A few weeks ago, OpenAI published more than 700 AI-generated mathematics papers, containing hundreds of claimed breakthroughs on previously unsolved problems.
The reaction has been extraordinary. Some mathematicians see it as one of the most important developments in their field’s history, while others accuse OpenAI of undermining mathematical research and potentially damaging the very process by which mathematics advances.
I will only quickly cover the main practical complaints, as these have been discussed extensively elsewhere.
• OpenAI is overwhelming mathematicians with work it hasn’t verified. The repository currently lists 719 papers, but fewer than half of the main results have computer-checkable formal proofs. OpenAI has already withdrawn three papers after finding a mathematical error and revised another 14. Releasing results at this scale shifts a huge burden of verification onto mathematicians.
• OpenAI is claiming the rewards of discoveries built on others’ work. These problems weren’t solved in a vacuum: the AI builds on insights developed by human mathematicians. A researcher might dedicate much of their career to a difficult conjecture, sharing their progress along the way only for OpenAI to publish a solution first, capturing much of the credit (and potential commercial benefit).
• OpenAI risks making mathematical research dependent on a handful of private companies. As I describe in Chapter 11 of Tiny Experiments, mathematics has traditionally advanced through a relatively open research community, where mathematicians build on one another’s discoveries. Proprietary AI systems could concentrate control over mathematical discovery in a few powerful companies, allowing commercial priorities to influence which problems get solved.
These are very real concerns, and as someone who started her career in the tech industry, I hope the teams at OpenAI take stock of what happens when they unleash their shiny new toy on an entire field.
But as a neuroscientist who studies curiosity, what I’d like to explore in today’s newsletter is the psychological and emotional impact of seeing your entire field seemingly overtaken by machine intelligence.
This impact is already visible. A mathematician whose doctoral research centred on a conjecture OpenAI claims to have solved, has spoken about sadness and anxiety about his future career. Another who had hoped to spend years exploring a problem before OpenAI published a claimed solution wrote: “I feel a profound sense of grief for the working conditions I believed I would have.”
Why might this hurt so much? Because discovering something ourselves is different from simply being told the answer.
Research shows that “aha moments” are associated with activity in the brain’s reward circuitry. Having the opportunity to wrestle with a problem before being taught how to solve it also helps develop a deeper understanding so we’re better able to apply it to new situations. Lastly, curiosity helps us remember not just the answers we sought, but unrelated information encountered along the way. The journey towards an answer can teach us things we weren’t even looking for.
That journey of discovery isn’t just about developing an understanding, it’s also about finding meaning in our work. People who copy AI-generated writing report less ownership, confidence and meaning than those who write independently. In that same experiment, those who drafted their own work and used AI to improve it didn’t show the same drop. It suggests that the impact of AI feels different when we’re creating with it rather than just receiving its output.
Of course, you don’t need to be the first person in history to understand something to experience the joy of learning. Learning from other people’s discoveries can be deeply rewarding too, and AI could open extraordinary new directions for research in general.
But feeling that our work matters requires more than generating answers. We also need opportunities to exercise our abilities, follow our curiosity, make progress and contribute something of our own.
If mathematicians increasingly spend their careers checking and explaining discoveries produced by proprietary AI systems, we risk taking away some of the very experiences that make mathematical research meaningful.
This raises a much bigger question, one that goes far beyond mathematics: are we so focused on making intellectual and creative work more productive that we’re forgetting what makes that work meaningful to the people doing it?
I do think we need to seriously consider what happens when we optimize for the number of problems solved without taking into account the people who dedicate their lives to solving them. Mathematics may be one of the first fields confronting this question at such a dramatic scale, but it certainly won’t be the last.
So I’d like to open up the conversation here. What happens to our curiosity, motivation and sense of meaning when machines can increasingly do the work we love? And how do we make sure that, as these technologies advance, we don’t lose the human experiences that made us want to pursue that work in the first place?