OpenAI AI disproves major computer science hypothesis

A 30-year-old mystery solved

OpenAI recently announced that its AI models successfully disproved the Exponential Time Hypothesis (ETH). This hypothesis has been a cornerstone of computer science since 1991, serving as a starting assumption for many fields including computational biology and database design.

The ETH is based on a puzzle called Boolean satisfiability. In this puzzle, a system tries to find a specific setting for a pile of true/false switches that satisfies a list of rules. For decades, scientists believed that as you add more switches, the time required to find the answer would always grow exponentially slower.

By finding a disproof, OpenAI's system suggests that there may be fundamentally faster ways to solve these types of problems. If the hypothesis is indeed false, many previous results that claimed certain tasks "can't be done faster" are now open questions for researchers to explore again.

How the AI verified the logic

To ensure the AI wasn't making a mistake, the results were processed through a program called Lean. Lean functions like a spellchecker for logic, confirming that every single step in a mathematical proof follows correctly from the one before it.

OpenAI's system identified the disproof on October 16, and the Lean program verified the result the following day. This multi-step process helps provide a level of certainty that raw AI generation usually lacks in high-stakes mathematics.

Despite this verification, OpenAI is being transparent about the current state of the research. They describe the findings as raw research data and note that turning this theoretical disproof into actual faster software will require significant additional work.

The human review process

Because the result is so significant, a working group of experts has been formed to confirm the claim. This group includes Russell Impagliazzo, Ramamohan Paturi, and Ryan Williams, who are notable figures in the field.

Notably, Impagliazzo and Paturi are two of the three people who originally proposed the Exponential Time Hypothesis. This means the very authors who built the theory are now tasked with deciding whether a robot has successfully knocked it down.

The experts will need to determine if the AI's logic means exactly what the researchers think it means. This process is expected to take months and may eventually lead to a formal conference to discuss the findings.

Broader scientific impacts

The disproof of the ETH is not the only scientific milestone OpenAI reported during this period. The company also claimed its models produced a rigorous existence proof for quantum chromodynamics, which is the theory describing the strong force.

However, that specific physics result is being handled differently. A group led by Nobel laureate David Gross is currently holding back the physics result while they write their own independent account of the discovery.

These milestones highlight a growing trend of AI being used to navigate complex, multi-dimensional problems in math and physics. They demonstrate the ability of large models to handle rigorous proofs that require deep logical consistency across many steps.

What comes next for research

While the news is a major milestone for AI capabilities, the immediate impact on everyday technology will be gradual. The transition from a theoretical disproof to a faster database or biological model involves a long pipeline of engineering and software development.

Researchers will now have to re-examine hundreds of papers that used the ETH as a starting assumption. This could lead to a new wave of discoveries in how we approach complex computational problems that were previously thought to be unsolvable in a reasonable timeframe.

The coming months will be focused on the peer review of these results. The verdict from the human working group will be the ultimate test of whether AI-generated proofs can be trusted to redefine the boundaries of computer science.