Teaching Humans to Work Better with AI

Humans and AI can team up to solve problems faster than either could alone. But we still don’t fully know what makes that teamwork click. Researchers say the missing piece is how people learn when they get AI advice. That idea hasn’t gotten much attention in past studies.

A big review looked at 74 experiments that paired people with smart software. The review found that, on average, the mixed teams didn’t beat the best solo player. When the team looked closer, they saw that many studies ignored simple tricks that help people learn, like getting results after each try. Those studies that gave outcome feedback tended to show a small boost in teamwork.

Adding clear AI explanations helped only when people could check the advice with feedback. Without that chance to learn, the explanations actually hurt teamwork. So the key is letting humans test the AI’s suggestions and see if they’re right.

Future work should build experiments that change how much people learn, not just the type of task. Only then can we know if we can truly boost human‑AI collaboration. The field needs a new focus that puts learning front and center.