When adult learners alternate between solo AI-supported quizzes (with dynamic hint gating) and small-group problem-solving sessions that use an always-on AI assistant, does misalignment between the two hint policies (e.g., strict gating solo, unrestricted AI use in groups) increase illusions of learning and overconfidence compared with a consistent cross‑context policy, and is this effect stronger for low‑prior‑knowledge learners?

Answer

Misalignment between strict, dynamically gated hints in solo AI quizzes and unrestricted AI use in small-group sessions is likely to increase illusions of learning and overconfidence relative to a more consistent cross-context policy, with stronger negative effects for low-prior-knowledge learners. The inconsistency encourages learners—especially in groups where AI is always-on and often used by a dominant solver—to interpret the ease and apparent success in the group context as evidence of personal mastery, discounting the constraints they experience in solo quizzes. A more aligned policy (e.g., requiring unguided attempts and some form of gating or structure before AI use in both contexts) should produce better metacognitive calibration by making the conditions for success more similar and by preserving productive struggle and individual retrieval practice across settings. However, the exact magnitude of the misalignment effect remains theoretical and would need empirical confirmation.