AI-Assisted Learning in High Schools: Evidence from a Nationwide Cluster Randomized Trial in Estonia

Kristjan-Julius Laak
Jaan Aru
Paul Yoo, Stanford University
Lily Fesler, Stanford University
Susanna Loeb, Stanford University
James Donovan
Date
Publication
OSF Preprints

This preregistration centers on evaluating the causal effects of an AI-based classroom-support intervention on students’ capacities, affects, and behaviors using a cluster randomized controlled trial. Randomized at the school level, treatment teachers are provided access to AI-supported learning modules that they can adapt and use for a topic they are covering in their class. Each module centers student activities around leveraging AI affordances to support learning (e.g., using AI to simulate experiments to test scientific mechanisms). The aim of the intervention is to induce students to use a customized AI tool in the context of integrating AI into schoolwork. The tool is a customized ChatGPT with specific system instructions grounded in learning sciences (e.g., to support autonomous motivation, see Other section). We measure outcomes using traditional survey instruments and task-based assessments and by deriving indicators of student behavior from student–AI interaction chat logs. Using natural language processing methods, we construct parallel measures of student capacities, affects, and behaviors from the text of students’ interactions with the AI system. A broader aim of the project is to assess the extent to which these chat log–based indicators align with and predict analogous constructs measured through traditional instruments.

Search Repository