This report centers on the current causal evidence on learning outcomes because causal studies are the best way to learn how a tool (like AI) impacts students and educators. Across the limited set of causal evaluations available to date, the first outlines are starting to take shape. Current AI tools often improve student performance and foster positive experiences when students have access to them, but these gains can weaken or disappear when students are assessed without AI support. For educators, AI tools can save time as well as improve instructional quality. Given the narrow contexts and applications examined in high-quality research to date, these findings should be interpreted as reflecting the limited range of tool uses studied as much as the tools’ underlying effects.
This report begins by summarizing the characteristics of the over 800 papers in the Research Repository, which primarily consist of recent preprints (i.e., research papers publicly available but not yet peer reviewed by experts in the field). Understanding what researchers are studying provides insight into where the field is heading, including the topics prioritized, the methods employed, and the outcomes examined. The report then introduces the learning science principles that frame how AI could help or hinder learning opportunities. It then reports overarching takeaways about what is currently known about AI in K-12 education based on the 20 high-quality causal studies.
Overall, this report provides a synthesis of the current causal evidence on AI tools in education, highlighting emerging patterns as well as areas of uncertainty in the literature. In doing so, it aims to provide education leaders with a clearer evidence base for navigating decisions in a rapidly evolving landscape. However, readers should interpret these insights as preliminary and expect them to evolve as more evidence emerges. The field of AI in education is developing rapidly both in products available and insights learned from research. Given this dynamic landscape, many conclusions will require updating as additional rigorous research becomes available. Readers may find it useful to return to the Research Repository regularly as the evidence base develops.