AI-enabled personalized STEM education in K–12 classrooms significantly improves student learning outcomes, especially when supported by AR/VR technologies and collaborative, teacher-guided learning environments.

Objective

The study aimed to determine the overall effectiveness of AI-powered personalized learning in K–12 STEM education and to identify which conditions — such as grade level, subject, learning model, and technology type — produce the best results for students.

Methods

A comprehensive meta-analysis was conducted, reviewing randomized controlled trial studies published from 2012 to 2024. The researchers identified 32 qualified studies involving AI-enabled personalized STEM approaches and extracted 99 independent effect sizes. Data were analyzed using a random-effects model to determine overall impact, while subgroup analyses examined moderating variables such as school level, AI application type, subject area, and instructional design.

Key Findings

• Overall, AI-enabled personalized STEM learning has a statistically significant positive impact (small-to-medium effect).

• AI interventions were most effective in junior and senior high school, where students have higher self-regulation and cognitive maturity.

• AR and VR technologies produced the strongest learning gains, much higher than intelligent tutoring systems or AI-powered educational games.

• Hybrid learning models that blend AI with teacher involvement, collaboration, and project-based learning outperform fully automated systems.

• AI had the largest impact in interdisciplinary STEM courses, followed by mathematics and information technology subjects.

• Generative AI technologies (like ChatGPT) did not show stronger effects than earlier forms of AI in K–12 STEM contexts.

Implications

• Schools should invest in AI tools that support teachers, not replace them — especially tools that promote interactive and collaborative learning.

• Educational leaders should prioritize AR/VR and blended instructional models, as they provide the best outcomes.

• Policies must support teacher training and curriculum redesign to ensure technology aligns with pedagogy and student developmental needs.

• AI’s benefits are real but not transformational by default — implementation quality matters more than novelty.

Limitations

• Some STEM disciplines (like chemistry) had very few studies, limiting generalizability across subject areas.

• Only English and Chinese studies were included, leaving out research from many global regions.

• The wide variety of AI system designs makes it difficult to pinpoint which specific technological features drive success.

• Qualitative aspects — such as student motivation or teacher perspectives — were not extensively examined.

Future Directions

• Include more diverse STEM areas and age groups, especially early childhood.

• Examine how intervention duration, teacher readiness, and classroom context affect outcomes.

• Conduct more qualitative research on learner experience, equity concerns, and long-term impact.

• Compare sustained learning gains beyond the initial novelty effect of new technologies.

Title and Authors

A meta-analysis of AI-enabled personalized STEM education in schools
By: Shanshan Li, Chengze Zeng, Huaiya Liu, Jiyou Jia, Min Liang, Yinging Cha, Cher Ping Lim, and Xiaomeng Wu

Published On

2025 (Accepted August 22, 2025)

Published By

International Journal of STEM Education