返回资讯中心
外部精选
软件工程
#开发者实时流

Terence Tao on how AI math solutions miss the point

In a matter of months, AI has transformed the field of mathematics. In September, OpenAI claimed that an AI model solved the million-dollar Navier-Stokes equation. Then, a few weeks later, OpenAI shared results for more…

Hacker News 最新digital552026/11/1 · 16 天前30 分钟阅读

以下正文同步自 Hacker News 最新,版权归原站所有,已转换为易读排版。

In a matter of months, AI has transformed the field of mathematics. In September, OpenAI claimed that an AI model solved the million-dollar Navier-Stokes equation. Then, a few weeks later, OpenAI shared results for more than 350 previously unsolved math problems. The news has prompted an existential conversation among mathematicians: In an age of AI, what’s the role of the human mathematician?

It’s a question that Terence Tao, largely considered one of the greatest mathematicians of this generation, is reckoning with right now. He speaks with Flora about how AI has upended the mathematics world, what the role of math is in this day and age, and why the real value of mathematics lies in the journey, not just the answer.

Read an excerpt of Tao’s forthcoming book, “Six Math Essentials,” which tells the story of math and how it shapes our world.

Invest in quality science journalism by making a donation to Science Friday.

Dr. Terence Tao is a professor of mathematics at the University of California, Los Angeles and the author of “Six Math Essentials.”

[MUSIC PLAYING] FLORA LICHTMAN: Hey, it’s Flora, and you’re listening to Science Friday. Mathematics has been top news in a way it rarely is, almost entirely because of AI. Last month, OpenAI claimed that one of their models solved the Navier Stokes Millennium Prize problem. Mathematicians have been working on this problem for decades, and according to OpenAI, their agents solved it in 88 hours. Then, a few weeks later, OpenAI announced results for over 350 previously unsolved math problems.

Now, there’s debate about how well these problems have been solved and the model’s ability to explain their work, but regardless, the news has prompted a kind of existential conversation among mathematicians. In an age of AI, what’s the role of the human mathematician? Dr. Terence Tao is considered one of the greatest mathematicians of our time. He was taking college-level math courses by age 9, was a full mathematics professor at 24, and won the Fields Medal, a.k.a. the Nobel Prize of math, at 31. He’s also the author of the upcoming book, Six Math Essentials, which tells the story of math and how it shapes our world. Terry, welcome to Science Friday.

FLORA LICHTMAN: You told The New York Times recently that in mathematics, this moment feels like soul searching at gunpoint. That’s a arresting statement. Walk me through it.

TERENCE TAO: Yeah, so mathematics has been very stable and quiet. And it hasn’t made the headlines, as you said, for over a century, pretty much. We are often focused inwards, trying to solve our own problems, mostly of interest to other mathematicians, although every so often, the things we do are of interest to scientists and engineers. So we largely do our own thing, and we haven’t had external actors come and try to play with our problems so much.

But suddenly we have these 800-pound gorillas in our profession feasting on our open problems, which, on the one hand, is great. We’re getting all these new answers to problems that we wanted solutions for, but it is disrupting a lot of things that we are traditionally interested in doing. For instance, when we solve problems, often, these problems are not interesting for their own sake. The Navier-Stokes regularity problem doesn’t help us advance weather prediction or any practical impact right away.

But often, the journey to solve these problems, the story, the collaborations along the way, the partial progress we write up, that is really what’s valuable. And it helps mathematics become more useful in the real world. And that’s been kind of lost in this big scramble to use these machines to grab solutions to these problems without getting any of the understanding that normally comes with it.

FLORA LICHTMAN: You said 800-pound gorillas– 800-pound gorillas feasting. Those are the AI models.

TERENCE TAO: Yes, yes. So they’ve become very, very good at mathematics. So mathematics has been traditionally considered a very hard topic for humans, but it has many of the features that make it particularly good for AI. In mathematics, the problems we work on can be objectively verified. You can check on a computer whether the answer is correct or incorrect, and that turns out to be part of what makes AIs very effective at these problems.

So, yeah, we had this sort of monopoly, if you wish. Professional mathematicians were almost the only people who could work on these problems. And now we’ve built machines that can give us these solutions. But the solutions weren’t actually what we wanted. They were just kind of nominal goals, flags on top of a mountain to move towards. One of my friends called them lighthouses, not destinations to go to for their own sake, but they illuminate the landscape around them.

And so, now, we’ve conquered these lighthouses without getting the understanding. So we now have this backlog of hundreds of proofs that are interesting. And slowly, mathematicians are going through them. But we don’t have any help from, say, OpenAI to help us. Normally, when a human mathematician solves a major problem, they can come to talks, we can call them up, they will answer questions, and we can have really productive conversations with them. There’s nobody at OpenAI I can call to help me understand this dump of 300 papers. We have to do it ourselves. And we will, but it’s going to take a lot longer than normal.

And in the meantime, there were many, many junior people who had research projects which have just been kind of been made obsolete by this dump. So a lot of junior people are scrambling. Some people are suddenly now very, very secretive about what they’re doing. They’re now terrified that if they mention their research directions, which is normally what you do in our subject, someone will just put them into an AI and solve their research project for them before they can actually produce some valuable understanding of that problem. So priority used to be very important in our profession, but now, priority is degenerating into who can feed their problem into an AI the fastest, which is a completely useless competition.

FLORA LICHTMAN: So what do you see the purpose of human mathematicians? What is the purpose of math? And has that changed or not?

TERENCE TAO: Right, so mathematics fundamentally is about understanding. I mean, ostensibly, it looks like it’s about answers. And if you take a math class in high school, it seems like that’s all you’re doing. You’re given homework where you have to say how you find x, or how many apples does Jane have, or whatever. And you hand back answers.

But that’s only part of what mathematics is. There’s a lot of real-world phenomena which are governed, at least in part, by mathematical models, so if you want to study climate, or traffic, or the energy grid, or can you build a more efficient solar energy cell, whatever. And so you do want answers to these questions, but in order to get the right answers, you need to understand what these models are and what their limitations are.

Sometimes the model you have, on paper, it works great, but it actually doesn’t describe the real world. In 2008, we had a massive financial crisis, in part because investment bankers were using a certain mathematical model to model mortgage defaults. And they were basing all these fancy derivatives and financial instruments on this model. And it turned out this model was only an imperfect representation of mortgage defaults. And it didn’t take into account, in particular, systemic shocks to the entire mortgage market. And basically, people were using models that they didn’t understand. And the rest is history, actually.

So I think one of the purposes of a human mathematical community is to provide understanding of the mathematical models that are now increasingly used everywhere, including AI. AI is built on lots and lots of quite advanced mathematics now. And there’s now this fear that we are deploying AI far faster than we’re understanding its strengths and weaknesses. So, yeah, we need the understanding to catch up. And this is a role that mathematicians will increasingly play in the future.

FLORA LICHTMAN: Why do you think this is so destabilizing for mathematicians?

TERENCE TAO: Right. So historically, we have organized our profession around the assumption that solving problems was difficult, and in particular, proving that a result was true or false was hard and scarce and valuable. And so we’ve organized our entire culture, our institutions, our prizes around this scarcity. So if a young postdoc solves a major unsolved problem, they will get papers in prestigious journals. They’d win prizes. They would get job offers, this type of thing. And this is how we have organized our field. Proofs were hard and scarce, and so that was the valuable thing.

But it was always a proxy. Most of the problems in pure mathematics are not useful for their own sake. Similar to how a high school math homework question, what is the side length of this triangle, you’re unlikely to have to use that particular problem in real life. It was always a proxy for just how good you were at problem-solving, or just dealing with abstract situations that require some quantitative thinking.

So we maybe have overindexed on ability to solve problems, but now we have these machines that can do that for us. So there’s now this historic decoupling between the traditional task of solving problems, which is what has motivated our field. And now we’re seeing that this objective is being decoupled from the more fundamental objectives of mathematics, understanding what is going on, how to apply things to the real world, knowing the limitations of our mathematical models, as well as our successes, and training the next generation of mathematicians to continue these traditions. And so the soul searching is, how do we keep the long-term health of our field, now that our main guiding proxy for progress has been optimized to death?

FLORA LICHTMAN: You’ve talked about bringing humanity to math. What does that mean to you? And what does that mean to you now, in this age of AI?

TERENCE TAO: Right. So math is half science, half humanities. But somehow, for marketing purposes, and particularly for applying for funding, we’ve emphasized the science half, and we have downplayed the human half. But math is a very social activity. Part of what gives a mathematical result value is not just its application to the real world, but to what extent it excites other mathematicians, or other humans.

I think mathematics not only is a technical subject that other mathematicians appreciate learning more about, but knowing a bit of mathematics just as a regular person helps make the world less scary. We are surrounded nowadays by technology we don’t understand, that we have these magic devices. We can communicate across the planet.

Now we have these AIs that can talk to us almost at the level of humans, or sometimes, it can sound superhuman. And sometimes, they fail, and now it becomes scary. But knowing mathematics helps you understand technology and demystifies it, and the world just becomes less scary. So I think mathematics has value in just helping people be at peace with the world around them.

FLORA LICHTMAN: That’s such a profound thought. And it also reminds me a little bit of these proofs that are coming out, right? Because that must be also why it’s so destabilizing. If you read these proofs, and you don’t have someone to ask about them, what does that even mean for understanding?

TERENCE TAO: Right. It’s like this artifact. It has all the outward markings of an achievement, but it has none of the substance. People can say, wow, this problem got solved, but I can’t explain it, and you can’t explain it. We can just gawp at it. And that’s no way to do science. And now these solutions are being dumped by actors who just don’t care about cleaning it up. They say, OK, someone else can explai

正文由 探How 从来源站点 RSS 同步,内容未经改写;遇到排版缺失或需要图片、视频时请以原文为准。

打开原文
去社区讨论 →