New study shows AI can cut learning by 17% or double it—depending on design

The paper 'The Effortless Trap' directly addresses the growing confusion among educators: should we allow or ban AI in the classroom? The authors argue this is a false dichotomy—the real question is placement. Using AI poorly replaces the cognitive work essential for learning, creating an illusion of mastery that collapses on unaided tasks. Strong causal evidence from a high school study shows the outcome flips entirely on design: an unguarded AI helper left students 17% worse on a later test compared to peers with no tool at all. The same model rebuilt to withhold answers erased the harm, and a well-engineered tutor roughly doubled learning outcomes.

To guide educators, the authors introduce a six-move model for learning any new concept: Prime, Probe, Point, Attach, Strengthen, and Test. The critical placement rule is to secure the first hard attempt and the final unaided check, using guarded AI only in between. A diagnostic sounds simple: if AI makes the task feel effortless, it's in the wrong place. The paper maps classical teaching moves and AI-supported interventions to each step, providing a practical foundation for lesson redesign that avoids the 'effortless trap' and preserves productive struggle.