Does Shared Expertise Improve Learning? Evidence from AI Teacher Training

How to cite

Goulas, Sofoklis; Megalokonomou, Rigissa; Sotirakopoulos, Panagiotis. 2026. "Does Shared Expertise Improve Learning? Evidence from AI Teacher Training." AEA RCT Registry. AEARCTR-0018640. https://doi.org/10.1257/rct.18640-1.0

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Sofoklis Goulas* , Rigissa Megalokonomou , Panagiotis Sotirakopoulos

* Primary investigator

Registry ID
AEARCTR-0018640
DOI
https://doi.org/10.1257/rct.18640-1.0
First registered
Last updated
Status
In development
Trial dates
2026-05-15 – 2026-07-31
Countries
Private
Registry record
http://www.socialscienceregistry.org/trials/18640

Abstract

This study examines how peer-group composition shapes human capital accumulation during the adoption of artificial intelligence technologies. We implement a randomized field experiment within a large-scale AI teacher training program involving 750 in-service teachers.
Participants are randomly assigned into groups of five teachers during collaborative learning activities embedded throughout the training program. The experiment varies whether teachers are assigned to peer groups composed of educators who share a common field of expertise (STEM, non-STEM, or primary education) or to mixed groups containing teachers from different educational domains.
The study investigates whether learning is more effective when peer interactions occur within a shared domain of expertise or across domains with heterogeneous knowledge bases. Within-domain groups may facilitate communication, coordination, and transfer of domain-specific tacit knowledge, while cross-domain groups may generate broader knowledge spillovers and exposure to diverse pedagogical approaches.
The intervention is integrated into collaborative AI-related activities, including lesson design, prompt engineering, peer evaluation, and assessment of AI-generated educational materials. The study measures human capital accumulation using assessments of AI competency, productivity, AI oversight capabilities, and the quality of AI-assisted educational outputs.
The primary outcomes include post-training AI knowledge, task performance, and subsequent adoption of AI tools in educational practice. Secondary outcomes examine collaboration patterns, participation, confidence in AI use, and heterogeneity in peer-learning effects across teacher backgrounds.
This project contributes to the literature on peer effects, human capital accumulation, and knowledge transfer by examining how shared expertise and cross-domain interactions shape learning during technological change.

Experimental design

Participants in a large-scale online AI teacher training program are randomly assigned into collaborative peer groups of five teachers. The study includes approximately 750 in-service teachers organized into 150 groups.
The experiment varies peer-group composition. In the within-domain condition, teachers are assigned to groups in which all members share a common field of expertise (STEM, non-STEM, or primary education). In the cross-domain condition (mixed condition), teachers are assigned to groups containing teachers from different educational domains.
The intervention is embedded within collaborative components of the training program, including group discussions, peer-feedback exercises, AI-assisted lesson-planning tasks, prompt-engineering activities, and evaluation of AI-generated educational materials.
Participants complete baseline and endline assessments measuring AI-related knowledge, productivity, AI oversight capabilities, and attitudes toward AI use in education. The study also collects process data from the training platform, including participation, collaboration patterns, and interaction with AI tools.
The primary objective is to examine whether human capital accumulation during AI adoption differs between peer groups characterized by shared expertise and those characterized by heterogeneous expertise and cross-domain interaction.

Design details

Randomization method
Randomization by a computer
Randomization unit
Teacher
Clusters
150 groups
Observations
750 teachers
Arms
-Mixed (cross-domain condition): 190 Within-domain conditions: -Non-STEM: 270 -Primary: 225 -STEM: 65
Primary outcomes
The primary outcomes include post-training AI knowledge, task performance, and subsequent adoption of AI tools in educational practice. Secondary outcomes examine collaboration patterns, participation, confidence in AI use, and heterogeneity in peer-learning effects across teacher backgrounds.

Cited by

No citation data available for this registration yet.

Keywords