RUME 27
2025
Alexandria, Virginia
Student Validations of Author-free AI-Generated Proofs
Page: 1062
We report on a small-scale clinical interview study of students’ validation of AI-generated proofs. During these one-on-one interviews, students were tasked with using ChatGPT to generate proofs and validate those purported proofs. Consequently, neither the interviewer nor the participant knew if the justifications being validated constituted a proof before evaluating it. As such, no two participants validated the same proof. This required the development of novel post-interview analysis methods, which led to some valuable insights into students’ proof validation processes. We view the primary contributions of this work to be both the methodology we developed, and our insights gained into how AI-generated proofs provide a different lens for students’ validation and error detection.