Аңдатпа
Artificial intelligence can provide individualized movement information at a frequency that is difficult to achieve through teacher observation alone. This study examined whether AI-supported corrective feedback could strengthen adolescents' physical education self-efficacy while improving movement execution. Eighty-four students aged 13–15 completed an eight-week controlled experiment involving bodyweight squats, forward lunges, and countermovement jumps. The AI group received pose-estimation-based feedback generated with MediaPipe Pose Landmarker and OpenCV, whereas the control group received standardized teacher feedback. The AI group increased self-efficacy from 4.17 ± 0.62 to 5.39 ± 0.56 points, compared with 4.19 ± 0.60 to 4.66 ± 0.61 in the control group, producing a significant group × time interaction, F(1,82) = 42.31, p < .001, ηp² = .340. Movement scores increased from 64.1 ± 6.8 to 82.3 ± 5.7 in the AI group. Within the AI condition, movement-score improvement and feedback engagement jointly predicted post-intervention self-efficacy, adjusted R²=.487. The findings indicate that interpretable, attempt-specific AI feedback can make mastery progress more visible and support both technical learning and efficacy beliefs in adolescent physical education.