Notices of the American Mathematical Society
Welcome to the current issue of the Notices of the American Mathematical Society.
With support from AMS membership, we are pleased to share the journal with the global mathematical community.
- Previous Issue
- Volume 72 | Number 5 | May 2025
- No newer issues
PDFLINK |
AMS Advisory Group on Artificial Intelligence
Communicated by Notices Associate Editor Gregory Lawler
AMS President Bryna Kra set up the Advisory Group on Artificial Intelligence and the Mathematical Community in mid-2023. Mark C. Wilson interviewed the chair, Akshay Venkatesh, in June 2024.
Q: What is the scope of the AMS Advisory Group on AI and the Mathematical Community?
A: What we are trying to do is think about what the AMS should be doing in relation to AI, and then pass those questions or issues on to the existing policy committees. It’s preliminary—a way to start the ball rolling. The effect of AI on math is evidently going to be a long-term thing. We’re just at the start of that now. I’m in charge of the committee, not because I know a lot, but just because I think it’s important for the community to think about it.
We wrote a collection of white papers corresponding to the AMS policy committees, which are meant to be just a starting list of issues for those committees to think about.
Q: Can you give a little more information on how the advisory group members were chosen?
A: We wanted to have people who had overlap with the policy committees so that they could be liaisons, and people who had thought about the issue in some way. I was involved in a Bulletin article (see below). Our other members are much more involved, in proof formalization and so on.
Q: What can AMS members do to engage with these issues? I noticed you have a feedback form on your committee website.
A: I think we could really use more resources—links to people who really used AI in some way in their teaching or research. If someone reading this has given a course on Math and AI, or is using AI in their research or teaching in some way, it would be wonderful if they could write something about it and let us know so it can be shared with the community. Something the AMS can do is be a forum, so we learn from each other.
Q: The Bulletin recently has two issues that you were partially responsible for curating, on the general topic of the impact of machines on mathematics. Can you say something about that?
A: That was based on a conference at the Fields Institute in Toronto, in 2022. We had a very interesting group there. We had mathematicians, but not only mathematicians. We had social scientists of various types, discussing also history and philosophy and anthropology of math, which I thought was incredibly helpful. What’s particularly nice about that collection of articles, is that it’s something of a dialogue between mathematicians and humanists. I really think that that dialogue is worthwhile keeping up for us, because we we can’t see ourselves from the outside.
Q: What is your rough idea of what the Advisory Group will do in the remainder of its term?
A: One thing we have discussed, and it would be nice if we can get somewhere on, is the issue of how AI providers are using academic papers to train large language models. I think it would be good to have more clarity and possibly more control about how our papers enter into that process. I know there are proposals now from publishers to license their content to companies that are training AI models. I think that’s something AMS has to think about. I’m sure the arXiv is being used to train all kinds of things, so it would be nice for us to have a bit more clarity about when can our work be used and whether we have rights over that. Having some transparency about what’s going on would be nice.
Q: Are you concerned about ethical issues posed by authors using AI tools?
A: The AMS committee on publications has already adopted the COPE (Note: a publisher organization, not the AMS committee with the same acronym) position statement on usage of AI and our advisory group recommended to the AMS Council some very minimal similar guidelines for reviewers, based on NSF guidelines: basically, reviewers shouldn’t be uploading content to AI. At the same time, I don’t believe it’s the role of the AMS to be incredibly prescriptive about this.
Q: A lot of people who are not native English speakers are using AI tools in order to write more smoothly, which could be a perfectly good use, couldn’t it?
A: I also know many people who have been writing like an AI for a long time before we’ve had AI! I think it’s clear we want to err on the side of being very transparent about our usage of things. I don’t necessarily see that there’s an issue in using AI to improve your abstract, but it will be good if people declare it.
Q: The 2025 JMM will have an AI theme. What are your hopes for this?
A: There’s a list of AI activities at the JMM—panels, special sessions, workshops—now available online. They cover a lot of ground! Some of them are about how we can use AI in our teaching and research—I think we can learn a lot from each other about how to effectively use these new tools. Others go in the reverse direction, thinking about the creation of new mathematics for AI. It’s a new area of science, and there’s probably a lot of mathematical theory that will be created to go along with it. I think it’s important for our community to be exposed to that, and that is something that I certainly hope will come out of some of the JMM sessions. I don’t know how successful or not that kind of theory will be, but it is worth us trying to think about it.
Q: Why are you personally so interested in this topic?
A: I think it will fundamentally alter what math looks like, and I would like us to think about that in advance. We have some choices to make, and it would be good if we make those choices more consciously.
Q: There are obviously some positives in terms of efficiency, quicker development of mathematical areas. What are some possible negatives that come to your mind if we don’t do it right?
A: I worry that as machines think more, we will think less! An interesting issue to think about here is that, the more cognitive support we get from machines, the more we can engage with very long and complicated proofs. I think we have to think about as humans, how we want to interact with such things. We already have a phenomenon where math is becoming more and more technically complicated. In the future, machines will both be doing the math and assisting us in understanding it. There’s a balance here that I’m sure will change a lot, and it would be nice if it changes in a way which is pleasant for humans.
As a community we should try to look further than a couple of years in the future. We have to start thinking now about when computers can do vast parts of what we can do, and how would we like that to look.
Akshay Venkatesh is a professor at the Institute for Advanced Study. His email address is akshay@ias.edu.
Mark C. Wilson is the editor-in-chief of the Notices. His email address is notices-ed@amsjournals.org.
Article DOI: 10.1090/noti3070
Credits
Photo of Akshay Venkatesh is courtesy of Andrea Kane.
Photo of Mark C. Wilson is courtesy of Mark C. Wilson.