An AI Breakthrough Has Made Decades of Computer Science Research ‘Irrelevant’

Decades of computer science research are now potentially obsolete, requiring many academics to re-evaluate their published work.

OpenAI and Anthropic have independently published proofs, generated with AI assistance, that appear to disprove the Exponential Time Hypothesis (ETH). This hypothesis, a foundational assumption in theoretical computer science, posits that solving certain problems requires exponentially increasing effort as problem size grows. Researchers, including Daniel Lokshtanov and Thore Husfeldt, are now grappling with the implications, as significant portions of their decades of research, built upon the ETH, may become irrelevant or require substantial revision. Both companies have released their findings on GitHub in natural language and machine-checkable Lean formats, and OpenAI is actively soliciting collaboration from computer scientists to verify and understand the proofs. The potential impact extends to areas like cryptography, which relies on the assumed difficulty of certain computational problems.

The disproving of the Exponential Time Hypothesis by AI represents a paradigm shift in theoretical computer science, challenging long-held assumptions about computational complexity. This event signals a potential inflection point where AI-driven discovery can rapidly overturn established scientific consensus, forcing a re-evaluation of entire research subfields. The speed at which these proofs were generated and published, coupled with the difficulty researchers are having in fully digesting them, highlights the growing gap between AI's discovery capabilities and human comprehension speed.

This development compels researchers to adapt to a new era of AI-assisted scientific inquiry, where foundational theories can be rapidly challenged and invalidated. The immediate aftermath involves a race to verify and integrate these new proofs, with significant implications for fields like cryptography that rely on computational hardness assumptions. The long-term effect could be a restructuring of theoretical computer science, with a renewed focus on problems that were previously considered intractable and the development of new algorithms and security protocols.