Key facts
- Mathematician Terence Tao warns that AI is solving difficult math problems at an unprecedented rate.
- Tao believes AI is depleting the supply of significant open problems faster than new ones can be found.
- AI models from OpenAI and Anthropic have recently solved problems that resisted human mathematicians for decades.
- Tao suggests marking problems as 'analysis-required' to value explained reasoning over just AI-generated solutions.
- This approach aims to sustain the ecosystem for future mathematical progress.
Terence Tao, a renowned mathematician, has raised concerns that artificial intelligence is solving significant mathematical problems at a pace that outstrips the discovery of new ones. Tao argues that the indiscriminate use of AI tools, while capable of solving current challenges, risks depleting the field's supply of valuable open problems that drive future progress.
AI labs, including OpenAI and Anthropic, have recently demonstrated their models' capabilities by solving problems that have stumped human mathematicians for decades. Examples include the 80-year-old Erdős unit-distance conjecture, which was disproven by an OpenAI model and subsequently by Anthropic's Claude Mythos, and a 90-year-old problem cracked by OpenAI shortly after a human researcher published a proof.
Tao's primary concern is that this rapid problem-solving by AI could flatten the 'difficulty landscape' of mathematics, making it harder to identify and pursue research that yields deeper understanding and opens new frontiers. He noted that the very rumor of a problem being worked on can trigger massive AI-powered efforts to solve it, potentially before original research can mature.
To address this, Tao proposes a system where certain problems are designated as 'analysis-required.' This would mean that a correct answer alone would hold little value unless accompanied by detailed reasoning that illuminates the problem's context and potential implications for related areas. He likens this to food banks refusing donations that are merely edible, emphasizing the need for substance beyond a basic solution.
Tao acknowledges that banning AI in mathematics is technically infeasible. His proposed solution, however, has not yet been adopted as policy by any institution, and its implementation faces practical challenges given the current trajectory of AI development.
Discussion