Catch up quickly: OpenAI released 722 manuscripts organized into 372 result groups (“families”) on long-standing mathematical problems on Tuesday, inviting scholars and researchers to review and expand on this material.
- The reception mixed scientific enthusiasm and genuine unease. A Rutgers University mathematician said on X that a result related to the Riemann Hypothesis would warrant an automatic Fields Medal if a human had done the work.
- Some mathematicians have questioned these results, as well as another OpenAI solution released last month, questioning whether they represent original advances or rely heavily on previous human input.
- Others have downplayed the usefulness of AI-generated mathematics. “You can easily discover new mathematics,” Stephen Wolfram, a renowned computer scientist and physicist who has closely followed AI for years, said at an event Tuesday for the National Mathematics Museum. “You can easily come up with trillions of theorems. The problem is that most of those theorems don’t interest anyone.”
Zoom: Like computer programming, mathematics gives AI something exceptionally valuable: a way to know when it is working correctly.
- A proof can be examined by mathematicians and, increasingly, translated into formal languages that computers can check line by line. Likewise, it is immediately possible to know whether autonomously written AI code actually works.
The big picture: Software engineers have already experienced this transition.
- AI coding tools have evolved from autocomplete and debugging to agents capable of writing substantial amounts of software and performing complex engineering tasks.
- This shift changed not only how code is written, but also what it means to be a programmer, bringing a healthy dose of respect, amazement, and dread to many longtime software engineers.
- Just as software engineers did before them, mathematicians have begun to question the implications of some of OpenAI’s findings for higher education and the training of a new generation of theorists.
Yes, but: Some software engineers have successfully defined the roles that humans still play in generating quality products. Even though new AI coding tools have dramatically increased the code an engineer can deliver, humans still play a crucial role in creating quality software.
- Mathematicians have begun to echo this point, noting that the most significant advances in their field often involve building a framework or structuring a complex idea. Many are skeptical about the ability of an AI system to replicate human originality.
Between the lines: Beyond the math, the results indicate the likelihood that AI disruptions will continue to expand into new areas, partly answering a long-standing question from AI skeptics.
- While researchers at competing startups question the usefulness of OpenAI’s mathematical findings due to their limited practical applications, some acknowledge that they provide evidence of rapid advances in AI.
Gn bussni
