Harvard Mathematicians Split Over AI as a Former Harvard AI Adviser Joins Anthropic's Oversight Trust

Harvard's math department is divided over how far artificial intelligence should go in research and teaching, and a Harvard physicist who helped bring AI tools to the Faculty of Arts and Sciences has taken a seat on the oversight body of an AI company, according to two reports this week in The Harvard Crimson.

The Navier-Stokes claim

In September, OpenAI released what it said was a solution to the Navier-Stokes existence and smoothness problem, which asks whether the equations governing fluid motion can develop mathematical breakdowns, The Crimson reported. The problem had been unresolved for roughly 90 years and is one of seven Millennium Prize Problems, each carrying a $1 million prize from the Clay Mathematics Institute.

The Crimson interviewed a dozen Harvard mathematics faculty, researchers and lecturers, who expressed differing views about what the claimed solution means and how AI should be used in the field. Some raised concerns that AI-generated proofs can lack proper citations and thorough explanations, making them difficult to evaluate.

"We should use AI as a tool to increase that human understanding, but not to replace it or give up our knowledge and let it only be stored in machines," said Melanie Matchett Wood, a Harvard professor and member of the independent Advisory Group on Mathematics and Artificial Intelligence. That group, formed in September, aims to provide guidelines for the release of mathematical research by AI companies.

Others said accuracy matters more than authorship. "If someone is able to tell you the equations of general relativity, it doesn't really matter whether it was Einstein or Hilbert or ChatGPT," said Ahsan Z. Khan, a research associate at Harvard's Center of Mathematical Sciences and Applications. "It's still general relativity."

Lorenzo Riva, a CMSA lecturer, said he has avoided using AI in his research in large part because of concerns that models could plagiarize his work. "That's something that authors should have control over: whether their papers are being used for training or not," Riva said.

Tool, not author

Several faculty told The Crimson that AI has a role in routine work but that they avoid relying on it for ideas, solutions or proofs. "I find its effectiveness for my research, so far, best served in things like literature search and coding," said mathematics professor Tomer M. Schlank. Schlank also cited "serious concerns" about the ethics of using AI models to solve mathematical problems, arguing that the profit incentive of AI companies conflicts with the field's pursuit of knowledge.

Michał P. Szachniewicz, a Benjamin Peirce Fellow in the department, said he uses AI to clean up lecture notes but prefers discussing problems with other mathematicians to "sitting and typing" with ChatGPT.

In the classroom

Brendan A. Kelly, director of the Introductory Math Program, said he hopes to balance adopting AI tools with preserving meaningful learning. Several introductory math courses introduced tech-free policies for classes and sections this fall, which Kelly said were intended to prioritize an education that "elevates students' skills, dispositions, and academic identities."

At a September summit co-hosted by the math department and the CMSA, 24 mathematicians developed recommendations calling for more in-person written and oral assessments and for training students to use AI tools in research, The Crimson reported. Laura G. DeMarco, the department's director of undergraduate studies, said the department was "working hard" to determine how AI fits into the collaborative culture of mathematics.

Stubbs joins Anthropic trust

Separately, Anthropic announced Wednesday that Harvard physics and astronomy professor Christopher W. Stubbs has joined its Long-Term Benefit Trust, The Crimson reported. The trust has authority to appoint a majority of the company's board and advises its leadership on decisions involving the risks and societal impacts of AI.

The announcement came less than three weeks after FAS Dean Hopi E. Hoekstra announced that Stubbs was stepping down as the faculty's senior adviser on generative AI. In that role, Stubbs, who was Harvard's dean of science from 2018 to 2024, helped secure access to leading AI models for FAS affiliates, including Anthropic's Claude, and helped design guidance for faculty and researchers, according to Hoekstra. He remains on the faculty and on the FAS committee on generative AI. Hoekstra named physics and astronomy professor Douglas P. Finkbeiner, who spent last year on leave working with Anthropic, as his successor.

"I use AI tools in my research every day and see what they make possible, and as an educator I have confronted some of the hard questions they raise. As a grandfather, I want to help ensure that AI is developed and deployed for the long-term benefit of humanity," Stubbs said in Anthropic's announcement. He declined to comment to The Crimson, directing inquiries to Anthropic.

Harvard's ties to the industry

Stubbs is one of several Harvard affiliates involved in the AI industry. Harvard Corporation member Mariano-Florentino Cuéllar served on Anthropic's trust from January until August, when he stepped down to become the company's first global affairs officer, The Crimson reported. Two of the three inaugural Catalyst professors, a program launched in 2024 to encourage faculty ties to the private sector, work in AI: computer science professor Boaz Barak, who also holds a technical staff position at OpenAI, and physics professor Michael Brenner, who researches AI for Google.

University President Alan M. Garber announced in September that Harvard was investing $150 million in research infrastructure, including expanding access to commercial AI models, according to The Crimson.