Teacher-mediated AI feedback in undergraduate piano teaching

Automated piano assessment can identify performance deviations, but teachers must translate these observations into practice tasks. This 16-week quasi-experimental study examined teacher-mediated AI feedback in 80 undergraduate music majors from two existing teaching groups, with one group per condition and 40 students per group. The same teacher taught both groups. During W1–W12, the experimental group received teacher-reviewed AI reports and passage-specific tasks in six biweekly cycles; the control group received conventional feedback. During W13–W16, scheduled AI-informed tasks ceased, while weekly lessons continued in both groups and platform access continued in the experimental group. Three independent experts, masked to group allocation, provided composite ratings at W0, W12, and W16. Across 240 paired records from 80 students, recorded platform composite scores correlated with the expert mean ( r = 0.972, student-cluster bootstrap 95% CI [0.962, 0.978]; ICC(A,1) = 0.971, 95% CI [0.961, 0.978]; MAE = 1.620). Dimension comparisons showed positive bias for technical accuracy (+ 1.032 points) and negative bias for expressive control (− 0.975 points). Model-based differences in growth favored the experimental group at W12 (6.57 points, 95% CI [5.19, 7.94]) and W16 (5.05 points, 95% CI [3.68, 6.43]). Practice time and teacher-estimated feedback uptake were higher, but the bootstrap confidence intervals for all exploratory indirect effects included zero. These preliminary associations describe the two teaching groups.