Empowerment or Disempowerment? A Meta-Analysis of Intelligent Applications on Academic Achievement in Autism Spectrum Disorder

Abstract

Artificial intelligence in education (AIED) applications have emerged as a promising solution to address the unique educational challenges faced by students with autism spectrum disorder (ASD). However, there is a lack of comprehensive meta-analyses examining the effectiveness of AIED applications on ASD students’ academic achievement and the factors that moderate its efficacy. This study bridges the existing research gap through a comprehensive meta-analytic methodology to quantitatively synthesize 45 experimental and quasi-experimental studies, systematically examining both the overall impact of AIED applications on ASD students’ academic achievement and the moderating factors affecting this relationship. The meta-analytic results demonstrated a moderately small overall effect size (ES = 0.48), with significant heterogeneity observed across different academic domains and AIED application modalities. Moderator analysis further revealed four significant predictors of intervention efficacy: (a) greater impact on ASD students over 18 years old; (b) stronger effects when interventions were implemented directly with ASD students; (c) increased efficacy in procedural and factual task types compared to conceptual or metacognitive tasks; and (d) more substantial gains in experimental durations shorter than three months. The study concludes by discussing the implications of these findings and suggesting practical directions for future research to further develop AIED applications for ASD learners.