A Beginning for Mathematics – How AI Will Reshape the Profession

AI Is Already Outpacing Human Arithmetic

The author notes that three years ago AI could not reliably add two numbers, a year ago internal models at OpenAI and DeepMind earned gold‑medal scores on the International Mathematical Olympiad, and today AI systems can autonomously resolve major open problems. This rapid progress implies that superhuman mathematical AI is imminent and will force a radical rethinking of the discipline.

The Core Problem: Production vs. Understanding

Mathematics has traditionally been evaluated by the production of theorems—papers and PhD theses that claim a new result and provide a proof. The essay argues that this operationalization is incomplete because a machine that enumerates proofs or conjectures can generate correct statements without delivering human understanding. The community’s current reliance on theorem‑counting therefore risks losing the deeper value of mathematics: insight, intuition, and the ability to explain ideas.

"The prospect of automating mathematics by enumerating all conjectures, and all proofs of ZFC, is probably not so disturbing to you. … Is there still a need for human mathematicians? I think so. This machine might produce answers we value, but it would not, in itself, produce human understanding of those answers."

What Mathematics Should Aim to Produce

The essay proposes two broad goals:

Produce and understand high‑quality mathematics.
Produce high‑quality mathematicians—people who can internalize, communicate, and extend mathematical ideas. These goals must be pursued even as the production of mathematical text becomes increasingly detached from the understanding of that text.

Institutional Changes Required

1. Rethink the PhD Evaluation Process

Shift from written thesis to rigorous oral defense. The degree would be awarded primarily on the candidate’s ability to explain the topic to examiners until they are convinced of the candidate’s understanding, regardless of whether the results were generated with AI assistance.
Maintain originality of the topic but ignore provenance. Whether a result originates from a human or an AI should not affect the award, as long as the candidate demonstrates deep comprehension.

"I propose the following reconceptualization of the goal of a mathematics PhD: to become a world expert on some interesting, deep topic, and to be able to convey that interest and understanding to others."

2. Prioritize Human‑Centred Skills

Reward talks, seminars, and sustained discussion over papers, because live explanation better evidences understanding.
Interview candidates for graduate admissions just as faculty hires are interviewed, ensuring that communication ability is assessed early.
Foster a robust seminar culture where speakers must satisfy the audience’s comprehension, not merely present polished PDFs.

3. Preserve the Social‑Relational Functions of Mathematics

Keep learning seminars, door‑knocking conversations, and community‑wide curiosity as core activities. These human interactions are the primary mechanism for turning AI‑generated results into shared insight.
Recognize that figuring out what is interesting is a uniquely human contribution that guides AI output toward valuable directions.

Anticipating Objections and Counter‑Points

Concern about validation: Some commenters worry that the sheer volume of AI‑generated proofs will outpace human verification, potentially leading to an “infinite” validation problem.
- The essay acknowledges this risk but argues that the community’s role will shift to understanding and explaining results rather than verifying every line of formal proof.
Fear of losing the “puzzle‑solving” appeal: Critics fear that delegating problem‑solving to AI will make mathematics uninteresting.
- The author counters that human curiosity will persist; the why and what does this teach us will remain the driving forces, even if the how is automated.
Potential bias toward oral performance: Some commenters note that not everyone excels at live presentation.
- The essay suggests a balanced approach: oral defenses complement, not replace, written work, and institutions should provide support for diverse communication styles.

Community Reactions on Hacker News

Support for oral defenses: Users like @wrs and @MarceliusK praised the idea of evaluating candidates through rigorous oral examinations, likening it to code‑review practices in software engineering.
Skepticism about AI‑generated proofs: @vld_chk warned that the validation timeline for AI proofs could become effectively infinite, threatening the reliability of the mathematical record.
Optimism about new opportunities: @Jun8 likened the situation to an ancient Olympiad where new technology (an “exoskeleton”) raises the bar, suggesting we should reward broader contributions such as community building and mentorship.
Calls for broader interdisciplinary discussion: @bonoboTP emphasized that the challenges facing mathematics will echo across many knowledge‑work domains and that coordinated reform is needed.
Emphasis on human motivation: @ksd482 highlighted that AI removes many bottlenecks, leaving motivation as the primary remaining hurdle for progress.

Why This Matters Now

The convergence of cheap, high‑quality AI math generators and the traditional reliance on written output creates a mismatch: papers no longer signal expertise. Without institutional adaptation, the profession risks devaluing human mathematicians and losing the cultural and educational functions that have historically justified mathematical research.

Concrete Steps Forward

Implement oral‑defense‑centric PhD criteria at graduate programs.
Redesign hiring and admissions to include live problem‑solving interviews.
Invest in seminar infrastructure that rewards clear, interactive exposition.
Create incentives for research‑program design—recognize scholars who identify and cultivate interesting problem families, even if AI assists in solving them.
Develop community‑wide validation protocols that blend formal proof checking with human‑led explanation sessions.

Closing Thought

Mathematics has always been “at the beginning.” AI will accelerate the production of results, but the enduring value of the discipline will lie in human understanding, communication, and the collective curiosity that turns isolated theorems into a coherent body of knowledge.