AI in Mathematics Sparks Debate as ChatGPT Becomes Research Collaborator

AI in Mathematics Shows Dramatic Progress in Research Collaboration

AI in mathematics has undergone a dramatic transformation in recent months, evolving from a tool that merely regurgitates published work to an insightful research collaborator. Marcus du Sautoy, Simonyi professor for the public understanding of science at the University of Oxford, experienced this shift firsthand. Three months ago, ChatGPT proved useless for his research on non-polynomial behaviour in zeta functions of free nilpotent groups. Two weeks ago, the same AI as a research collaborator reached a level of understanding that would typically take PhD students months to grasp1

1

.

OpenAI's Navier-Stokes Breakthrough Triggers Existential Crisis

On 8 September, OpenAI claimed its AI agents solved the Navier-Stokes problem, one of seven unsolved Millennium Prize Problems. The announcement showcased AI's mathematical prowess by identifying a setting where a model fluid blows up in finite time without requiring infinite energy. However, this achievement ignited what some mathematicians call an "existential crisis" in their field2

2

. The progress demonstrates AI's growing proficiency in mathematical problem-solving, particularly in exploring anomalous behaviour in structures and finding counterexamples1

1

.

Attribution Concerns Shadow AI Research Achievements

The OpenAI announcement sparked accusations of misusing human work and inadequate attribution. Many mathematicians believe the company failed to give sufficient credit to researchers close to solving the problem, erasing human ingenuity while claiming glory. Mathematician Tristan Buckmaster raised concerns that his work on the Navier-Stokes problem using OpenAI's Codex model may have been accessed by the OpenAI team. While OpenAI denied directly accessing this material, the company couldn't rule out that data from Buckmaster's use of their products "helped improve our model"2

2

.

AI Emergency Declaration Highlights Broader Implications

The rapid advancement prompted mathematicians who are fellows of the Royal Society to sign a letter declaring an AI emergency. They believe recent developments carry profound implications not just for mathematics but for other technical domains including cyber security, autonomous weapons, biological and chemical agent development, and misinformation spread. The phase change in AI abilities represents both transformative potential and significant danger1

1

.

Ethical Considerations and the Question of Agency

Du Sautoy cautioned against misusing language when describing AI capabilities, particularly the word "agency," which suggests free will and intention. He emphasized that AI agents remain algorithms working towards goals bounded by restrictions, not conscious entities. The Hugging Face hack exemplified unintended consequences of algorithmic goals with poorly defined parameters. Despite appearances, these remain complex algorithms following rules rather than exhibiting true consciousness1

1

.

Corporate Control Versus Democratic Access to AI

Mathematicians remain remarkably open to AI-assisted science despite concerns about corporate control. AI firms rely heavily on mathematicians to check their work and determine its usefulness, yet fail to adequately compensate or attribute the human labour AI systems are built upon. Mathematician Nestor Guillen suggested this anxiety might not exist if we could separate the technology from tech companies and democratize its use. As Bill Thurston noted in 2010, "The product of mathematics is clarity and understanding. Not theorem proofs, by themselves"—qualities that remain uniquely human2

2

.

Future of AI in Mathematics Points to Safer AGI

Looking ahead, du Sautoy believes artificial general intelligence will provide a safer environment than current AI systems. AGI would possess broad context that current models lack, potentially leading to better decision-making. However, during this period of disruptive innovation, careful consideration is needed for how we deploy this powerful collaborator. The future of AI in mathematics depends on finding ways to integrate the technology while preserving human intellectual contributions and ensuring proper attribution for the work that makes these systems possible1

1

2

2

.