Guest Talk
Who’s Listening? Levels of Automation for AI That Hears Students’ Math Talk
Guest Talk by Dora Demszky on AI Support for Classroom Math Discussions
14.12.2026
11:15 am - 1:15 pm
Marsstraße 20, Room 605, 80335 Munich
How much of the work should AI take on when it listens to students’ small-group math discussions? Should it transcribe speech, flag important moments, interpret student understanding, or recommend actions to teachers? In this guest talk, Dora Demszky applies the levels-of-automation framework by Parasuraman et al. (2000) to these questions and argues that each level requires its own approach to design and evaluation.
The talk draws on MathOutLoud, a dataset of about 800 video recordings of small-group math discussions in grades 2 to 10, which will be publicly released in January 2027. She will also present a comparison of how LLMs and teachers notice student thinking, as well as a study of teachers using Talk Matters, a tool that pairs classroom video with LLM-generated analyses. Attendees will have the opportunity to try the tool.
Dora Demszky is an Assistant Professor at the Stanford Graduate School of Education, with a courtesy appointment in Computer Science, and leads the EduNLP lab. Her research uses natural language processing and computational methods to support student-centered teaching, teacher learning, and equitable classroom discourse.
The talk is organized by the TUM Center for Educational Technologies and jointly funded by MCML. Registration is required.