正文 · AI 翻译
IT Home reported on October 27 that, according to an October 26 report by Tide News, Fields Medal winner and University of Chicago mathematics professor Deng Yu delivered a keynote speech at last weekend’s Intellectual and Mr. Science forum, systematically setting out his thoughts on mathematical research in the age of AI.
Deng Yu said AI has two advantages over humans: it has an unlimited breadth of knowledge, and the cost of trying things and making mistakes is extremely low. But we must hold to one principle: in mathematical research, as in anything else, humans remain the principal actors. AI is a tool that we use to conduct research.
Deng Yu believes that if AI is treated as a researcher, it should follow academic standards. If AI proves a highly significant problem, a very reasonable requirement is to disclose its chain of thought. Yet AI companies refuse to disclose it, which plainly goes against academic convention. “This AI company is unwilling to do so. Of all the proofs they have published, only a very small number have actually included the chain of thought. In most cases, they have used so-called prevention of distillation as an excuse to refuse to disclose the chain of thought and related information.”
Deng Yu said that AI companies cannot freely distill humanity’s online literature and knowledge when they are the ones doing it, but then, when others simply want to understand their chain of thought—when in fact no one wants to distill their models—decide that others want to distill them and start saying this and that are off limits.
Deng Yu proposed using open-source mathematical models to ensure that the mathematics community has models of its own under its own control, without having to depend on such AI companies. The use of these models should also follow academic standards, including those concerning data use, disclosure of chains of thought, and traceability.
IT Home noted that on September 11 local time, 25 Fields Medal winners, including Terence Tao, Deng Yu, and Peter Scholze, jointly issued a statement as its original signatories. Titled “A Severe Misalignment of AI in Mathematics,” it warned of an emerging “severe misalignment” between AI companies and the mathematics community.
The statement said that over the past few months, the mathematical capabilities of large language models have improved significantly, enabling them to solve many major open problems across mathematics. However, the push by AI companies to use mathematical problems as benchmarks to solve is harmful to both mathematical science and the mathematics community. The goals of AI companies are severely misaligned with those of the mathematics community, and these issues are part of broader alignment problems affecting other scientific and creative professions, and even society as a whole.
The statement noted that these AI solutions are often released hastily, without time for formal write-ups, for isolating new methods and ideas, or for citing relevant work by others. As in all creative professions, this raises serious questions of attribution and plagiarism. Moreover, without mathematicians willing to take responsibility for developing AI-conceived ideas and integrating them into the mathematical canon, those ideas can never be fully realized, and the crucial human chain of transmission among mathematicians will be lost.