Enhancing Learning Outcomes through AI-Based Tutoring Systems: A Study on Student Motivation and Academic Achievement

Abstract

Purpose: To determine whether an artificial intelligence (AI)-based tutoring system (AITS) is more effective in terms of academic success and motivation, as well as to investigate causative influences of motivation. Techniques: It was a pre-registered randomised trial in 24 classes (N=602; Grade 7-10), with assignment to AITS or business-as-usual either at the student or class level. The intervention provided adaptive sequencing, stepwise feedback, mastery thresholds, and spaced review in 8-12 weeks. The outcome measures included Post-test achievement that was curriculum-based; Intrinsic Motivation Inventory and MSLQ subscales were the secondary outcome measures.

The ANCOVA and multiple imputation linear mixed models were analysed and then multilevel mediation and moderation followed. Findings: AITS brought about a 5.1-point (d[?]0.40; p<.001) posttest-controlling effect. Interest/enjoyment and perceived competence went up (d=.20-.45). The achievement effect was mediated by interest ≈24%. The effects were greater in students with lower baseline scores and rose with usage (to approximately 12 hours) after which it was diminishing in returns. Results were strong to clustering correction, different scaling and sensitivity tests. Conclusions: Under normal classroom time, AITS has the potential to improve performance through the improvement of motivational states and effective engagement, especially with occurrence in lower-baselin learners. It should be implemented to focus on minimum viable dosage, step-level feedback of high quality, and analytics-informed coaching. The future work must also be cross-subject and cross-term, include retention results, and have equity and cost-effectiveness audits. All materials will be shared.