Both OpenAI and Anthropic have presented AI-generated proofs that contest the exponential time hypothesis (ETH), a cornerstone in theoretical computer science. This development could obsolete decades of research, prompting both disruption and excitement in the academic community.
- AI-produced proofs question the exponential time hypothesis.
- Hundreds of prior research papers may require reevaluation.
- Academics express mixed feelings of disruption and opportunity.
What happened
OpenAI and Anthropic released AI-generated proofs disputing the exponential time hypothesis (ETH), which posits that some problems need exponential time to solve as their size increases. These proofs, shared on GitHub in both natural language and machine-checkable formats, challenge a fundamental assumption in theoretical computer science and algorithm design.
The academic community has not yet fully verified or understood these proofs, but major experts including those who initially proposed ETH and those who previously refuted variants of it are collaborating with OpenAI to assess the claims. This effort includes gathering researcher feedback and planning related scientific conferences.
Why it matters
The ETH has underpinned much of the theoretical groundwork in complexity theory for decades. If disproven, large amounts of existing research, including about 50 published papers by some experts, could be invalidated or require significant revisions. This represents an unprecedented shake-up in the field.
While some researchers express distress over their work becoming obsolete, others view this as an exciting scientific breakthrough that opens up new areas of inquiry previously hidden by accepted assumptions. The situation reflects the evolving dynamic between AI capabilities and human-driven scientific exploration.
What to watch next
Verification and validation of the AI-generated proofs by independent researchers is the immediate priority. The outcome will determine the scope of disruption across computer science and mathematics and possibly reshape future research directions.
The broader academic community’s response and adaptation strategies, including how researchers handle the psychological and practical implications of rapid AI advancements, will also be important to monitor. OpenAI’s commitment to funding conferences on this topic could facilitate collaboration and understanding.