Aura Kasih
A recent announcement by OpenAI regarding a purported solution to one of mathematics’ most esteemed and elusive challenges has sent reverberations through the academic world, not just for its technical achievement, but for the deeply concerning methodologies and ethical questions it has unearthed, potentially reshaping the landscape of scientific collaboration and discovery.
The Navier-Stokes existence and smoothness problem, a cornerstone of fluid dynamics and a Millennium Prize problem offering a substantial monetary award for its resolution, has long been a formidable barrier to human intellect. For nearly a century, mathematicians have grappled with its intricacies, seeking to understand the fundamental principles governing the motion of fluids. OpenAI’s claim, articulated in a recent blog post, details how one of its unreleased artificial intelligence models, leveraging a distributed network of approximately 10,000 AI agents, reportedly arrived at a solution within an astonishing 88 hours. This feat, presented as a monumental milestone in AI’s capacity for abstract reasoning and problem-solving, simultaneously highlights the accelerating pace at which artificial intelligence is encroaching upon disciplines once considered exclusively the domain of human cognition.
However, the circumstances surrounding this breakthrough are far from celebratory, casting a long shadow of controversy over the achievement. The narrative suggests that OpenAI’s intensive focus on the Navier-Stokes problem was not a meticulously planned, long-term research endeavor, but rather a reactive, last-minute surge of computational power triggered by intelligence that other researchers were nearing a solution. This alleged "scoop" has ignited accusations of academic espionage and a disregard for the established norms of scientific discourse, prompting anxieties among researchers about the potential for a chilling effect on the open and collaborative spirit that underpins academic progress.
As articulated by Professor Abhishek Saha of Queen Mary University of London, OpenAI’s approach deviates significantly from the conventional practices that govern the mathematical community. The inherent nature of mathematical research is often characterized by a slow, iterative process of developing ideas, rigorous peer review, and the gradual refinement of proofs. The notion of a powerful entity, alerted to progress, rapidly deploying vast resources to preemptively claim a discovery, fundamentally clashes with this ethos.
The timeline of events is particularly contentious. Just one day prior to OpenAI’s public announcement, Tristan Buckmaster, a mathematics professor at New York University, published findings related to a similar problem, in collaboration with Levent Alpöge, a researcher at Anthropic, a prominent AI competitor. Buckmaster has since detailed a disturbing interaction following his awareness of OpenAI’s burgeoning interest in the problem. He claims to have reached out to OpenAI to inquire about their research timeline and the data underpinning their model’s development. This inquiry, according to Buckmaster, quickly devolved into an adversarial exchange, with an OpenAI representative allegedly issuing a veiled threat: "If you don’t want me to be nice, then I don’t have to be nice." This statement, coupled with a suggestion that Buckmaster should publish his findings and credit OpenAI’s internal model while dropping Alpöge as a co-author, has fueled suspicions of coercive tactics and an attempt to manipulate the scientific record.
Buckmaster’s account further alleges that OpenAI’s responses became increasingly evasive and hostile when he questioned whether his data, specifically his sessions on the Codex platform which he utilized for his research, might have been accessed. OpenAI, in its official blog post and subsequent statements, has vehemently denied the direct use of any specific user data, asserting that "the researchers and the agents did not see any of their work through any means until they released it publicly – in particular, no specific user data was accessed in order to solve this problem." However, the company has not entirely dismissed the possibility of indirect influence, acknowledging that "while unlikely, we cannot rule out that de-identified data derived from their usage of our products helped improve our models." This nuanced admission, while stressing the significant differences between the two proofs, has done little to assuage the concerns of the academic community.
The intricate interplay of research timelines, overlapping investigations, and the opaque nature of AI-generated output makes a definitive determination of precisely what transpired exceptionally challenging. The competitive drive to solve a problem of such profound significance, particularly one associated with a substantial financial reward, is understandable. Yet, specific elements of OpenAI’s narrative raise persistent questions. The company characterizes its effort as a hurried and exceptionally costly undertaking, involving millions of dollars, with the stated objective of "report[ing] on the substantial progress of our AI models" rather than claiming the prize money. Furthermore, evidence suggests that OpenAI’s significant engagement with the Navier-Stokes problem may have commenced relatively recently, around September, with little prior public disclosure of their efforts. This raises the pertinent question: why the sudden and immense urgency?
OpenAI’s explanation for this accelerated pursuit is, in essence, an acknowledgment of opportunism. According to Sébastien Bubeck, an OpenAI researcher, the company initiated its work after encountering "rumors on Twitter" about progress being made on Millennium Prize problems. This prompted the internal question: "We have such a strong model. Why don’t we try to solve also a Millennium Prize problem?" The company later stated that it only subsequently realized these rumors pertained to Alpöge and Buckmaster’s work. While Bubeck has disputed certain aspects of Buckmaster’s account, including the claim that he requested the removal of Alpöge as a co-author, the broader context of a reactive, competition-driven approach remains a point of contention.
Beyond the most explosive allegations of data misuse, the very essence of OpenAI’s acknowledged conduct has unsettled many mathematicians. The perceived race to preempt other researchers is antithetical to the collaborative ethos that defines modern mathematics. While "scooping"—the act of publishing a result before another researcher who is also working on it—does occur, it is typically a difficult and often unintended consequence of independent research efforts. Professor Saha elaborates that the highly specialized nature of cutting-edge research often means that only a select few possess the requisite expertise to rapidly replicate or supersede another’s work, making swift scooping less feasible.
The principle of openness is a deeply ingrained virtue within mathematics. Matthew Ballard, a professor of mathematics at the University of South Carolina and associate director for scientific activities at the Institute for Computer-Aided Reasoning in Mathematics (ICARM), emphasizes that "mathematics depends heavily on an informal norm of trust." He explains that researchers routinely share nascent ideas and ongoing work with colleagues to refine their thinking, operating under the expectation that such exchanges will not devolve into competitive races. The possibility that AI systems might glean insights from researchers’ queries, even inadvertently, is profoundly disquieting. Jeremy Avigad, director of ICARM and a professor at Carnegie Mellon University, notes that "the thought that AI systems might steal ideas from our queries is chilling." Mathematicians have historically felt secure discussing their work openly, free from the fear of their ideas being co-opted. The advent of AI, capable of potentially "setting a swarm of agents on solving the problem" based on the slightest hint, is poised to foster a more guarded research environment, a prospect that many find disheartening.
OpenAI’s inability or unwillingness to definitively confirm whether its models were influenced by the work of other mathematicians exacerbates the unease. Brendan Hassett, a professor at Brown University, views this ambiguity as problematic, noting that "given the history of the AI companies appropriating copyrighted work without permission or payment, it is natural for people to ask these questions." He argues that such companies should be held accountable for providing verifiable assurances that user interactions will not be leveraged to enhance their models, and this accountability should extend to demonstrable proof.
The long-term implications of this incident remain uncertain. Yang-Hui He, a fellow at the London Institute for Mathematical Sciences, expresses concern that the pursuit of mathematics by large technology companies may lead to increased secrecy. While generally optimistic about AI, he fears a regression to a more clandestine research environment, reminiscent of historical periods funded by aristocratic patronage, where discoveries were not readily shared.
For the majority of researchers, the immediate impact may be limited. Professor Saha suggests that major AI labs are unlikely to dedicate vast resources to every mathematical problem, as the publicity generated might not justify the immense investment. This suggests that the current scenario, driven by the allure of a high-profile Millennium Prize problem, might be an outlier rather than a harbinger of a wholesale shift in research priorities.
The pursuit of publicity appears to have been a significant motivator for OpenAI, potentially explaining their decision to target a problem known to be under investigation by a researcher at a rival organization, even if that researcher was acting independently. Oxford professor Andras Juhasz characterizes the event as "clearly a PR victory for OpenAI." However, he questions the sustainability of this approach, pondering whether it signals a potential decline in AI companies’ engagement with fundamental research mathematics, now that their models are demonstrating the capacity to tackle such complex problems. Human mathematicians engage in scooping as well, Juhasz concedes, but OpenAI’s actions have demonstrated that this can occur on an unprecedented scale. The sudden deployment of "10,000 mathematicians" onto a problem signifies a new paradigm of competitive research.
Ultimately, this apparent public relations triumph for OpenAI could prove to be a Pyrrhic victory. The company has once again showcased the advanced capabilities of its models at the vanguard of mathematical discovery. However, in doing so, it appears to have alienated the very academic community it has sought to impress, potentially fostering an atmosphere of suspicion and eroding the foundational trust upon which scientific progress is built. The future of AI’s involvement in mathematics hinges on its ability to navigate these ethical complexities and re-establish a collaborative, transparent relationship with the global research community.
Lisa Mariana
Ayu Aulia
Safa Marwah