AI Versus Human Feedback in Gamified Color Education: A Four-Arm Cluster Randomized Controlled Trial

Empirical evidence for AI formative assessment combined with gamification remains scarce in subjective aesthetic disciplines. This study examined whether such integration enhances metacognitive regulation in undergraduate color education and whether the effect is conditional on learners’ prior AI credibility belief. A four-arm, four-wave cluster-randomized controlled trial (N = 540, 24 classes, eight cross-allocated teachers) compared gamification with AI feedback, gamification with dose-matched human feedback, gamification only, and traditional instruction; the primary outcome was metacognitive regulation measured 4 weeks post-intervention. Gamification produced a small-to-medium effect over traditional instruction (d = 0.34, p_Holm = 0.015). AI and human feedback did not differ on average (d = 0.07; underpowered). Autoregressive mediation confirmed metacognitive regulation as a longitudinal mediator of intervention effects on motivation and achievement. On an exploratory basis, prior AI credibility belief moderated the AI-versus-human indirect effect on motivation (p_Holm = 0.034) but not achievement (p_Holm = 0.054), with higher-belief learners benefiting; given the limited power of the AI-versus-human comparison, this conditional pattern is reported as a tentative finding rather than as confirmatory evidence. These findings extend self-determination theory and the self-regulated learning framework to aesthetic education and provide preliminary support for the deployment of prior-trust-aware AI feedback, pending replication.