Meta-analysis finds AI tools help university students learn across skill levels

The authors say earlier research on the question gave inconsistent results and that no one had systematically evaluated generative AI in higher education with a meta-analytic approach. Discipline moderated the effect on higher-order cognitive skills, so the authors recommend that course designers consider discipline characteristics and learning content.

A second analysis, association rule mining, found that generative AI positively influenced lower-order cognitive skills when in-class interventions lasted within a week. Small-group instruction in STEM disciplines looked promising for building higher-order cognitive and non-cognitive skills. The authors say the findings should be read with caution.

The research responds to inconsistent findings on generative AI in higher education. Using Bloom's Taxonomy as the frame, the authors pooled earlier studies to test whether the tools improve lower-order, higher-order and non-cognitive skills.

Moderator analysis showed that academic discipline shapes how generative AI affects higher-order cognitive skills. The authors conclude that educators should match AI applications and activities to the discipline and its learning content.

The association rule mining pointed to specific conditions. Short in-class interventions within a week aligned with gains in lower-order skills, and small-group teaching in STEM fields aligned with higher-order cognitive and non-cognitive gains. The authors say these results can inform policy and instructional approaches in higher education.