OpenAI and Anthropic have released AI-derived proofs suggesting that the Exponential Time Hypothesis (ETH), a foundational assumption in computational complexity theory, may not hold true, prompting debate and collaboration across the scientific community.
- OpenAI and Anthropic AI models propose solutions refuting ETH.
- MIT and top researchers collaborate to evaluate and discuss implications.
- Potential shift in understanding of P versus NP and algorithm efficiency.
What happened
OpenAI released AI-generated solutions to three major mathematical problems, including a surprising refutation of the Exponential Time Hypothesis (ETH). This hypothesis has been a critical component of computational complexity theory, asserting that some problems require exponentially growing time as inputs scale. Shortly after, Anthropic released its own AI-based counterclaim, further challenging ETH.
To help interpret these developments, OpenAI enlisted prominent computer scientists such as MIT professor Ryan Williams, alongside other experts who helped organize feedback and collaborative working groups. These groups aim to explore the validity of the new proofs, discuss academic norms around AI-driven contributions, and prepare the community for further analysis and conferences.
Why it matters
ETH has influenced how scientists understand computational difficulty and the feasibility of solving complex problems quickly. It is closely related to the famous P versus NP question, which examines whether problems whose solutions can be verified quickly can also be solved quickly. Most researchers have believed P ≠ NP, with ETH providing additional supporting evidence.
The AI labs’ claims suggest new algorithms that solve certain complex problems in sub-exponential time, effectively calling ETH into question. If this finding holds, it forces a reevaluation of long-standing assumptions about algorithmic limits and problem hardness, potentially reshaping fields like cryptography which rely on these theoretical guarantees.
What to watch next
The scientific community will be closely watching ongoing peer reviews, working group investigations, and possible refinements or extensions of the AI-derived proofs. Conferences facilitated or funded by the labs are expected to play a key role in fostering detailed examination and discussion of these revolutionary claims.
Long term, the outcome of these developments may influence the direction of AI research, computational complexity theory, and practical algorithm design. Researchers will scrutinize whether these AI-discovered algorithms represent optimal solutions or if further improvements can push the boundaries of what is computationally feasible.