Does Artificial Intelligence-Powered Educational Agents Affect Students’ Learning? Evidence From a Meta-Analysis

• In Event: AERA Roundtable Session Sunday 1:45 pm - Gold 3
  In Roundtable Session: Meta-analytic and Systematic Reviews of Learning Environments

Abstract

This meta-analysis examined the effect of AI-powered educational agents on student learning outcomes using 41 effect sizes from 37 studies. The overall effect was small-to-medium (Hedges' g = 0.255, 95% CI [0.060, 0.450]). Significant variations across studies were found, with the largest effects for feedback-based interventions (g = 0.664) and immersive environments (g = 0.556). Longer interventions (10-15 weeks, g = 1.008) showed stronger effects compared to shorter ones (<1 week, g = 0.067). These results suggest AIEAs interventions positively impact learning, with notable differences based on intervention type, duration, and environment.

Authors

• Shuai LIU, East China Normal University

• Hao Lei, East China Normal University

• Zhuotao Lu, Qingdao University

• Xin Liu, East China Normal University