"Hallucinations are not a bug in my classroom. They are the assignment." That's Giorgio Lagna's framing for one of the 13 AI tutors he's built into his biology course at Santa Clara University. In this episode of Educast 3000, the UCSF research scientist and college lecturer joins Ryan Lufkin and Zach Pendleton to walk through what happens when you design AI not to answer questions but to refuse them, not to be right but to argue back. The conversation covers his "Crucible" course design, the DISCO framework research he published this year showing measurable equity outcomes, and why he believes higher ed is making a strategic mistake by treating AI as an integrity problem instead of an assessment problem.

In this episode:

Toll booth gems, boss battle gems, and certification gems explained
Why deliberate AI errors teach better AI literacy than warnings about hallucinations
The cost math on AI-powered oral exams (NYU's run came to 42 cents per student)
How community college students are now running CRISPR research with AI co-investigators
What Pope Leo XIV's first encyclical said about AI and education the day before this recording

Pedagogy Under the Microscope: https://substack.com/@glagna
Structured AI Integration for Equitable STEM Writing: A Pilot Study of the DISCO Framework: https://www.tandfonline.com/doi/abs/10.1080/0047231X.2026.2625111
UCSF profile: https://profiles.ucsf.edu/giorgio.lagna

Further readingPope Leo’s ‘Magnifica humanitas’: AI must serve humanity not concentrate power: https://www.vaticannews.va/en/pope/news/2026-05/pope-leo-xiv-encyclical-magnifica-humanitas-ai.html