Constitutional AI for Proof Feedback at Scale
Project Director:
EDC Staff:
Funded by:
Renaissance Philanthropy and XTX Market’s AI for Math Fund
Partners:
University of Florida; Marquette University; St. Olaf College
Services:
Design & Development
Region:
United States
Duration:
2026-Present
Challenge
Writing a mathematical proof means explaining why a statement is true through a logical argument. Students need an enormous amount of practice to learn to write proofs, and ideally they would get personalized, immediate, and expert feedback on all of that practice. Given finite time and resources, faculty cannot be expected to meet all of those needs, even as students progress through the math major.
A previous National Science Foundation (NSF)-funded collaboration between lead partner University of Florida, EDC, Marquette University, and St. Olaf College worked to solve this problem by developing HaLLMos.com, an online tool that provides AI feedback on students’ proofs for introductory college mathematics courses. Now, the Constitutional AI for Proof Feedback at Scale project extends this approach to more undergraduate mathematics courses, bringing expertise from a wide variety of mathematical fields to the design of feedback.
Key Activities
The project team is carrying out the following activities:
- Convene mathematician Fellows to articulate what good feedback should do and how expectations for mathematical proofs vary across Linear Algebra, Topology, and Analysis
- Expand HaLLMos’s feedback principles and collection of expert feedback examples to support proof writing in courses across the full undergraduate mathematics curriculum
- Refine AI feedback through Fellows’ review of student proofs and responses, preserving student thinking while attending to mathematical rigor and opportunities for revision
- Release an open collection of student proofs paired with expert feedback and a reusable approach for communities to guide AI with disciplinary expertise
Impact
- The work will expand access to feedback that supports students in revising their own mathematical proofs across the undergraduate curriculum.
- The work will provide openly shared resources and a replicable approach for incorporating expert knowledge into AI systems in other fields.
