AI Proves Exponential Time Hypothesis False, Rendering Decades of Computer Science Research Obsolete

In a seismic shift for theoretical computer science, leading AI firms OpenAI and Anthropic have claimed to have disproved the Exponential Time Hypothesis (ETH), a foundational premise that has underpinned vast swaths of algorithm design and complexity research for nearly three decades. This breakthrough, which suggests that solving certain problems does not necessarily require exponentially more time as their size increases, has effectively rendered significant portions of existing academic work obsolete. Peter Jonsson, a professor at Linköping University, described his upcoming conference presentation as 'dead' upon hearing the news, noting that half of his paper’s arguments no longer hold water. This event marks a pivotal moment where artificial intelligence has not merely assisted in research but has fundamentally overturned long-standing scientific consensus. The implications of this discovery are profound and immediate. Researchers such as Daniel Lokshtanov from UC Santa Barbara estimate that approximately 50 of his published papers may now be considered 'pointless' or in need of a complete revisit. Lokshtanov characterized the development as a 'giant bomb dropped into the middle of the field,' acknowledging the dual nature of the shock: it is both intellectually thrilling and professionally devastating. Similarly, Thore Husfeldt, head of the algorithms research group at the IT University of Copenhagen, who had formulated an earlier version of the hypothesis, stated that much of his recent work is now 'irrelevant.' The speed at which these AI-generated proofs have emerged has outpaced the community’s ability to independently verify them, creating a vacuum of certainty in a field that relies heavily on rigorous peer review. Despite the initial turmoil, many researchers are embracing the opportunity to re-evaluate their assumptions. Husfeldt expressed enthusiasm, comparing the situation to someone illuminating a dark room with floodlights, allowing scientists to explore previously inaccessible corners of computation. Victor Lagerkvist, an associate professor at Linköping, plans to revisit problems he had previously abandoned under the assumption that ETH was accurate. The AI-generated proofs, published on GitHub in both natural language and the machine-checkable Lean format, have sparked a global effort to unpack their validity. OpenAI has engaged key academics, including those who originally proposed the hypothesis, to help verify the findings, signaling a new era of human-AI collaborative verification. However, the rapid pace of AI-driven discovery has introduced new challenges. The sheer volume of content, including what some describe as 'slop-level' results, has overwhelmed researchers. OpenAI’s recent release of 722 manuscripts on 400 problems has been criticized as a hasty rollout, potentially motivated by commercial interests rather than pure scientific advancement. This has led to psychological strain within the academic community, with some departments considering hiring occupational psychologists to support staff dealing with the anxiety of their work being rapidly obsoleted. The incident highlights a growing tension between the speed of AI innovation and the slower, more deliberate pace of traditional scientific validation. Beyond theoretical implications, the falsification of ETH raises concerns for cryptographic security, which relies on the assumption that certain mathematical problems are computationally intractable. While this specific result does not directly break existing cryptosystems, it serves as a warning to designers that foundational assumptions about computational complexity may be flawed. As the community scrambles to understand the full scope of this 'once-in-a-generation result,' the focus shifts to how AI can be integrated into the scientific process without compromising the integrity of peer review. The coming months will be critical in determining whether this breakthrough leads to a more robust understanding of computation or further destabilizes the field.