09-25-2026, 11:44 AM
The post “Ways to improve Mathematics (an introduction)” argues that the arrival of AI should prompt mathematicians to reconsider not only how mathematics is done, but what the mathematical community actually values. The author begins from the premise that even extremely capable AI will not eliminate the human desire—or need—to understand mathematics. Proving a theorem and understanding why it is true are different achievements, and the author uses Fermat’s Last Theorem and class field theory to emphasize that a simple statement can conceal extraordinarily deep mathematical structures. Human understanding therefore remains a substantial intellectual undertaking requiring years of study.
The author then argues that AI exposes a growing mismatch between mathematical understanding and the activities traditionally used as proxies for it. Producing proofs, publishing papers, giving seminars and establishing priority have historically been treated as evidence of mathematical achievement, but AI may increasingly be able to perform some of these activities. This creates an opportunity to ask whether the profession should reward understanding itself rather than merely outputs associated with understanding. The author sees this as particularly important because mathematics is already experiencing rapid growth and increasing specialization.
Four systemic problems are identified. First, the journal system is overloaded by the growing number of papers and an increasingly strained refereeing process. Second, seminars often fail to communicate why a problem matters, what the major methods are, or where the field might go next. Third, increasing mathematical specialization has produced insularity, making communication between different areas increasingly difficult. Finally, ego and incentives, particularly the emphasis on being first to prove a result, can conflict with the deeper goal of developing mathematical understanding. The author suggests that AI gives the community a reason to reconsider all four rather than simply attempting to preserve existing practices.
The discussion in the comments extends these ideas into possible future models of mathematical research. One particularly interesting proposal is a large, structured mathematical “wiki” containing results with different levels of verification—such as Lean-verified, human-verified and human-digested mathematics—with journal articles becoming more focused on explaining important developments and giving intellectual direction. Other commenters argue that AI could facilitate “experimental mathematics,” while mathematicians concentrate more on discovering frameworks, connections and important questions. The broader message is that mathematicians should actively participate in shaping the transition rather than simply waiting to see what AI does to the discipline.
Key takeaways
ARTICLE
The author then argues that AI exposes a growing mismatch between mathematical understanding and the activities traditionally used as proxies for it. Producing proofs, publishing papers, giving seminars and establishing priority have historically been treated as evidence of mathematical achievement, but AI may increasingly be able to perform some of these activities. This creates an opportunity to ask whether the profession should reward understanding itself rather than merely outputs associated with understanding. The author sees this as particularly important because mathematics is already experiencing rapid growth and increasing specialization.
Four systemic problems are identified. First, the journal system is overloaded by the growing number of papers and an increasingly strained refereeing process. Second, seminars often fail to communicate why a problem matters, what the major methods are, or where the field might go next. Third, increasing mathematical specialization has produced insularity, making communication between different areas increasingly difficult. Finally, ego and incentives, particularly the emphasis on being first to prove a result, can conflict with the deeper goal of developing mathematical understanding. The author suggests that AI gives the community a reason to reconsider all four rather than simply attempting to preserve existing practices.
The discussion in the comments extends these ideas into possible future models of mathematical research. One particularly interesting proposal is a large, structured mathematical “wiki” containing results with different levels of verification—such as Lean-verified, human-verified and human-digested mathematics—with journal articles becoming more focused on explaining important developments and giving intellectual direction. Other commenters argue that AI could facilitate “experimental mathematics,” while mathematicians concentrate more on discovering frameworks, connections and important questions. The broader message is that mathematicians should actively participate in shaping the transition rather than simply waiting to see what AI does to the discipline.
Key takeaways
- AI may make proving easier without making understanding easier.
- A simple mathematical statement does not necessarily imply a simple explanation.
- Mathematics should distinguish more clearly between producing proofs and understanding proofs.
- The traditional journal model is under increasing pressure from publication volume and refereeing demands.
- Mathematical seminars should spend more time explaining motivation, context, methods and future directions.
- Increasing specialization risks making mathematics more fragmented and internally isolated.
- The community may need to redesign incentives so that understanding and exposition receive greater recognition.
- AI provides an opportunity to fix longstanding weaknesses in mathematical institutions, rather than merely threatening existing practices.
- A future mathematical infrastructure could combine formal verification, human verification and human-readable explanations.
- AI could potentially enable a new form of experimental mathematics, generating and testing enormous numbers of conjectures and constructions.
- Human mathematicians may increasingly focus on connections, frameworks, conceptual explanations and choosing important questions.
- The central question is not simply “Can AI prove mathematics?” but “What kind of mathematical culture do we want when AI can produce proofs?”
ARTICLE
┌────────────────────────────────┐
│ KONSTANTINOS MICHAILIDIS │
└────────────────────────────────┘
│ KONSTANTINOS MICHAILIDIS │
└────────────────────────────────┘