“If human problem-solving disappears by virtue of the AIs becoming strictly and substantially better at it, then most of the time currently spent by modern mathematical researchers will have to be spent on an activity that is altogether pretty different. Whether such an activity is viable as a professional endeavour is something I am unsure of, but strongly encourage others to think about.” – Jacob Tsimerman – Professor of mathematics at the University of Toronto
The immediate concern raised by advanced AI systems in mathematics is not simply whether they can assist with proofs, but whether they could displace the central activity that defines contemporary mathematical practice: sustained human problem solving. Modern research culture is built around individuals and teams tackling precise questions, deciding whether a statement is true or false, and constructing proofs or counterexamples that other humans can understand and evaluate1. If increasingly capable systems come to dominate this problem-solving layer, the principal daily work of many researchers would be reconfigured, with substantial consequences for career structures, training pathways, and the self-understanding of the discipline.
Problem solving as the organising activity of modern mathematics
Jacob Tsimerman’s remark sits within a longer thread in which he characterises problem solving as an immense and pervasive part of current research practice1. He explicitly foregrounds tasks of the form ‘is T true; if so, prove it; if not, disprove it’ and ‘find an example of S if one exists’ as core to what many mathematicians spend their time on1. This is broadly consistent with philosophical and sociological studies of mathematical work, which emphasise the central role of theorem-proving tasks, conjecture testing, and example construction in structuring research agendas10. The technological turn in mathematics described in recent scholarship hinges on AI systems entering exactly this layer: interactive theorem provers, automated theorem provers, and large language models are all designed to attack formal statements, generate proofs, and explore structural examples10. Against that backdrop, Tsimerman’s concern is not about peripheral automation of routine calculations but about the fate of the defining activity around which careers, prestige hierarchies, and intellectual identities have been organised.
Recent evidence for AI’s growing problem-solving capabilities
In the last few years, empirical results have supported the claim that AI systems are closing the gap with human mathematicians on well-posed problem-solving tasks. Work on neuro-symbolic agents such as AlphaGeometry demonstrates that AI can tackle competition-level geometry problems at success rates comparable to human participants, sometimes finding shorter or more elegant proofs than those previously formalised in proof assistants4,8. Large language model based proving agents, including systems like Aletheia, have achieved correct solutions for a non-trivial fraction of research-level problems when coupled with formal verification backends5,15. Surveys of automated theorem proving and AI-enhanced proof search in systems such as Lean report that machine learning guided tactics can discharge thousands of lemmas at speeds far exceeding human capabilities2,3. Together, these developments underpin the claim, documented in recent essays on mathematicians in the age of AI, that proving power has moved from negligible to competitive with human researchers within a remarkably short time frame5. Although present systems remain error-prone and require significant human supervision, the trajectory substantiates Tsimerman’s scenario as a live possibility rather than speculative science fiction7,11.
Viability of new professional activities once problem solving is automated
Tsimerman’s central worry is not merely technological but professional: if human problem solving becomes less central because AI is strictly and substantially better at it, can mathematicians pivot to a different main activity that still supports a viable research career1. Philosophical analyses of the technological turn suggest several candidate activities, including supervision of AI reasoning, interpretation and conceptual reorganisation of machine-generated proofs, and governance of collaborative proof projects that rely heavily on formal verification platforms3,10. Some authors argue that mathematicians could increasingly act as designers of high-level theories, curators of conjecture landscapes, or critics who assess the mathematical significance of results discovered with AI assistance12,14. However, the empirical literature also highlights difficulties in making these roles professionally robust: AI-generated proofs often lack interpretable structure, hindering the extraction of new concepts and intuition2,3; over-reliance on tools risks erosion of human problem-solving skills, complicating the ability of researchers to maintain independent judgement3,9. In that landscape, the question of whether an alternative activity can anchor a long-term career remains unsettled, justifying Tsimerman’s explicit uncertainty and his call for collective reflection1.
The strategic tension: replacement versus collaboration
Across the research record, there is a clear tension between two narratives. One frames AI as an autonomous problem solver that could perform a large fraction of research-level tasks with minimal human input; another emphasises AI as a collaborator or co-reasoner that extends human capacity while leaving conceptual leadership with people2,4,9. Some technical reports on autonomous mathematics argue that, despite impressive progress, current systems still struggle with reliable reasoning and cannot yet formulate frontier problems unaided15. Practical guides to AI-assisted research go further, claiming that mathematical research is not fully automatable and declining to speculate confidently about whether this will change13. On the other hand, capability conjectures in recent position papers explicitly state that AI tools will reasonably soon be able to perform a reasonable fraction of research-level tasks, under reasonable supervision and cost7. Popular treatments in outlets such as Nature quote leading researchers who expect AI to autonomously contribute at or beyond the level of the greatest mathematicians6. Tsimerman’s statement can be read as occupying a middle position: it neither asserts inevitability nor impossibility but treats substantial AI superiority in problem solving as a contingency worth planning for. The strategic challenge, then, is to design institutional and epistemic frameworks in which collaboration remains meaningful even in the face of potential task-level replacement.
Debates over interpretability, rigour, and mathematical value
One of the most contested aspects of AI-driven mathematics concerns the interpretability and pedagogical value of machine-generated proofs. Studies of automated theorem proving emphasise that, although formal correctness is guaranteed once a proof is verified, the resulting derivations may be opaque, making it difficult for humans to glean insight or build new theory from them2,3. Some commentators worry that a culture of outsourcing detailed proof steps to AI and proof assistants will reduce emphasis on rigour in human exposition, as the nitty-gritty becomes the responsibility of machines12. Others argue that shifting human effort towards explanation, conceptual synthesis, and communication could enrich the discipline, provided that formal verification systems continue to enforce correctness in the background4,12. Empirical findings from education settings suggest both benefits and risks: AI-assisted tutoring can improve problem-solving speed and retention, but excessive reliance may weaken independent proof-construction skills2,9. Within this debate, Tsimerman’s focus on enjoyment and motivation is notable1. There are many mathematicians whose primary satisfaction comes from the struggle and eventual success of problem solving itself1. If that struggle is delegated to AI, then even a formally correct and conceptually rich mathematical ecosystem may lose part of its human appeal, raising questions about the value and purpose of continuing in the profession.
Quantitative models of division of labour and productivity
Several studies attempt to quantify the impact of AI tools on mathematical productivity and to model the evolving division of labour between humans and machines. Surveys of practising mathematicians report high adoption of AI tools and significant perceived gains in productivity, with mean ratings for tool usage and efficiency improvements both above 4.0 on Likert scales3. Experimental work with educational platforms notes improvements of roughly 15\text{-}20\% in problem-solving speed for students using AI-assisted systems, but also decreases in independent proof-writing ability on the order of 15\%2,9. Theoretical analyses treat the research process as a pipeline in which tasks such as conjecture generation, example search, lemma proving, and formal verification can be progressively automated10,11,14. In simple productivity models, if AI systems handle a fraction \lambda of routine proof steps and lemma verification, leaving humans to focus on high-level conceptual work, overall output might scale roughly like (1+\lambda)\times P_h, where P_h denotes baseline human productivity. However, this stylised picture ignores possible nonlinear effects: reductions in human problem-solving practice may diminish the capacity to recognise deep patterns or to evaluate AI proposals critically over time3,9. For Tsimerman, the key point is that even if such models suggest impressive aggregate gains, the structure of individual careers and day-to-day activities would be radically altered once \lambda approaches values near 1.0, where machines dominate the pipeline.
Institutional, ethical, and governance implications
Beyond individual motivation, the prospect of AI-superior problem solving raises institutional and ethical questions. Who receives credit for results when AI systems contribute decisive problem-solving steps, and how should authorship be recorded4,10,14. What mechanisms ensure that reliance on AI does not mask subtle errors, given that current models can hallucinate or mis-specify the problem they purport to solve11,15. How should training programmes adapt, balancing the need for students to acquire classical problem-solving skills with the reality that many routine tasks may be automated in their future careers2,9,13. Governance discussions in recent position papers highlight issues ranging from data provenance and the ethics of scraping mathematical texts for training to the risk that AI systems could centralise expertise in a small number of institutions controlling the largest models11,14. Tsimerman’s call for others to think about and envision alternative professional activities can be interpreted as an ethical request to anticipate such shifts rather than leaving them to market or technological inertia1. Designing norms, credit systems, and evaluation metrics that remain robust under high AI involvement is part of steering the future of mathematics, rather than simply enduring it11.
Why the question matters for the future of mathematical culture
The deeper significance of the statement lies in its focus on culture and vocation. Mathematics is not only a collection of theorems but also a community whose members choose the discipline partly for the distinctive experiences it offers: confronting hard problems, developing personal techniques, and contributing proofs that reflect individual style and insight10,12. If human problem solving becomes marginal because AI tools are strictly and substantially better, that lived experience may change in ways that are not captured by productivity metrics or citation counts. Some researchers may relish new roles as interpreters, curators, or architects of machine-generated theory; others may feel that the field has lost the challenges that drew them to it, rendering the profession less attractive. Existing analyses of mathematics in the age of AI urge mathematicians and philosophers to engage with these questions now, while technology is still in flux7,10,11. Tsimerman’s uncertainty about the viability of a different primary activity is thus not a confession of defeat but an invitation: to chart forms of mathematical work that remain meaningful, intellectually demanding, and professionally sustainable, even if the central locus of problem solving shifts from human minds to artificial ones1,5,6.
References
1. jacob tsimerman on X – 2026-05-04 – https://x.com/Jacob_Tsimerman/status/2051116022585770170
2. [PDF] The Influence of Artificial Intelligence on Mathematics – Ijrpr – https://ijrpr.com/uploads/V6ISSUE5/IJRPR47370.pdf
3. The Impact of Artificial Intelligence on Scientific Research in Mathematics – https://content.scirp.org/pdf/oalib_1115381.pdf
4. The impact of AI on mathematical research – https://www.allscientificjournal.com/assets/archives/2026/vol11issue2/11055.pdf
5. Mathematicians in the Age of AI – Emergent Mind – https://www.emergentmind.com/papers/2603.03684
6. ‘It is incredible’: How AI is transforming mathematics – 2026-05-19 – https://www.nature.com/articles/d41586-026-01553-1
7. Mathematics in the age of AI – 2026-08-24 – https://arxiv.org/html/2608.16753v1
8. Towards the Automatic Mathematician – https://link.springer.com/chapter/10.1007/978-3-030-79876-5_2
9. Synergy of Artificial Intelligence and Mathematics – https://www.irjet.net/archives/V11/i7/IRJET-V11I7116.pdf
10. The Technological Turn in Mathematics – PhilSci-Archive – https://philsci-archive.pitt.edu/28906/1/The%20Technological%20Turn%20in%20Mathematics,%202.04.2026.pdf
11. Shaping the Future of Mathematics in the Age of AI – 2026-03-26 – https://arxiv.org/html/2603.24914v1
12. Mathematical Beauty, Truth and Proof in the Age of AI – 2025-04-30 – https://www.quantamagazine.org/mathematical-beauty-truth-and-proof-in-the-age-of-ai-20250430/
13. A Practical Guide to AI-Assisted Research in Mathematics … – arXiv – https://arxiv.org/html/2603.15914v1
14. How AI is changing the nature of mathematical research – 2026-03-09 – https://www.amazon.science/blog/how-ai-is-changing-the-nature-of-mathematical-research
15. [PDF] Towards Autonomous Mathematics Research – UC Berkeley – https://math.berkeley.edu/~fengt/Aletheia.pdf
