What's the point of MIT? OpenAI's math proofs have thrown students and researchers into chaos.

Inside the math department at the Massachusetts Institute of Technology , more than 100 students, professors, and postdocs solemnly shuffled into a lecture hall, as if they were going to group therapy. Two days before, on Sept. 8, OpenAI announced that its artificial intelligence agents solved one of the famed “Millennium Prize Problems,” involving the Navier-Stokes equations, which describe how fluids move. Solutions to each problem, posed in 2000, come with a $1 million prize and celebrity status in the field. For MIT’s mathematicians, especially doctoral students, the technological milestone marked the start of a full-fledged freak-out — and they needed a forum to discuss AI. Last week, their alarm continued when OpenAI published proofs to nearly 400 additional previously unsolved problems. It will take some time to sort out the significance, and accuracy, of these proofs. But the overarching feeling among grad students is a fear of being displaced, if not replaced. “The personal stakes are kind of highest for us,” said Dora Woodruff, a third-year math PhD student at MIT. “We’re the ones who are actually going to be on the job market in a couple years. We’re the ones thinking about whether we want to spend our lives in this field, dedicating [it] to this kind of work.” The upheaval comes at a time of deeper reckoning for MIT, where AI is almost all anyone can talk about on campus. President Sally Kornbluth said in an email to the community that the school has a “duty to prepare our students to know when and how to use AI effectively, wisely, ethically, and responsibly, and when not to use it at all,” and now each class syllabus comes with an AI use policy. AI is already changing campus life, according to a 40-page report the university published in August. Administrators had tasked a committee of faculty and students with studying the impact of AI on teaching, research, and learning, and asked them to suggest how the university should respond. The Stata Center at MIT. Jessica Rinaldi/Globe Staff The report found that AI agents had become advanced enough to complete any written undergraduate assignment and that office hours were now effectively obsolete, with students trading conversations with teaching assistants for chatbots. Other schools have weighed their own approaches. A Brown University faculty group published a report in July that recommended establishing clear AI rules. At Harvard University, David Deming, the dean of the undergraduate college, said one of his priorities this year would be to develop usage policies. Some schools, like the University of Chicago, have taken staunch stances in the opposite direction — banning the use of laptops in law school classrooms. But as administrators weigh long-term changes in response to AI, math students are feeling the immediate impact. With the published proofs, OpenAI, the maker of ChatGPT, showed its technology could answer some of the most complicated questions in the field. The pursuit of such problems is what drives MIT students like Woodruff, who is quick to explain concepts that are beyond most people’s understanding of algebra, like the fact that she studies combinatorics (counting and patterns) and has marveled at 100-page proofs about polytopes (geometric objects). In recent months, that excitement had been engulfed by AI fears, and she knew others in the department felt the same. To prove it, she turned to what she knew best: numbers. In August, Woodruff, along with Hanna Mularczyk and Anna Brandenberger, two fifth-year PhD students, conducted a survey of the department, asking graduate students how they felt about the state of math in the age of AI (45 out of 120 people responded). In the lecture hall, Woodruff started the meeting by sharing some of the survey results. Three out of four respondents said they feared AI will make it harder to get a tenure-track job. The majority of participants (71 percent) were now looking to pivot, responding that their chances of continuing into “academic math” had decreased. AI made the field boring, one student said, but another shared they were afraid of falling behind if they didn’t buy in. One by one, professors spoke about their personal love for mathematics and their careers. But students said they now see a changed field that has become less collaborative and more unstable as a race against AI has emerged. The gathering ended with an ice cream social on the lawn outside. On campus, between classes, undergraduates said it was harder to find other students to pull all-nighters with to work on the notoriously hard problem sets. A few years ago, that wouldn’t have been a consideration, but now Claude or ChatGPT can spit out a fairly accurate answer in minutes. Woodruff said, laughingly, that the department had a stereotype — “antisocial mathematicians” — but AI’s threat has provided common ground across campus. Humanities students have consoled Woodruff, who dabbles in creative writing herself, to say there is a similarity between what mathematicians and artists are facing: “This sort of encroachment of AI on what we do, people questioning the value of what we can produce,” Woodruff said. Last year, a student came to Mularczyk, who helped conduct the math department survey, for help with a problem in multivariable calculus. He wondered if he had solved a problem correctly, before admitting that “Chat said to do it that way” (referring to ChatGPT). “I actually had to defend myself against Chat about a calculus problem, which is not really fun,” Mularczyk said. Gone are the days of learning by doing, it feels to Mularczyk, who was taught that “the way you learn to be a mathematician is to struggle through math.” Now, students are encouraged to use AI in many math classes, where instant results can come without explanation. “There’s maybe this instinct to be like, ‘Here’s all the toys, play with them,’ ” Mularczyk said. “Oftentimes, our goal is not to have an oracle built into your computer to tell you what to do.” For years, Pavel Etingof, a tenured professor in the department, said he considered the job of mathematicians to “expand the boundaries of the field by proving new theorems.” He knows that will no longer be the case for his students in the age of AI. “Proving new theorems is very difficult and requires ingenuity, it requires big computations and sometimes it requires complicated abstract technical tools, and also it requires a lot of knowledge and background,” he said. “There will be less of that now because powerful AI models can do it faster and faster, and better.” He sees the role of a mathematician shifting — now to a translator, rather than a technician who cranks out proofs. “This work might be done by AI, but somebody needs to interpret this and communicate this,” Etingof said. “I’m hopeful that [the] mathematical profession will still exist and, in fact, become more interesting even than before.” The long-term prospects of how the industry will shift are important conversations, but for doctoral students, there are far more urgent questions. Mularczyk scrolled through the most recent set of proofs from OpenAI, “praying that a problem we were working on was not on there,” which might mean that years of work would come to a halt. “With that work goes all of the exploration, understanding, and discussion involved in that process, gone, dead in the water,” Mularczyk said. A job search now, especially in academia, feels fleeting. Many are not sure what plan B would look like. But scattered throughout campus, including one just off Killian Court, advertisements on sandwich board signs offered one suggestion for graduate students: “a six-month immersive experience” at OpenAI. Michaela Towfighi can be reached at michaela.towfighi@globe.com . Follow her on X @michaelatowfghi .