AI Comes to Pre-K: A New Multi-Modal, Dynamic AI System to Identify Malleable Features of Classroom Instruction

Prior projects in education primarily have examined how to automate the coding of classroom videos using a predetermined set of constructs drawn from standard classroom coding schemes and commensurately predefined AI solutions. This project breaks new ground by developing a dynamic AI that uses multi-modal components of classroom videos to better understand, model, and discover key facets of Pre-K instruction. In effect, success enables education scientists to ask novel measurement questions about classroom video corpora post hoc without the excess cost or labor of needing manual coding or developing a specific AI to code a certain scheme, which too requires massive amounts of annotated training data. If successful, the flexibility of this AI system will allow it to return query results beyond what is captured in existing coding schemes and answer questions that emerge as the education field evolves over time. The study team also includes PIs Dr. Jason Corso (Professor of Robotics), Robin Jacob (Youth Policy Lab), and Sandra Tang (Youth Policy Lab), as well as Pamela Davis-Kean (ISR) and Annie Taylor (Education Policy Initiative).