Study: AI help lifts homework scores but drags down exam results

A study reported by The Economist found a split between two measures of student performance: when children used AI assistance on homework, their homework scores rose, but their scores on exams — taken without AI — fell. The pattern points to a gap between assisted output and retained understanding. Work that looks stronger with a tool in hand does not necessarily reflect what a student has actually learned.

The finding matters for how schools evaluate and deploy AI. If homework scores improve while exam performance declines, the tool may be substituting for the effortful practice that builds durable skills rather than reinforcing it — a well-known effect where offloading cognitive work weakens recall and transfer. It suggests homework grades become a misleading signal of mastery once AI is in the loop.

Note: only the headline was available in the source provided here. The underlying study’s specifics — sample size, ages, subject, effect magnitudes, and methodology — are not included and would need to be pulled from the full Economist piece to summarize responsibly.