Yann LeCun, chief A.I. scientist at Meta, said “the current state of machine learning is that it sucks.”Credit.Fabrice Coffrini/Agence France-Presse — Getty Images
Dear Commons Community,
Participants at this year’s Joint Mathematics Meetings explored everything from the role of A.I. to the hyperbolic design of a patchwork denim skirt.
The world’s largest gathering of mathematicians convened in Seattle from Jan. 8 to Jan. 11 — 5,444 mathematicians, 3,272 talks. This year the program diverged somewhat from the its traditional kaleidoscopic panorama. An official theme, “Mathematics in the Age of A.I.,” was set by Bryna Kra, the president of the American Mathematical Society, which hosts the event in collaboration with 16 partner organizations. In one configuration or another, the meeting, called the Joint Mathematics Meetings, or the J.M.M., has been held more or less annually for over a century. As reported by The New York Times.
Dr. Kra intended the A.I. theme as a “wake-up call.” “A.I. is something that is in our lives, and it’s time to start thinking about how it impacts your teaching, your students, your research,” she said in an interview with The New York Times. “What does it mean to have A.I. as a co-author? These are the kinds of questions that we have to grapple with.”
On the second evening, Yann LeCun, the chief A.I. scientist at Meta, gave a keynote lecture titled “Mathematical Obstacles on the Way to Human-Level A.I.” Dr. LeCun got a bit into the technical weeds, but there were digestible tidbits.
“The current state of machine learning is that it sucks,” he said during the lecture, to much chortling. “Never mind humans, never mind trying to reproduce mathematicians or scientists; we can’t even reproduce what a cat can do.”
Instead of the generative large language models powering chatbots, he argued, a “large-scale world model” would be the better bet for advancing and improving the technology. Such a system, he said in an interview after the lecture, “can reason and plan because it has a mental model of the world that predicts consequences of its action.” But there are obstacles, he admitted — some mathematically intractable problems, their solutions nowhere in sight.
Deirdre Haskell, the director of the Fields Institute for Research in Mathematical Sciences in Toronto and a mathematician at McMaster University, said she appreciated Dr. LeCun’s reminder that, as she recalled, “the way we use the term A.I. today is only one way of possibly having an ‘artificial intelligence.’”
Dr. LeCun had noted in his lecture that the term artificial general intelligence, or A.G.I. — a machine with human-level intelligence — was a misnomer. Humans “do not have general intelligence at all,” he said. “We’re extremely specialized.” The preferred term at Meta, he said, is “advanced machine intelligence,” or AMI — “we pronounce it ‘ami,’ which means friend in French.”
Dr. Haskell was already sold on the importance of “using A.I. to do math, and the huge problem of understanding the math of A.I.” An expert in mathematical logic, she plans to use a theorem-proving program to create the equivalent of a textbook: a collection of results that can be used by A.I. systems to generate and verify more complex mathematical research and proofs.
For Kenny Banks, an undergraduate at the University of North Carolina at Greensboro who attended the J.M.M., artificial intelligence does not appeal as a tool for guiding exploration. “I think the mathematics that people currently love is driven by human curiosity, and what computers find interesting cannot be the same as what humans find interesting,” he said in an email. Nevertheless, he regretted not squeezing any A.I.-related talks into his itinerary. “The math + A.I. theme was definitely of interest, it just ended up not working with all the things I had planned!”
Other highlights are available here!
Tony
This entry is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license.
