Review of 188 studies finds explainable AI tied to better learning measures in college

The authors searched Scopus, Web of Science and PsycINFO for work from 1 January 2015 to 30 October 2025. They started with 3,426 records and sent 616 studies to full-text review. The findings fall into three clusters: cognitive load and neural efficiency, attention and working memory, and metacognition and affective interaction.

Pooled correlations ranged from 0.27 to 0.36, with moderate heterogeneity (I² about 58%). A correlation of this size is modest and describes an association, so it does not show that explainable AI causes the learning differences.

The authors examined explainable AI in cognitive psychology within higher education, focusing on adaptive learning systems. They searched three databases covering 1 January 2015 to 30 October 2025 and followed PRISMA 2020 reporting guidelines.

From 3,426 initial records, 1,936 papers were excluded and 616 studies went to full-text review. The final set of 188 studies covered 62,000 participants from higher education institutions around the world.

A science mapping analysis found three clusters: cognitive load and neural efficiency (C = 0.82, D = 0.76), attention and working memory (C = 0.64, D = 0.58), and metacognition and affective interaction (C = 0.42, D = 0.39). Pooled correlations in these clusters were positive, from r = 0.27 to 0.36.

The Bayesian psychometric model put the overall effect at μ = 0.33, about r = 0.32 (95% CrI 0.27 to 0.38). Heterogeneity was moderate at I² of about 58%, and the authors concluded that explainable AI showed validity within cognitive psychology.