MIT panel challenges how mathematicians earn credit in the AI era
A journal’s submission surge and greater emphasis on interviews show how the debate over AI proofs is reaching professional practice.
Loading page…
A journal’s submission surge and greater emphasis on interviews show how the debate over AI proofs is reaching professional practice.
Listen to this story
At an MIT panel on Nov. 16, mathematicians focused on how journals and universities can judge whether people understand and can explain work produced with AI. The discussion pointed to practical changes in submissions, research communication, and interviews, while leaving open whether mathematical priority should be rethought. Panelists’ criticism of AI proofs was their assessment, not independent verification of OpenAI’s claims about AI-generated solutions.
Bryna Kra said a specialty journal she manages hired another managing editor after a large increase in submissions reporting AI-generated or AI-assisted results.
MIT News reported that OpenAI released 722 manuscripts in October, claiming they contained entirely AI-generated solutions to major unsolved problems.
Kra pointed to Mathematical Discourse, a peer-reviewed video journal for research talks, as one example of growing value placed on explanation and exposition.
Producing a mathematical proof and understanding it are becoming separate tests of achievement, mathematicians argued at a November 16 panel at MIT. Their discussion, covered by MIT News on November 20, reached beyond the quality of AI-generated results to a practical question: how should journals and universities judge the people behind mathematical work?
The event, “AI and Society: The Future of Math,” brought together MIT’s Jacopo Borga, Harvard’s Michael Hopkins, Northwestern’s Bryna Kra and New York University’s Tristan Buckmaster. Science writer Kevin Hartnett moderated. The panelists disagreed about abandoning established rewards, while describing changes already reaching journal submissions, hiring interviews and students’ approaches to research.
The distinction starts with what a proof contributes. Panelists questioned whether reaching a result is enough when other mathematicians cannot follow the ideas behind it. Buckmaster argued that mathematics needs human understanding to be important. Hopkins criticized the AI-generated papers under discussion as containing “strange ideas,” drawing a distinction between AI mathematics and human mathematics.
The backdrop includes large claims of machine-produced discovery. MIT News reported that OpenAI released 722 manuscripts in October, claiming entirely AI-generated solutions to major unsolved problems. Those claims, and mathematicians’ objections to the quality and intelligibility of the resulting proofs, framed the panel’s discussion. The panelists’ remarks were assessments of that work, not independent verification of the claimed breakthroughs.
Kra proposed reconsidering priority: the prestige attached to being first to prove something. She argued that communicating how a proof came about could itself be pioneering work. Hopkins pushed back, saying researchers would still want to be pioneers even if the field discarded priority. Their disagreement concerned what mathematics should reward, not simply whether researchers should use AI.
Speed should not be a priority. It’s not about the proof production anymore. We need to find other ways to assess understanding.
Bryna Kra, Northwestern University mathematician, at the MIT panel
Kra described a concrete workload change at a specialty journal where she is a managing editor. A large increase in submissions, primarily reporting AI-generated or AI-assisted results, prompted it to hire another managing editor. She also said some publications were placing more value on exposition: writing that explains the reasoning behind a conjecture or proof.
That emphasis extends beyond conventional papers. Kra pointed to Mathematical Discourse, a new peer-reviewed video journal intended to publish research talks. She also said interviews were becoming more important in admissions, hiring and promotion. Written ideas alone were no longer enough; candidates needed to demonstrate that they understood those ideas and could explain them to another person.
Hopkins offered an example of explanatory work with value beyond the current AI dispute. One of his students was working to prove a theorem from the 1970s whose original language and construction had become difficult to understand. He compared that distance to the unfamiliarity of some AI proofs, arguing that curating and translating knowledge offered substantial opportunities.
For students, the panel described no single route forward. Hopkins said his group included both AI users and “AI vegans” who avoided the tools, and he was supporting both. He singled out a student with strong mathematical instincts but difficulties executing ideas, whose skill at prompting AI produced interesting insights. Hopkins suggested this could represent a new kind of mathematician.
Borga urged giving overwhelmed students space to learn a subject he described as unusually slow. Buckmaster’s advice was to stay in mathematics rather than flee to another field: find new ways to be creative and communicate ideas.
Loading discussion...
Join the conversation
Explain which contribution deserves more recognition, or why they should count equally.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.