Deng Yu questions AI companies: You take mathematicians' ideas, what do we get in return?

At a time when artificial intelligence is advancing by leaps and bounds, the mathematics community is facing an unprecedented crisis and a major debate. On October 25, Deng Yu, a Fields Medalist and professor at the University of Chicago, set out his thoughts on mathematical research in the AI era at a forum organized by "Intellectuals" and "Mr. Science." Deng Yu's judgment begins with OpenAI's "mathematical offensive." OpenAI once published more than 300 problems and more than 700 articles. Are these works AI's "original creations," or an extension of human mathematics? Deng Yu said he had probably read a few of them. "By my understanding, a considerable portion of these works are very major advances in mathematics. Most of them are based on humanity's existing deep understanding of the relevant problems and techniques. The emergence of these articles itself has a considerable promoting effect on mathematical research, but it cannot simply be regarded as creation out of thin air, detached from human mathematics." Where exactly is AI stronger than humans? Deng Yu believes that one obvious advantage is that "it has an unlimited breadth of knowledge"; another advantage is that "AI has an extremely low cost of trying, or rather, cost of trial and error"—"in this respect, if humans try to compete with AI on speed, there is certainly no hope at all." Since AI has these advantages, why still insist on "humans as the subject"? Deng Yu stressed that this is a very basic premise: "In mathematical research, in everything, humans are always the subject, and AI is a tool, a tool we use for research. We must always be human-centered, using AI to advance our own research and to advance the problems we ourselves are interested in." In his view, under this assumption, humans play the role of setting the direction, and AI can be used to carry out many attempts and trial and error. "In the short term, a situation will emerge in which humans and AI co-evolve. This is a process of mutual learning and mutual promotion, so there is no reason to reject tools, but the premise is to use them correctly." The stronger the tool, the more rules matter. If AI is treated as a virtual researcher, should it disclose "how it came up with the idea" the way humans do? Deng Yu said: "Since it is treated as a researcher publishing research results, it must be judged by academic standards. For AI, the thinking behind a proof is obviously the chain of thought. So if AI proves a major problem, a very reasonable requirement is to publish AI's chain of thought. But this company discloses it only in a very few cases and refuses in most cases on the pretext of so-called anti-distillation." Some will say that commercial companies do not have to abide by academic norms. Deng Yu responded from a commercial perspective: "Commercial companies provide products, and we mathematicians can be users. We pay for your products and input our ideas into your products; you use the ideas we input to train a new generation of products, and then use these new-generation products to solve open problems in the mathematics community and receive all the honors. So what do we get as users? We get nothing. Therefore, even if we look at this from a commercial perspective rather than an academic-norms perspective, there are still problems with this matter." What counts as "correct use"? Deng Yu said: "First, AI should serve as a tool, used by mathematicians or mathematics enthusiasts—users using a tool; AI companies, as tool providers, should not personally step in to replace users in conducting research. Second, users should be able to choose, or be able to easily choose, whether the data input to AI is allowed to be used by AI for training; the right to choose belongs to users. Third, there needs to be traceability." Assuming strong AI can independently solve very difficult problems, what will mathematicians still need to do in the future? Deng Yu stressed: "Under any circumstances, mathematical research will always have humans as the subject. AI can solve problems, but humans can at least pose problems, guide AI toward which direction of application to pursue, and point the way for AI." "AI is like a car. We drive the car, set the destination for the car, point out the direction, and then drive to the destination. The car can drive autonomously and go very far on its own, but if we are not in the car, what meaning does that have for us?" he said by analogy. Mathematics is closer to reality—will that make pure mathematics disappear or merge into other disciplines? Deng Yu said: "Mathematics being more closely connected with real-world problems absolutely does not mean that mathematics will become a subsidiary of physics or biology, still less that pure mathematics will disappear or that mathematicians will be merged into other departments. Mathematics is a whole. We cannot, just because we want to be closer to applications, consider only the parts that can be directly applied and ignore the parts that may not have direct applications; it is impossible to set aside one part of mathematics to develop another part." He stressed: "The most unique quality of humans is the ability to think. I remember Pascal said, 'Man is a thinking reed,' and the point lies in thinking. AI in the future can answer many questions, but it cannot replace our independent thinking. If we give up independent thinking just because we have AI, I think we would be no different from animals."