The exact text we saved

This is the version exactly as it was saved, character for character. The SHA-256 hash below is a 64-character code worked out from that text. Run the same text through SHA-256 and you get the same code; change a single letter and the code comes out completely different. Each version also stores the previous version’s hash, which chains them together so no version can be swapped out quietly.

TRUVACE RECORD VERSION
record: TRV-2026-2782
version: 1
kind: certified
reason: Certified into the record
timestamp: 2026-11-12T17:23:18.643630Z
status: published
lens: g_space
sector: education
headline: The Impact of Artificial Intelligence-Supported Instruction on Student Learning in STEM: A Systematic Review and Meta-Analysis
dek: The rapid integration of artificial intelligence (AI) technologies into educational contexts has introduced innovative instructional approaches, particularly within Science, Technology, Engineering, and Mathematics (STEM) education. Although an increasing number of empirical studies have examined AI-supported instruction, existing findings remain heterogeneous, making it difficult to draw firm conclusions about its overall effectiveness. This study aims to systematically synthesize experimental and quasi-experim…
gain_title: A meta-analysis of 35 experimental and quasi-experimental studies found students taught with AI-supported instruction in STEM scored higher than peers taught traditionally (Hedges' g = 0.67, 95% CI 0.49 to 0.85), with the largest gain among high schoolers (g = 1.099).
problem_title: (none)
trace_subject: (none)
gain_reading: A meta-analysis of 35 experimental and quasi-experimental studies found students taught with AI-supported instruction in STEM scored higher than peers taught traditionally (Hedges' g = 0.67, 95% CI 0.49 to 0.85), with the largest gain among high schoolers (g = 1.099).
gain_evidence: statistically significant overall positive effect of AI-supported instruction on student learning outcomes in STEM education (g = 0.67, 95% CI [0.49, 0.85], p < 0.001) | the highest effect size was observed at the high school level (g = 1.099
problem_reading: (none)
problem_evidence: (none)
quick_read: A systematic review and meta-analysis published in the Journal of Intelligence in June 2026 pooled 35 experimental and quasi-experimental studies of AI-supported instruction in STEM education. Under a random-effects model the overall effect on student achievement was g = 0.67 (95% CI 0.49 to 0.85), a moderate to large improvement over traditional teaching.

The effect varied by school level. High school students gained most (g = 1.099), while the effect was smaller at elementary school (g = 0.465), middle school (g = 0.392) and university (g = 0.578). Interventions lasting more than one month and up to two months did best (g = 0.833), whereas those lasting one to seven days showed a weaker effect that was not statistically significant (g = 0.256).

Results were highly uneven between studies, with heterogeneity of about 83% in the fixed-effect analysis. Differences between science (g = 0.676), mathematics (g = 0.650) and technology/engineering (g = 0.501) were not statistically significant, so the authors say they should be read with caution.
limitation: Studies varied widely in their results (I² about 83%), much of the variance is unexplained, and most interventions were short to medium term, so long-term effects on learning are unknown.
tag: Evidence-backed gain
key_points: Searches of Web of Science, Scopus, ERIC, ScienceDirect and Google Scholar covering 2005 to 2025 turned up 450 records; 35 studies met the inclusion criteria. | Random-effects estimate: g = 0.670 (95% CI 0.491 to 0.848); fixed-effect estimate: g = 0.593. | High school students gained most (g = 1.099), versus g = 0.465 in elementary school, 0.392 in middle school and 0.578 at university. | Interventions lasting over one month and up to two months showed the largest effect (g = 0.833); five-hour-or-shorter programmes also had a significant effect (g = 0.621). | Interventions lasting one to seven days showed a weaker effect that was not significant (g = 0.256, p = 0.070). | Science (g = 0.676) and mathematics (g = 0.650) scored slightly above technology and engineering (g = 0.501), but the gap was not statistically significant (p = 0.088). | Publication-bias checks (Egger p = 0.095, Begg p = 0.0517, trim-and-fill with one imputed study) suggested little distortion; the fail-safe N was 2,404.
rundown: The authors searched English and Turkish sources in September 2025 and screened 364 records after removing 86 duplicates. They retrieved 43 full-text reports, excluded 8 for insufficient data or poor methodological quality, and analysed 35 studies, 29 of them published after 2020. Three researchers coded the studies, with inter-rater agreement of κ = 0.93.

Only one effect size per study was used, taken from the outcome closest to overall academic achievement, calculated from post-test means and standard deviations. The AI tools covered included intelligent tutoring systems, adaptive learning platforms, chatbots and simulations. Class size also mattered: effects were largest in medium classes of 50 to 99 students (g = 0.687) and smaller in classes of 49 or fewer (g = 0.448).

The authors read the results as conditional rather than universal: AI worked best within well-designed instruction, with teacher involvement and matched to learners' readiness. They call for studies of long-term retention, implementation quality and generative AI.
sources:
- peer_reviewed | Journal of Intelligence | https://doi.org/10.3390/jintelligence14060109 | 2026-06-15
prev: 0000000000000000000000000000000000000000000000000000000000000000

SHA-256 hash

3b27ca4db9ba3e053dc9cd621163447a23e3a995248f5d4378dcd08fc0e37450

Previous version’s hash

0000000000000000000000000000000000000000000000000000000000000000