We learned the wrong lesson from the AI-frontier math controversy
Some combination of human and AI recently solved, or claims to have solved the “Navier-Stokes” millennium math problem. You will be forgiven if you don’t already have a deep understanding of this, but it is a math problem dealing with fluid and air dynamics. As with many frontier math challenges, the impacts on human life are often downstream from the work product, but there are implications for things like forecasting and aerodynamics. This is actually one of several areas where AI has been used to further mathematics in recent months.
The proposed solution has been completely overshadowed by the drama of the publication and intellectual property. Mathematician Tristan Buckmaster had been working with Levent Alpoge from Anthropic on a novel process for this problem using a number of AI tools. It is possible a team from OpenAI was also working using internal models and arrived at a similar set of solutions, it is also possible OpenAI harvested ideas from chatlogs and used them to direct its own agents. The resulting dust-up has crossed over from tech-gossip into mainstream news and academia (we all love drama). This has culminated (for now) in an open letter from some of the world’s leading mathematicians about the alignment between LLMs and the field of mathematics.
Buckmaster has understandably been vocal about what happened, telling The Verge “They don’t care anything about us as a community. It’s all about this petty drama between two trillion-dollar companies that are acting like children.” James Robinson, a mathematician from the University of Warwick was quoted in The Guardian “It’s frustrating to see these big tech companies burning this fuel up just so that they can show off about how great their latest model is.”
I don’t pretend to know the truth of what is happening here in terms of how much OpenAI discovered vs. how much they took, I am writing to make a different point–this seems like a big deal and I think we are mistaking the drama for the story.
Just a few weeks ago AI leaders were coming to terms with the public souring on them and their products. Dario Amodei the CEO of Anthropic mused out loud that “the thing that will work (in changing public sentiment) is actually curing cancer.” Sam Altman from OpenAI went further “I don’t think a cure for cancer is enough.” Everyone seemed to settle on the truth that the labs needed to actually produce the scientific discoveries they had been promising.
We didn’t get a cure for cancer, but we did kind of get the scientific breakthroughs we were promised and we told everyone we were waiting for. Even in a world where OpenAI behaved poorly and stole notes from Buckmaster and Alpoge, everyone was using LLMs and the work would not have been possible without them. We are increasingly dealing with high stakes discovery and there is a lot to talk about, but we need to be having the right conversation. When the Soviets stole secrets to develop their own atomic weapons the key issue was not whether their scientists deserved credit the issue was–now they have nukes.
Maybe some of this public outrage about credit for this breakthrough is warranted, but I’m betting most of us were not actually all that deep into the politics of author order in mathematics journals before this controversy. I am going to stop short of psychoanalyzing this, but it really seems like a bunch of people don’t like AI and so whatever is produced or associated with AI becomes toxic to them. If Anthropic did cure cancer would we welcome that news or be mad that the wrong people got credit? If work at OpenAI led to a breakthrough in fusion energy which helped us curb global warming would we celebrate it or be upset it was done in a noisy data center?
I don’t think these are hypothetical scenarios. A few years ago these models failed miserably at even basic mathematics. Today they are solving or helping to solve problems we have been at for decades or even centuries. It is hard to imagine other advancements are not imminent. Before those arrive we need to assess our own readiness. Will we accept innovation if it is political, uneven, and interpersonally ugly? I sure hope so because if we attach a purity standard to innovation there are a lot of really useful things that were invented or developed by some really bad people that we will have to throw away.