Researchers Anand and Idan asked participants to teach themselves a technical subject. Half used books and other classic resources, half used an AI model. On average, AI helped, and some people gained much more than others.

Good grades and prior knowledge said little about who would gain. What made the difference was a skill the authors call AI interaction competence: asking the model good questions, sorting its answers and checking them. People strong in this skill gained a lot. The encouraging part: a simple aid, concept maps that help structure the work, narrowed the gap between the groups.¹

The study is still a preprint and awaits peer review. Its direction fits well with what field experiments on AI at work have found.

Talent Takeaway

It reminds us that working well with AI is a skill we can learn. An afternoon in which a team practices three steps together (ask, sort, check) can go a long way. A shared one-page template with good questions helps everyone keep pace.

¹ Anand, B., & Idan, L. (2026). Generative AI and the productivity divide: Human–AI complementarities in education and knowledge work [Preprint]. arXiv. https://doi.org/10.48550/arXiv.2605.18143

Further reading: Brynjolfsson, E., Li, D., & Raymond, L. (2025). Generative AI at work. The Quarterly Journal of Economics, 140(2), 889–942. https://doi.org/10.1093/qje/qjae044