Artificial intelligence for inclusive education of neurodiverse learners: a systematic review of psychological and social factors

Objectives This systematic review examines psychological and social factors in AI-powered inclusive education for neurodiverse learners, including those with autism spectrum disorder (ASD), attention deficit hyperactivity disorder (ADHD), dyslexia, and other developmental variations.Methods Following PRISMA guidelines, we searched five databases (Scopus, Web of Science, IEEE Xplore, ACM Digital Library, and ERIC) for peer-reviewed studies published 2014–2025. From 871 initial records, 37 duplicates were removed and 456 excluded as ineligible, leaving 378 for title/abstract screening. Of these, 290 were excluded and 88 reports sought for retrieval; 32 could not be retrieved, leaving 56 full texts assessed for eligibility. Thirty-four were excluded and 6 after quality appraisal (threshold ≥ 8/10), yielding 16 included studies.Results Studies addressed eight psychological factors (attention and engagement, emotional recognition, communication processing, learning motivation, cognitive flexibility, behavioral regulation, memory processing, and anxiety regulation) and eight social factors (social communication, skills training, educational environment, peer relationships, family involvement, cultural context, professional support, and inclusive design). Findings suggest AI may improve social communication, emotional learning, and adaptive skills in some neurodiverse learners. However, responses varied between and within individuals, reflecting the heterogeneity of autism, ADHD, and dyslexia. Effect sizes were inconsistent and samples small, limiting generalizability.Conclusions This review examined a small and still-developing evidence base for AI to support individualized education recognizing neurodiversity as natural variation. However, geographic concentration (East Asia 43.8%, North America 31.3%), small sample sizes, and methodological heterogeneity limit generalizability. Future research should prioritize larger samples, longitudinal designs, and inclusion of underrepresented regions.