Stanford’s SCALE (Systems Change Advancing Learning and Equity) Initiative has quietly been doing some of the most valuable work in AI and education: reviewing hundreds and hundreds of AI education research papers to amplify the ones that show real causality and quality.
Back in March, their first Evidence Base Review found that only 20 papers (out of 800+) cleared their bar, so the update they released this week is a real leap: the number of high-quality causal studies has more than doubled, bringing the total to 52. Moreover, the vast majority of the new studies are focused on student-facing AI, just at a moment when the entire Edtech field is working to discover what works for students.
Just as important as the numbers is how carefully SCALE reports them. They’re intentionally not trying to collapse the field into one magic effect size via meta-analysis, an approach that has been largely unsuccessful:
- A widely shared meta-analysis claiming ChatGPT had a “large positive effect” on learning was retracted this summer
- Researchers auditing another found that some of its most-cited studies measured how students performed while using ChatGPT rather than what they learned afterward
- A sweeping analysis of 67 AI-in-education meta-analyses found severe publication bias across the board
- 39 meta-analyses, 39 positive findings: An audit finds serious flaws in research on AI in education
Instead, SCALE takes the opposite approach, with narrow claims, honest caveats, and a willingness to say “we can’t tell yet” when the evidence isn’t there.
