#364: Four Years of AI-Powered Formative Assessment: Exploring the 2-Sigma Effect and Fairness

Bloom’s (1984) two-sigma problem highlights the substantial learning gains achieved through individualized tutoring. With advances in artificial intelligence (AI), scalable personalized support may help approach such effectiveness while ensuring fairness across diverse learners. This study reports findings from a four-year deployment of an AI-powered formative feedback tool in a technical college English course. Two sequential studies were conducted: Study 1 synthesized data from seven semesters (Spring 2022-Spring 2024) to examine the longitudinal impact of AI feedback on learners’ concept learning and subsequent task performance. Study 2 used a quasi-experimental design (Fall 2024-Spring 2025) to compare learners with and without AI support, analyzing demographic differences in outcomes. Results showed that iterative AI feedback enhanced conceptual representation and improved later performance. Between-group comparisons favored AI-supported learners for concept learning, with no demographic disparities detected. Overall, AI-powered formative assessment produced learning gains approximating one sigma while maintaining equitable outcomes across learners.