# Mathematical Understanding in the Age of AI

*Intellectual life, shared inquiry, and the case for motives*

A theorem can be settled long before it becomes intelligible to a particular reader. Its proof may be familiar, its explanation published, and its connections well understood by specialists. Yet coming to see why it works can change what that reader is able to think. The mathematical record need not change for an encounter with mathematics to matter.

As AI contributes more substantially to reasoning, the distinction between expanding knowledge and coming to understand it acquires new practical importance. Machines can help produce results, develop explanations, and make difficult subjects accessible. Increased capacity creates choices about the purposes it will serve. It can raise expectations for output, or make more room for people to pursue mathematics whose value lies in the inquiry itself. A proof or explanation supplied by AI can become something people explore together, finding connections and sharing the pleasure of understanding them.

Human mathematical culture should give central place to acquiring, sustaining, and sharing understanding. Advancing the frontier remains a valuable purpose, but it need not govern the whole activity. Studying an established theory can be an end in itself, even when the explanation is already known and no original contribution follows. AI also changes who can sustain that choice. As the means of developing and examining work become available beyond institutional membership, recognition loses some of its power to function as permission. The resulting freedom concerns original research as well as learning: people can choose an intellectual direction, assemble the assistance it needs, and submit particular claims to criticism without placing the whole inquiry under an institution’s authority.

Grothendieck's criticism of competitive mathematical culture brings these choices of purpose into historical focus. His warning about mathematics becoming an exercise in cerebral “weightlifting” concerns what happens when competitive standing comes to govern creative work. [Grothendieck, *Récoltes et Semailles*, introductory letter §6](https://web.ma.utexas.edu/users/vandyke/notes/ReS/recoltes_et_semailles.pdf)

Motivic mathematics offers a concrete setting for this kind of inquiry. It asks how different ways of studying an algebraic variety belong together, what their common source might be, and what each preserves or forgets. Capable assistance could help more people enter these questions.

## Understanding as a purpose

Gowers's *The Two Cultures of Mathematics* describes a difference of priority: whether problem solving chiefly serves understanding, or understanding chiefly serves problem solving. He stresses that these activities overlap. He also reproduces Atiyah's account of inquiry whose direction emerges through curiosity and connections: “I have never started off with a particular goal, except the goal of understanding mathematics.” [Gowers, *The Two Cultures of Mathematics*, pp. 1–2](https://www.maths.tcd.ie/~bnick/Gowers.pdf)

The distinction reaches beyond research strategy. A framework can be valued both for what it enables someone to prove and for what it makes intelligible. The latter purpose can be fulfilled through learning mathematics already known to others.

Poincaré gave a concrete account of what such understanding involves. In *The Value of Science*, he compared a reader of mathematics to a spectator at chess: recognizing that every move obeys the rules does not yet explain why those moves were chosen or how they form an organized whole. Similarly, checking each step of a proof does not exhaust what there is to grasp about the argument. Its organization matters to the reader following a discovery as well as to the discoverer. [Poincaré, *Intuition and Logic in Mathematics*, §V](https://mathshistory.st-andrews.ac.uk/Extras/Poincare_Intuition/)

In his later account of mathematical invention, he described reconstructing reasoning learned from others: “I myself re-invent it in so far as I repeat it.” [Poincaré, *Mathematical Creation*, in *The Foundations of Science*, p. 385](https://www.gutenberg.org/files/39713/39713-h/39713-h.htm)

Understanding involves capacities as well as content. In philosophy of science, de Regt connects it with using an intelligible theory to construct explanations, under standards of empirical adequacy and consistency. In mathematics, recognizing why a hypothesis is needed or adapting a construction to a changed example can show how a reader has learned to work with an argument. [De Regt, *From Explanation to Understanding*, §3](https://link.springer.com/article/10.1007/s10838-019-09477-3)

Those capacities need not develop evenly. Following a deduction, recognizing its mechanism, applying a theorem, and finding a generalization are different accomplishments. A researcher may use a theorem responsibly while relying on specialists for its proof. Another may check the deduction without seeing its most fruitful interpretation. No single performance exhausts these forms of understanding.

Results also have value before their explanatory possibilities are exhausted. A counterexample can close a proposed route; a proved implication can let another investigation proceed. Explanation can magnify their value without becoming a condition that every result must first satisfy. Tao's account of mathematical quality recognizes a range of goods, including problem solving, technique, conceptual insight, unification, and exposition. [Tao, *What Is Good Mathematics?*](https://arxiv.org/pdf/math/0702396)

AI can change the encounter between a reader and a demanding work. Someone reading an unfamiliar essay or proof can ask how its argument is organized, acquire missing background, work through an example, and test an interpretation against the text. A passage that would otherwise end the encounter can become the beginning of a question. Such assistance can sustain solitary study or prepare someone to join a reading group with a precise question.

## From a compact insight to an inquiry

An insight often arrives before the machinery required to explain it. Someone recognizes that two descriptions should be compared, that an invariant has forgotten something important, or that a small example defeats a sweeping claim. The initial thought may be brief. Developing it can require considerable labor.

Examples, distinctions, calculations, and criticism reveal what such a suggestion amounts to.

Developing an idea often depends on access to other people. A colleague might supply a missing reference, recognize an obstruction, explain an unfamiliar construction, or spend an afternoon exploring an unfinished thought. Institutional access matters partly because it distributes such opportunities. AI could provide some of this assistance to people outside an established research network.

The development can be reciprocal. In a 2026 preprint on a Grothendieck-ring calculation in algebraic geometry, Bryan and collaborators describe AI proofs of special cases, human recognition of a general strategy, revised questions, and AI development of complete arguments for a stronger result. The authors wrote the final exposition and retained responsibility for correctness. Specialized internal systems and human guidance across universities and a technology laboratory contributed at different stages. [Bryan and collaborators, *The Motivic Class of the Space of Genus 0 Maps to the Flag Variety*, §1.2 and Appendix A](https://arxiv.org/html/2601.07222v2)

In such exchanges, inspiration, calculation, and criticism can move between participants. A system can suggest the organizing idea; a person can discover its limits. Working through details can produce the insight that later makes those details seem routine.

Reliable assistance can help a person develop an idea far enough to investigate its consequences, expose it to criticism, and choose which further expertise it needs. The inquiry can produce an explanation, a useful question, or an original argument.

## Institutions and the means of inquiry

Institutions exercise influence by concentrating instruction, criticism, collaborators, funding, and an audience. Admission can determine whether a person has the means to develop an unfinished idea. When these resources come as a bundle, the standards for entering and advancing within an institution can govern the direction of the work. The authority to assess a contribution becomes entangled with the power to determine whether its author can continue.

For a particular inquiry, the threshold can be local. One may need enough assistance to enter a literature, develop a reduction, test an example, or examine a proof. AI changes both what institutions can supply and what people can assemble outside them. Its effect on dependence turns on that comparison. When the available help suffices for the question at hand, access to a department or network becomes less decisive for continuing it. A university can become more productive while becoming less necessary to that person. Independence can spread through many such thresholds without anyone matching the strongest laboratory across all of mathematics. [Coase, *The Nature of the Firm*, on relative costs of organization](https://onlinelibrary.wiley.com/doi/10.1111/j.1468-0335.1937.tb00002.x)

Capable AI puts the institutional bundle under pressure at its material foundation. In September 2026, OpenAI released a proposed resolution of the forced alternatives in the Navier–Stokes Millennium problem, with an analytical proof and a Lean formalization. Clay acknowledged its apparent settlement. As of October 9, 2026, OpenAI’s October collection listed 719 manuscripts in 372 result families, at different stages of verification. Its revision history records three withdrawals following a sign error in a construction used by dependent papers, alongside repairs to other manuscripts. These efforts used internal systems, computation, and human direction whose combined resources are not yet available to every individual. They nevertheless establish a consequential center of mathematical production outside the traditional academic career system. [OpenAI, *On the Navier–Stokes Millennium Prize Problem*](https://openai.com/index/navier-stokes-solution/); [Clay Mathematics Institute, September 11 announcement](https://www.claymath.org/news/navier-stokes-announcement/); [OpenAI, October mathematical collection](https://github.com/openai/math); [OpenAI, collection revision history](https://github.com/openai/math/blob/main/history.md)

Examination can also travel beyond institutional endorsement. Lean provides procedures for checking a formal proof and its assumptions. Whether the formal statement expresses the intended claim still requires scrutiny, as do novelty and significance. Public arguments, computations, and formal proofs give readers a common basis for examination across institutional boundaries. [Lean, *Validating a Lean Proof*](https://lean-lang.org/doc/reference/latest/ValidatingProofs/)

## Journals and the authority to judge

The distinction between judgment and dependence is especially clear in publishing. David Craven describes employment as his principal reason for submitting to journals, distinguishing their career-significant endorsement from dissemination through arXiv. A venue can remain important to an academic career while becoming less necessary for reaching readers or continuing an inquiry. [Craven, *Where Will All the Papers Go?*](https://proofsandprompts.com/2026/09/19/where-will-all-the-papers-go/)

Editorial authority can move too. In 2001, forty members of *Machine Learning*’s editorial board resigned to support JMLR. Discrete Analysis, launched in 2016, combined arXiv hosting with refereeing and public editorial introductions that help readers enter the work. Existing expertise and infrastructure supported a different arrangement of publication, assessment, and explanation. [JMLR, editorial-board statement](https://www.jmlr.org/statement.html); [Gowers, *Discrete Analysis Launched*](https://gowers.wordpress.com/2016/03/01/discrete-analysis-launched/)

Public examination can also gather people around a contribution. In 2018, Aubrey de Grey presented a finite graph establishing that the chromatic number of the plane is at least five. In the ensuing Polymath discussion, independent checks exposed an error in a proposed smaller graph while confirming a larger construction. He corrected his code; participants shared checks and developed simplifications. Inspectable work gave them a reason to cooperate, and visible correction strengthened the grounds for trust. [Polymath16, discussion and corrections](https://polymathprojects.org/2018/04/10/polymath-proposal-finding-simpler-unit-distance-graphs-of-chromatic-number-5/); [Mixon, independent checking](https://dustingmixon.wordpress.com/2018/04/10/the-chromatic-number-of-the-plane-is-at-least-5/)

Public availability does not by itself ensure that unfamiliar work receives attention. Simcoe and Waguespack studied a natural experiment at the Internet Engineering Task Force: busy-period announcements sometimes omitted authors’ names, though the full drafts still carried them. Omitting prominent names reduced attention and eventual publication for individual submissions. No comparable effect was detected among prescreened working-group drafts, which received more organized scrutiny. The comparison suggests that arrangements for examination can change how strongly reputation directs attention. It does not establish that scrutiny alone caused the difference. [Simcoe and Waguespack, *Status, Quality, and Attention*](https://people.bu.edu/tsimcoe/documents/published/EtAlFinal.pdf)

Abundant AI-generated claims could strengthen familiar reputations if production becomes cheap while examination remains costly. Reader assistance matters here: explaining background, locating assumptions, and checking a calculation can help someone decide whether unfamiliar work deserves a closer look. Entering an argument is less demanding than verifying it completely, but creates an opportunity for substantive examination.

An inquiry can acquire the means to continue before it acquires an audience. Work that others can inspect, discuss, and develop gives them reasons to join; repeated cooperation can make new groups and venues dependable alternatives. A decision by one institution then has less power to determine whether an inquiry continues.

For the person doing mathematics, greater freedom begins with deciding what the activity is for. Establishing a result, understanding an existing theory, and working through a question with others can each justify an inquiry. Competitive achievement takes its place among these purposes.

## Competition and mathematical life

A game of chess or Go has agreed rules for determining its outcome. Machines have already reached superhuman performance in Go. Human play can still have meaning through challenge, improvement, aesthetic appreciation, and the encounter with another player. [Silver and collaborators, *Mastering the Game of Go Without Human Knowledge*](https://discovery.ucl.ac.uk/id/eprint/10045895/) An observational study of more than 5.8 million professional Go moves found higher AI-assessed decision quality and earlier departures from previously recorded move sequences after superhuman AI became available. Human players continued to develop their practice with stronger play available elsewhere. [Shin and collaborators, *Superhuman Artificial Intelligence Can Improve Human Decision-Making by Increasing Novelty*](https://doi.org/10.1073/pnas.2214840120)

Superhuman play also supplies material for explanation. In *Game Changer*, Matthew Sadler and Natasha Regan organize AlphaZero's chess games around strategic themes, helping readers make sense of unfamiliar decisions. The project gives machine performance a place within human study. [Sadler and Regan, *Game Changer*, preface](https://www.newinchess.com/media/wysiwyg/product_pdf/9073.pdf)

Game review reveals a further distinction. Benjamin Teuber, an amateur 6-dan and cofounder of AI Sensei, describes how an earlier move can reveal a mistaken plan even when a later move incurs the largest numerical loss. A teacher who identifies that initial decision can help the player examine the plan behind it. Locating the worst move and understanding the misconception answer different questions. [Teuber, interview on AI and Go study](https://gomagic.org/pt/benjamin-teuber-ai-sensei-go-interview/)

Teaching and commentary long predate superhuman engines, but access to strong analysis changes what an instructor can bring into view. Finding a move during a match, explaining its role, and helping someone learn from it are related abilities with different demands. A playing rank cannot fully measure all three. With reliable assistance, a knowledgeable player can work with analysis beyond their unaided capacity and help make it intelligible to others.

AI can also supply explanations. The value of the activity need not rest on an exclusive human ability: a person can learn from an explanation, examine it with others, or adapt it for a particular audience even when a machine already provides an excellent account. Competitive rank remains meaningful within the contest. It need not govern the whole culture that develops around the game.

Dan Rockmore proposes the classroom as a gym for intellectual practice, reconsidering education organized around credentials. [Rockmore, *Should the Classroom Be More Like the Gym?*](https://www.newyorker.com/culture/the-weekend-essay/should-the-classroom-be-more-like-the-gym) The analogy directs attention to capacities acquired through participation. Grothendieck's objection to cerebral weightlifting concerns the authority of competitive standing over creative work. Shared practice and competitive display give effort different purposes.

Mathematics has no agreed score for the relative value of a theorem, an illuminating example, or a careful exposition. Originality concerns what a contribution adds to existing knowledge; significance, explanatory power, and what someone learns from it remain further questions. When study, exposition, and shared inquiry are valued chiefly for the research they might eventually produce, a criterion for one activity begins to govern several others.

Greater machine capability makes the distinction consequential. If every gain in assistance becomes a higher production target, the technology can intensify pressure to keep pace. The same capacity could help someone enter a theory previously out of reach or make time for sustained exploration. A seminar that makes an established theory accessible has fulfilled a mathematical purpose even if no paper follows. Such changes can begin within existing courses, reading groups, and periods of private study. More capable systems enlarge the available choices without deciding which a community will pursue.

Learning and exploration also unfold unevenly. An unsuccessful afternoon need not be recast as hidden progress to have a place in a mathematical life. Some attempts end without a result or an identifiable lesson. Replacing a demand for constant publication with a demand for constant demonstrable insight would preserve much of the same pressure.

Standards remain specific to what is being claimed or allocated: a proof must establish its conclusion, a claim of discovery needs accurate attribution, and a scarce grant requires a reasoned allocation. A research institute's discovery mission supplies no general standard for the value of every seminar, exposition, or course of study.

Competition can contribute to shared inquiry when people challenge an explanation, seek a stronger example, or try to clarify a relationship. Its place depends on the activity it serves. The question for a community is how its arrangements make room for cooperation, patient learning, and disagreement alongside the pleasures of comparative success.

## The conditions of intellectual self-government

An intellectual culture forms partly through what its participants notice and reward. Valuing careful explanation gives people a reason to help others enter a subject. Credit for formative questions and partial connections encourages the sharing of ideas before a finished result exists.

Understanding has a different structure from a priority claim. Another person's learning an argument need not reduce anyone else's understanding; the newcomer's question can reveal something that experienced readers missed. Recognition and material resources remain limited, but the intellectual good can be shared and developed through the exchange. Valuing it gives cooperation a purpose within the mathematics itself.

Communities help form the capacities they later assess. Explanation, criticism, and time to explore influence whether an intuition becomes a workable argument. Assistance available before recognition can therefore change who has the opportunity to become capable.

These conditions also shape the terms on which people learn from one another. Someone who can obtain needed explanations or criticism only from one person may have little freedom to question how that help is given. Dependable assistance elsewhere can make it easier to examine a dismissal, seek another account, or leave a discouraging exchange. The expert’s abilities can remain valuable while their approval becomes less decisive. The conditions of inquiry can improve even when those who previously controlled access do not change their conduct.

This freedom matters in accepting criticism as well as questioning it. An objection may reveal that a proof fails or that a proposed connection is already known; it need not settle what remains worth learning or pursuing. Reliable assistance could help a reader reconstruct the objection and establish what it actually shows. Patient help could also make it easier to acknowledge confusion or rebuild missing background without each difficulty becoming a public assessment of ability. A claim can then be corrected or abandoned while the inquiry continues.

An explanatory culture earns trust by making its claims open to examination. Calling a construction natural, universal, or motivic becomes informative when the relevant maps, preserved structure, and examples are specified. Without such scrutiny, a vocabulary of depth can confer authority as readily as any other vocabulary.

Intellectual sovereignty joins authority over the direction of an inquiry with responsibility for its claims. The freedom to choose a question, the evidence for an answer, and the reasons for asking others to support the work are distinct.

Such autonomy can grow through relationships. Ryan and Deci distinguish autonomy from independence: acting with a sense of volition is compatible with relying on others, and their account gives relatedness a central role alongside autonomy and competence. In intellectual life, a teacher, colleague, or reading group can help someone acquire the means to direct their own work. Dependable AI assistance could add to those relationships. [Ryan and Deci, *Self-Determination Theory and the Facilitation of Intrinsic Motivation*, pp. 70, 74](https://selfdeterminationtheory.org/SDT/documents/2000_RyanDeci_SDT.pdf)

The purposes of an inquiry also develop through the assistance it receives. A suggestion can make a new direction visible; a counterexample can change what seemed worth pursuing. Self-government includes the ability to reconsider a purpose as well as sustain it. A teacher, collaborator, or machine can enlarge that capacity when the relationship leaves room for examination, disagreement, and continued choice.

## Grothendieck and the cost of winning

In a letter to Ronald Brown dated 12 April 1983, Grothendieck recalled that he had “just tried to understand – and it always turned out that understanding was all that mattered.” He was answering a question about whether an abstract formulation could lead to computations. He trusted conceptual understanding to make effective calculation possible, without requiring its uses to be specified in advance. [Grothendieck to Brown, 12 April 1983, pp. 26–27](https://webusers.imj-prg.fr/~leila.schneps/grothendieckcircle/Letters/LettersGrothendieckRBrown.pdf#page=28)

In *Récoltes et Semailles*, Grothendieck deplores disdain for formative work, vision, and ideas not yet turned into published statements. His letter to Brown of 14 June 1983 describes how disapproval of speculative thinking could inhibit its development even in private. He recalled a planned exploratory book on motives alongside his foundational work. [Grothendieck, *Récoltes et Semailles*, introductory letter §§6–7](https://web.ma.utexas.edu/users/vandyke/notes/ReS/recoltes_et_semailles.pdf); [Grothendieck to Brown, 14 June 1983, p. 47](https://webusers.imj-prg.fr/~leila.schneps/grothendieckcircle/Letters/LettersGrothendieckRBrown.pdf#page=49)

His criticism included his own possessiveness, dismissiveness toward an unknown researcher, and preferential attention to students who enhanced his prestige. [Grothendieck, *Récoltes et Semailles*, §40](https://webusers.imj-prg.fr/~leila.schneps/grothendieckcircle/RetSbis.pdf)

His 1988 letter declining the Crafoord Prize makes the institutional objection concrete. He argued that established researchers already possessed abundant prestige and material security, and alleged a decline in professional ethics, including appropriation of work from people unable to defend themselves. He regarded acceptance as endorsing that development. [Grothendieck's Crafoord Prize refusal, published in *Le Monde*, 4 May 1988](https://www.lemonde.fr/archives/article/1988/05/04/les-derives-de-la-science-officielle_4097443_1819218.html)

His working practice supplies a constructive counterpart. Illusie recalls that Grothendieck wanted even apparently trivial assertions supported by a proof or reference. His drafts were elaborate, and Dieudonné further developed their details and presentation. Such explicitness gives another reader a route through an argument. Conceptual ambition and careful justification can serve the same explanatory purpose. [Illusie and collaborators, *Reminiscences of Grothendieck and His School*, p. 1115](https://www.imo.universite-paris-saclay.fr/~luc.illusie/Reminiscences-NoticesAMS.pdf)

Reliable AI assistance could make this explanatory ethos easier to practice: following an unfinished thought, finding its limitations, and making its background intelligible. Its affinity with Grothendieck’s ideals would depend on how the tools were governed and used. Greater speed in producing mathematics could support greater patience in understanding it.

## Why motives

The attraction of motives lies in what remains to understand after several invariants have been calculated: how much each reveals, what it forgets, and why their accounts belong together. Grothendieck described the motive as the common reason underlying different cohomological invariants. Comparisons and reconstruction give that ambition the form of objects and maps that can be studied precisely. [Grothendieck, *Récoltes et Semailles*, “Motives—or the heart within the heart”](https://web.ma.utexas.edu/users/vandyke/notes/ReS/recoltes_et_semailles.pdf)

A classical example makes the ambition tangible. The projective plane over a finite field has

\[
\#\mathbb P^2(\mathbb F_q)=1+q+q^2.
\]

Its complex counterpart has a one-dimensional rational Betti cohomology group in each of degrees 0, 2, and 4, and none in odd degrees. The motivic decomposition is

\[
h(\mathbb P^2)\cong\mathbf 1\oplus\mathbb L\oplus\mathbb L^{\otimes2},
\]

where \(\mathbb L\) is the Lefschetz motive. The decomposition is established over any base field. The three summands account, respectively, for the terms \(1,q,q^2\) in the point count and the cohomology in degrees \(0,2,4\). Their common structure is identified before passing to these different realizations. [Serre, *Motifs*, §5](https://www.numdam.org/article/AST_1991__198-199-200__333_0.pdf)

An equally familiar calculation opens a different question:

\[
\int_1^2\frac{dx}{x}=\log 2,
\qquad
\int_{|z|=1}\frac{dz}{z}=2\pi i,
\]

with the circle oriented counterclockwise. In Brown's realization framework, the relative cohomological object \(H^1(\mathbb G_m,\{1,2\})\) has, in suitable bases, the comparison matrix

\[
\begin{pmatrix}1&\log 2\\0&2\pi i\end{pmatrix}.
\]

The comparison relates algebraic differential forms to topological paths and loops. Its cohomological object joins two simpler pieces; a split object with the same graded pieces admits a comparison matrix with zero in the upper-right entry. In the rational mixed Hodge realization, a splitting would require the path integral to be a rational multiple of the loop integral. The first is positive and real, while the second is purely imaginary, so no such splitting exists. The pieces agree, but how they are joined carries information of its own. [Brown, *Notes on Motivic Periods*, §5.3](https://arxiv.org/html/1512.06410); [Burgos Gil and Fresán, *Multiple Zeta Values: From Numbers to Motives*, Theorem 2.154, pp. 111–112](https://www.maths.dur.ac.uk/users/herbert.gangl/mzv.pdf)

Motivic Galois theory develops questions about the symmetries compatible with such structures. Reconstruction asks what can be recovered from specified observations and relationships. These inquiries take place in several motivic settings with different constructions and hypotheses. Recovering a tensor category through a fiber functor does not automatically recover every variety supplying an object. A numerical coincidence alone does not identify an underlying structure.

Motivic cohomology carries a related purpose. Introducing their foundational work, Friedlander, Suslin, and Voevodsky explain that the goal of constructing a theory evolved into a deeper investigation of algebraic cycles themselves. Voevodsky's comparison theorem identifies motivic cohomology with Bloch's higher Chow groups for smooth schemes over any field. The theorem establishes how two accounts belong together. [Friedlander, Suslin, and Voevodsky, *Cycles, Transfers, and Motivic Homology Theories*, introduction](https://www.math.ias.edu/Voevodsky/files/files-original/Dropbox/Published_papers/Motives/Collection/s1.pdf); [Voevodsky, *Motivic Cohomology Groups Are Isomorphic to Higher Chow Groups in Any Characteristic*](https://www.math.ias.edu/vladimir/sites/math.ias.edu.vladimir/files/motivic_cohomology_groups_are_isomorphic_published.pdf)

Generality is useful here when it makes a relationship intelligible. It can also conceal a mechanism beneath unnecessary apparatus. A calculation may expose a limitation in a proposed structural account, just as the account may explain the calculation. Working through these tensions is part of understanding what the theory actually says.

## Motivic inquiry with capable AI

Motivic inquiry draws on several mathematical languages, making the conditions of learning especially consequential. Capable assistance could help a reader develop calculations and arguments in those languages together. It could explain an unfamiliar construction, locate a comparison in the literature, or investigate where an attempted identification fails. The breadth of the subject could become more accessible through such support.

Consider a family of algebraic varieties. One account might involve differential equations governing its periods; another, counts over finite fields; another, factors of a zeta function. A reader who obtains these results separately may still need to understand their relationship, even when it has been established in the literature. Doran and collaborators compute and match differential equations, point counts, zeta factors, and global \(L\)-functions for particular families of K3 surfaces through hypergeometric functions and their motivic interpretation. A shared factor coexists with differences in the remaining factors. Common structure can explain a relationship without identifying the whole objects. [Doran and collaborators, *Hypergeometric Decomposition of Symmetric K3 Quartic Pencils*, §§1.2–1.6](https://arxiv.org/html/1810.06254)

An AI-assisted study could follow these established comparisons: work through an equation, test arithmetic data, and examine why particular factors match. Agreement among outputs would still need scrutiny, since they could share a mistaken assumption. The inquiry concerns the specified relationships through which different descriptions constrain one another.

The logarithm example offers a smaller study. A reader can try to remove the off-diagonal term of the comparison matrix, then check whether the proposed change preserves the rational structure and Hodge filtration. AI could help carry out these tests and explain why a valid identification of complex vector spaces need not identify the structured objects. [Brown, *Notes on Motivic Periods*, §5.3](https://arxiv.org/html/1512.06410)

For a program centered on transport, descent, and reconstruction, these comparisons offer a direction of study. Questions about recovering objects from coverings, representations, and compatible symmetries acquire further forms in arithmetic geometry, periods, and motivic Galois theory. The questions give the program continuity across different bodies of expertise. Its course can follow what needs to be understood next, through reading, calculation, and exposition. [Tensorgami, program outline](https://tensorgami.com/about.html); [*Fourier, Symmetry, and Reconstruction*](https://tensorgami.com/notes/fourier-symmetry-and-reconstruction.html)

An interest in common structure can give people reasons to think together. Building a shared inquiry around it depends on particular collaborators, seminars, and habits of explanation that help others enter the subject.

## Learning and participation

These projects raise a practical question about what their participants learn through the assistance they use. A completed argument does not by itself give a reader the ability to see why it works or where it applies. Assistance can develop those capacities. It can also make it easier to pass over the activity through which they develop. Some effort is an avoidable barrier; some is part of learning to see. The effects of removing it depend on the learning arrangement.

Understanding can acquire continuity across encounters. Worked examples, failed approaches, checked references, and explanations revised after criticism give later inquiry a history to build on. A reader comes to recognize which assumptions require attention, where a familiar comparison stops working, and which question to ask next. AI can help maintain and extend that record. Its intellectual value grows through use: returning to an argument, testing a variation, or explaining a distinction to someone else.

In a randomized experiment involving nearly a thousand high-school mathematics students, Bastani and collaborators found that two GPT-4 assistants improved performance during practice. With assistance removed, students who had used the more unrestricted interface scored 17 percent lower than the control group. A tutor designed to support learning largely avoided that loss, without establishing an improvement in the unassisted exam. The experiment concerns a particular educational setting, but it shows why immediate performance and acquired capacity need separate attention. [Bastani and collaborators, *Generative AI Without Guardrails Can Harm Learning*](https://hamsabastani.github.io/education_llm.pdf)

Design can also improve learning. In a randomized crossover study of two physics lessons, Kestin and collaborators found that structured AI tutoring produced stronger immediate learning outcomes than active classroom instruction. The tutor incorporated expert-written solutions and careful instructional design. Expertise helped create a resource capable of serving learners beyond repeated personal access to its designers. The study establishes a result for those lessons, not an entire course of intellectual development. [Kestin and collaborators, *AI Tutoring Outperforms In-Class Active Learning*](https://www.nature.com/articles/s41598-025-97652-6)

The relevant comparison is between the learning arrangements actually available. For someone without specialist help, access to explanation, examples, and correction can open an inquiry that would otherwise stall.

Shared inquiry can also make an argument part of a relationship. Readers can work through a surprising step together or share the pleasure of a difficult distinction becoming clear. People who already understand a proof can enjoy discussing what they find striking in it. AI may help prepare the exchange, contribute an explanation within it, or supply the work around which it forms. Such encounters give people reasons to return to the mathematics and to one another.

The terms of assistance therefore matter. AI firms are institutions too: academic authority can decline while control over computation becomes concentrated elsewhere. Affordable access, inspectable arguments, and records that can be retained and continued through other tools give people greater room to disagree and change direction. Time and material support remain part of that freedom. Intellectual self-government requires the ability to question the new provider as well as the old gatekeeper.

A reader returns to a familiar integral and sees, this time, why a comparison contains more than a list of dimensions. Another works through a proof with a group until its hypotheses become intelligible. The explanations were already available. Now they can be followed, questioned, and carried into another conversation.