AI Is Disrupting Our Proxies for Expertise

Summary

The article argues that AI’s growing success on prestigious mathematical problems is exposing a weakness in how mathematics measures progress and rewards expertise. It discusses a declaration signed by almost 5,000 mathematicians, including 25 Fields Medalists, which warns that solving problems is only a proxy for the deeper goal of conceptual understanding and insight. The author distinguishes between puzzle-solving, which is legible and prestigious, and idea-generating, which creates concepts whose value may take years to assess. Traditionally, difficult puzzles helped validate new ideas and made mathematical skill visible to outsiders; AI may now solve some targets “the hard way,” without producing intuitive concepts that advance human understanding. The author says it remains unclear whether frontier models can generate genuinely new mathematical ideas, and argues that humans may still need to develop accessible human proofs alongside AI-generated proofs. Chess and tool-assisted speedrunning suggest a possible future in which human and AI mathematics occupy separate spheres while AI also improves human performance. The article applies the same concern to software engineering, where AI-generated code may make GitHub projects and high-volume coding less reliable as evidence of skill. It concludes that both fields will need new, legible ways to value human contribution or a cultural separation between AI and human work.