry-level work juniors used to learn from, folding it into workflows run by seniors. Juniors lose the hands-on struggle that builds expertise, and the two groups don't even agree on whether this is a problem Does generative AI prevent juniors from getting entry-level work?. This is how a gap can widen even while every individual task gets easier: experts already have the judgment AI can't supply, and newcomers lose the route to developing it.

So the useful question isn't "upskilling or deskilling?" Interviews with Dutch knowledge workers found four outcomes, not two: skills can develop, be maintained, erode, or be revalued, depending on how each person uses the tool and which parts of their job change How does generative AI actually change worker skills?. A broad review across work, education, information access, and healthcare reaches the same conclusion at a larger scale. Whether AI reduces or increases inequality depends on access, how it is built into work, and incentives, not on the technology itself Does generative AI inevitably worsen or reduce inequality?.

What you might not have expected: the evidence that AI equalizes and the evidence that it divides aren't really in conflict. They measure different time spans. Short experiments capture the leveling effect on one task. Longer, interview-based studies capture what happens to how skills get built. If you want to tell which one is happening in a given workplace, ask whether people still do enough of the work themselves to learn from it, or whether AI does that part for them.