A mathematician asks whether AI brings math a golden age or a dark age

Francesco Maggi, in a guest post on Terence Tao's blog, argues that AI's effect on mathematics is not predetermined. It depends on how institutions choose to build and use these tools now. He treats LLMs as a real breakthrough, especially for generating proofs, while insisting that mathematics is a human cultural artifact that needs more mathematicians, not fewer.

His sharpest concern is an understanding gap. A Lean-verified proof can run to hundreds of thousands of lines, and its natural-language translation is often messy and unfaithful, so a result can be formally correct and still unreadable. That points to what he calls the real darkness: knowledge becoming alien to human understanding. In a dark age, output keeps rising in the short term, but independent human activity weakens and knowledge concentrates among a few people who increasingly cannot follow it. The golden age is the mirror image, where the tools empower a new generation, widen the knowledge people can actually grasp, and keep science open and collaborative. He expects reality to blend the two.

Maggi raises other stakes that practicing researchers will recognize. Authorship, peer review, and citation are unsettled for AI-assisted work, and some researchers may withhold ideas to keep them out of proprietary systems. Frontier companies' compute advantage over universities could concentrate scientific influence even without anyone intending harm. His proposed responses are concrete: broader researcher access to frontier tools, disclosure and reproducibility standards for AI-assisted results, and publicly governed LLM infrastructure.

Why it matters

If you do research with these tools, the risks here are practical and personal: whether your results stay readable to other humans, who gets credit, and whether access concentrates with a handful of labs. The post reads as a short list of standards worth pushing for before the habits harden.