Mixed learning groups broaden teachers’ AI use
Sofoklis Goulas, Rigissa Megalokonomou, Panagiotis Sotirakopoulos · Goulas, Megalokonomou & Sotirakopoulos · SSRN · 13 Oct 2026
Source finding
Seven hundred and fifty teachers were randomly assigned to five-person groups with the same or mixed specialties. With the same course, instructor and AI access, mixed groups later reported AI use in teaching that was 0.19 standard deviations broader. The increase was driven mainly by creating grading rubrics.
Our interpretation
Training only people with similar jobs together may hide practical applications. The finding concerns the spread of reported use, not result quality or business value.
Method and evidence
Preregistered randomized experiment in a nationwide AI training programme in Greece. Seven hundred and fifty practising teachers were assigned to five-person groups: primary, secondary STEM and secondary non-STEM, either mixed or same-specialty. Curriculum, instructor and tool access were held constant. Good evidence for the short-term effect of group composition.
Limitations and interests
AI use was self-reported at the end of the core curriculum. Objective use, persistence, teaching quality, student outcomes and productivity were not established. Greek education setting; transfer to companies and other occupations is open. Working paper, not peer-reviewed. No conflicts declared; support from a Monash University grant.
What businesses can do
Mix specialties in AI training. Then assess specific new uses and their quality; the study measures reported use, not teaching quality or productivity.
Where to start
Abstract and methods for group assignment, identical training and post-course applications. Original paper on SSRN