Fundamental Mathematics in the Age of AI -- The Residue, the Journey, and the Ecology

Recent developments indicate that Large Language Models (LLMs) are demonstrating the capacity to solve and refute complex mathematical conjectures and problems, as evidenced by OpenAI's August 2026 announcement. This capability is prompting a re-evaluation of mathematical discovery, though the core argument suggests the focus of concern is misdirected. The true 'product' of mathematics is human understanding and collective sense-making, not merely the quantifiable 'residue' of theorems and proofs that AI can now generate.

Why it matters

The increasing capability of AI to perform complex intellectual tasks traditionally requiring human expertise signals a fundamental shift in how knowledge domains might operate. Understanding the distinction between AI-generated 'residue' and human-driven 'product' is crucial for institutions to strategically leverage AI while preserving and nurturing critical human faculties like understanding, context, and discovery.

Key insights

• Large Language Models (LLMs) are now capable of refuting established conjectures and solving previously intractable mathematical problems.
• The cost of using LLMs for advanced mathematical tasks is reportedly in the range of a few thousand dollars.
• The emergence of AI in advanced mathematics is prompting introspection regarding the future of discovery and generating legitimate anxiety.
• The paper argues that AI primarily addresses the 'countable part' of mathematics (theorems, proofs), which is considered the 'residue' rather than the ultimate 'product'.
• The true 'product' of mathematical work is identified as human understanding and the collective process of deciphering and acting upon the world.
• The process of looking and producing results (residue) and the journey of human understanding are described as two arcs of a single loop.

Source

arXiv — Computers and Society — https://arxiv.org/abs/2608.12816