The Right to Think

Gaia Carenini, PhD student at University of Cambridge

In the first day of philosophy, my professor walked up to the blackboard and wrote a single word in capital letters: MAN. Then he turned around and asked us what it meant.

Silence. We were fifteen, and afraid of saying something banal. In the end we arrived at an answer: man is a rational animal. An animal that thinks.

I know very well that this definition is incomplete. We are also body, affection, fragility, the need for one another; and no one is worth more for how well they reason. And yet there is one thing I still hold to be true: thinking is our way of inhabiting the world. It is through thought that we give a name to what happens to us, and it is through thought that we choose how to respond, and who we want to be.

That is why, when I see that little by little, almost without noticing, we are beginning to delegate our thinking, I feel a knot in my stomach.

I am a doctoral student in mathematics. I speak from where I stand, and from here, this is what I see.

Much of my work consists of not understanding. And not understanding is heavy. I set out down a path, and it breaks off. I build an example, and discover I was wrong. There are days when the only step forward is managing to say precisely where I am stuck: like someone who, in the dark, finally finds the door, even if it is still closed. From the outside it doesn’t look like much. And yet this is how I learn to see what I could not see before.

Anyone who has learned something difficult knows this. An explanation can be correct and still remain foreign to us: we repeat it, we use it, and it is not ours. Then comes an example, a question, and suddenly the reasoning rebuilds itself in our hands. It is one of the quietest joys I know. From that moment we are different, even if no one notices. As the fox tells the Little Prince, what is essential is invisible to the eye.

I am not saying that machines should stay out of our work. Checking a calculation, consulting the literature, verifying a case we don’t have time to do by hand: tools like these have always accompanied us. The difference lies in the gesture. It is one thing to be given a hand while you walk; it is another to be carried in someone’s arms through the very stretch where you learn to walk.

And I must be honest: that stretch is very easy to skip, and I myself know how strong the temptation is. A summary before having read. A solution before having tried. An opinion before having understood what convinces us. You arrive, and you have not passed through anywhere.

In mathematics, the solution to every exercise is often at the back of the book. We assign it anyway, because someone still has to learn how to get there. If we call it a success when the solution is handed in, even when the person handing it in has learned nothing, we have forgotten why we teach. And we have forgotten that, on the other side, there is a person.

You will tell me that research is a different matter, that what counts there is the result, the much-vaunted progress. But research, too, is already inside this story. In May 2026, a counterexample produced by an OpenAI model to an Erdős conjecture on unit distances was announced, and mathematicians verified it. In October, a collection of hundreds of manuscripts generated by a model arrived. I do not deny it, nor do I want to belittle it: a correct result remains a correct result, and some of the arguments put forward are remarkable. But my question is a different one, and I do not know the answer. I only know that we cannot stop asking it. What will become of a young mathematician who will be able to skip, every time, the very difficulties through which she would have learned?

They tell us that our task, in the future, will be to choose the directions to explore and to build theories. I don’t know why a machine should not be able to do that better than we can: we haven’t tested it yet. But there is a deeper problem. The ability to judge how interesting a direction of research is does not come from nothing: it grows through attempts, through problems faced, through learning what it is natural to ask. If we make this apprenticeship superfluous, we hand to the generation that is coming answers that we have not taught them to evaluate. We leave them alone before what they cannot recognize. It is a failure, even if the number of papers keeps rising.

And it is not only training that is lost. You can spend an hour at a blackboard and walk away with a better question. Explaining a still-confused idea to a colleague forces us to look for an example, to listen to an objection, to discover what we had not understood. In that work, without noticing, we also learn to pay attention to others, to cultivate a shared knowledge, to create culture. A department can go on publishing while its rooms empty of voices. This is what I fear: a house full of results and devoid of anyone to think them together, of someone who can sense their worth. A house in which no one feels welcome anymore.

So far I have spoken of what happens to each of us. But no one chooses alone: the habits of many become the rules of all.

If it becomes normal to produce many more papers in less time, those quantities become selection criteria, and those who work differently risk being left out. At that point, to speak of free choice is no longer honest.

And if the expected performance requires paid tools, those who cannot afford them start further behind, and we end up mistaking the ability to pay for the ability to do research. This we cannot accept: it risks widening the imbalance between the countries of the world even further. And there is more. A community that builds its work on commercial services accepts depending on decisions made elsewhere, about costs and conditions of access.

I hear it said more and more often that, from now on, the worth of a mathematician will be measured by visibility and communication. Some neurodivergent people have found in research a place that welcomes their rhythms and their ways. For some it is perhaps the only one. Asking everyone for the same style of public presence may close that door. The quality of a thought is not measured by the confidence with which the person carrying it stages it. The inclusiveness of our community is a fundamental value, and I believe we must protect it.

In all these cases we are asked how to adapt. But if we ask only that, we have already accepted that the transformation must happen this way. I do not accept it. I do not have all the answers, and I know that some of mine may turn out to be wrong. But I believe we can decide which uses of artificial intelligence to admit and which to refuse, and that the criterion is the one from before: to distinguish what accompanies thought from what replaces it. Saying so does not mean opposing progress.

I know I speak from a fortunate position. I was able to study and to have time for questions that served no immediate purpose. I am grateful for it, and I know I did not deserve it more than others. For this reason I feel I owe something to those who have not yet had these possibilities. Those who are beginning are afraid, and it is right that they should be: they cannot bear alone the cost of opposing new expectations of productivity, and no one should ask it of them. Those who lead a group, allocate funding, or evaluate an application can open that path, or close it. I ask them to open it. And if that has a cost, let it be they who pay it, not the youngest.

Someone may object: if a machine does it better, what is the point of defending this effort? I answer that mathematics is not a mere list of true statements. It is what remains in a person when they understand why they are true. A theorem that no one understands is a result, but it is not yet someone’s knowledge. And we do not make this effort in order to be the best. We make it because mathematics cannot be separated from the community that thinks it.

We mathematicians are among the first to face this change, at least on this scale. Translators went through it before us, but the best of them stayed, because the artistic side of their craft seemed untouchable. With us, instead, the change has reached everyone: even the best seem able to do little compared with AI. For this very reason we are given the chance to show that a community is capable of defending its own humanity, even at the cost of producing less.

Concrete choices are needed. To protect, in training, the work done in the first person. To recognize, in evaluation criteria, the time it takes to understand. To support those who do research without handing over to others the heart of their work. These are decisions that belong to our institutions and to those who lead them, not to AI companies.

I return to that blackboard, to that word: MAN. I fear that our humanity is being taken from us, or rather that we ourselves are letting it go, one shortcut at a time, until we no longer notice. And I do not want that to happen, neither to me nor to those who will come after. I do not accept that giving up on thought should become the price to pay for continuing to do mathematics.


Received 23 October 2026.

9 responses to “The Right to Think”

  1. M Avatar
    M

    I find your text very accurate on many points, and I share many worries.

    One important point I could emphasize is that we cannot try to influence how other people want to proceed. I think we need to accept that math is not the same anymore. Like Euler had to accept how we did it in 2022, and like some cave man discovering for the first time how to make fire had to accept that our world is quite different to theirs.

    I think one of the best things todo, if one wants to make math a fruitful profession, is to focus on understanding, theory building, simplification, community building, and also (very importantly) on useful and maybe futuristic applications. The way to do this might be fundamentally different compared to 2025, but I think it will be a business model to stay in control over our knowledge, and to nurture the ability to think. Maybe we have to explore the boundaries of what AI can do, and learn how to comprehend this, how to think in a new way, how to teach people to truly comprehend this, how to make use of it without giving the decision about our fate to autonomous algorithms and big data. The difficulty of navigating our current dilemma shows already to me that we will have a lot of things todo that AI cannot just take away from us. Yes, AI can now do research, so let’s figure this out without giving in.

  2. Scott Guthery Avatar
    Scott Guthery

    The issue is not what you want to think about. You can think about whatever you like, with or without AI. The question is: What do you want to get paid for, whether by salary for teaching mathematics or by grants for doing mathematical research? If you aren’t using AI in these endeavors, you may find that your efforts come to be perceived as having less value.

    1. Anonymous Avatar
      Anonymous

      The headline is obviously not about a legal right, but a moral right for humans to run the show.

      Also, much of the article is about students’ rights to have a proper education. Academia could absolutely forbid AI on campus and make arrangements with the relevant authorities to jam cell phones at least during exams.

      But many don’t, because they want part of the industry money that originates from laundering and reselling stolen IP.

    2. Nilima Nigam Avatar
      Nilima Nigam

      It -may- be that ‘If you aren’t using AI in these endeavors, you may find that your efforts come to be perceived as having less value’ by some. Others may assign a higher value. The question presupposes that the only value worth discussing is an economic one.

      It may be that every single human who goes to university will be unemployable without formal training in AI tools. It may be that within a decade we see that introduction of AI tools in university is detrimental to higher-order thinking abilities, with skills related to AI being easily picked up in the workplace. It may be that the granting system collapses entirely, since agents could cheaply identify new research directions and work on them sans human intervention. At this juncture these are all entertaining hypotheticals. We’re projecting into the future about the economy.

      The author’s essay is about moral and cognitive agency, a somewhat different proposition.

  3. John Mitchell Avatar
    John Mitchell

    This piece is beautifully written, and I hope that neither A.I. (in the abstract sense) nor the A.I. industry annihilate the ability of humans to engage in the wonder of mathematical discovery and understanding.

    On the practical side, though, I think the public at large would find a couple of the author’s points baffling.

    1. “[W]e do not make this effort in order to be the best.” Some do, and it’s best to be honest about that. Bill Thurston rightly described academic mathematics as “fiercely competitive”.

    2. “[W]e are given the chance to show that a community is capable of defending its own humanity.” Most people outside the mathematical community (at least in the U.S.) probably don’t even know what that means. To them, “math” and “humanity” probably seem like opposite poles of a magnet. The public tends to care, if at all, only that mathematics has applications to science and technology (including, ironically, the A.I. technology that just washed over the entire field). If humans are still needed for that (as I hope), then society will support such humans; if not, then there may no longer be such a thing as a professional mathematician.

    1. Giovanni Mongardi Avatar
      Giovanni Mongardi

      About point 2, maybe we should learn from philosophers and all of the humanities how to make a case for beautiful works of intellect with no direct applications to the economy.

    2. S Avatar
      S

      1. Does that mean that mathematics is reserved only for winners or likely winners of the various prizes that litter the field? Prize culture in mathematics has created corruption, and has enabled AI companies to “buy” mathematicians as if they were raw materials for their models. Such a view also ignores the fact that there are different ways of being the best in any field, and there is no guarantee that someone will achieve what is considered the best in mathematics by actually aiming to be “the best” rather than, say, pursuing and exploring what they think is interesting or even, on a more mundane level, what is considered useful or worthwhile. A significant amount of good *and* useful mathematics, past and present, was discovered by chance and exploratory approaches, not by design or competitive requirements, and such contributions have included those from genuinely interested and talented amateurs, who would otherwise be excluded.

      2. The second point overstates the level of public interest in and understanding of mathematics, which is (and generally always was) probably quite low, and also makes unsupported claims about the public only caring about mathematical research that has applications in science and technology. I think the latter claim is false, but it is empirical and no evidence was provided. There is no formula for determining whether a given mathematical topic or theory today will have an application in science or technology today or tomorrow or in a hundred years. Mathematics has always been the most open ended of all the sciences, and should remain so, and I think this is something the public would recognise. And although public understanding of (the content of) mathematics is probably quite low, the public would be very sympathetic to the notion of mathematics as a highly specialised craft (or set of crafts) where the craft skills, in all their diverse forms, continue to be cultivated with human hands and minds, and not lost by delegating it all to AI.

      The question of whether “such a thing as a professional mathematician” should exist, the answer to which many would clearly say yes, should not be decided by AI companies, but by the mathematical community, and I think much of the public would agree.

  4. Giovanni Mongardi Avatar
    Giovanni Mongardi

    Very beautiful and well written. I completely agree with it.

  5. ThoughtBuilder Avatar
    ThoughtBuilder

    The entire concept of ‘computers replacing humans’ in mathematics comes from a simple distortion of natural language. Namely: there’s no such thing as ‘pure mathematics’.

    The author of this is on the right track by talking about ‘human understanding’, but then repeats the absurdity about ‘theories’ being created by AI (see below). Also, there is no point to human understanding.. of a theory that is not ‘memorable’ in the human sense. Closer inspection shows that this is exactly what is ‘meant’ by calling something original (rather than just random).

    Imagine an infinitely powerful computational intelligence that can form definitions at least 10,000 pages long (by putting together syntax in random ways), and then generate even more complicated theormes and proofs from those definitions. That would be completely ‘pure’ mathematics (not being ‘biased’ to any set of axioms), and it would not be interpretable by humans: however it is, in a sense, a higher form of intelligence. You can argue that the theories are much harder than human mathematics, and for a human to understand these definitions represents ‘greater achievement’.

    But this is not ‘mathematics’. The entire history of mathematics is the attempt to reduce pure formalism (above) into what is maximally human and simple, and memorable by humans. The implicit goal isn’t simply to generate ‘complex’ problems requiring more mental power, but the unity of the complex and the simple. This is what is meant by a ‘new’ concept: a new concept is a concept that ‘subordinates’ as much complexity as possible to simplicity, so a vast sea of ‘difficult’ proofs are actually made trivial. Human ‘embody’ simplicity, by being the weakest, most easily confused form of intelligence that can ‘just about’ understand complex formalism.

    All mathematics is applied, in different senses of that word. All of it involves a larger construction problem, to construct a system of ideas that plays nicely with formal, and informal (in this case maximally human) requirements.

    Conceptualization at a high level depends on humans, since humans embody it, and it is actually about humans. What are ‘creative’ ideas are not random ideas, but ideas that humans (the most limited cognitive window) find memorable.

    On the other hand, academic ‘pure math’ has been distorted since WWII since it defines this notion of mathematics as based on bureaucratic instrumentalism, namely solving problems that were deemed as important by some humans, based on their human agendas. The solutions themselves are intrinsically forgettable, and most likely no one will ever remember those solutions in just a few years. The ‘goals’ became reductionistic and formalistic, there is no long-horizon thinking, no ideas. (It also doesn’t make sense: a ‘conjecture’ is just a piece of information about an object. The larger question is about forming better ‘representations’ of that object, to play nicely with theoretical features, physical features, numerical, and otherwise.)

    There is no point to doing anything that isn’t memorable and applicable (as G.-C. Rota said, it’s always the definitions and not the ‘results’ of mathematics that are useful to other fields). Terminology is often an inversion of reality: I find ‘applied’ mathematics to be often friendlier to long-term theory-building, and to construction programs that unify the physical and abstract in a single object (e.g., neural coding, computational biology). Riemann, Poincare, Elie Cartan, Weyl, Kolmogorov, were all rooted in self-consciously applied math (the source of their ideas was the non-mathematical and partially symbolized example, understood in physical and dimensional terms before clear pieces can be found). Hilbert was someone who was as far as possible from a mental athlete as can be imagined, and who directly embraced the fact that his cognitive processing window was smaller — which was the very source of ideas (to reduce the complex to the simple and naive). In applied math and theory-building, no one views AI as more than a minor tool.

    The whole concept of computers ‘automating’ or replacing human mathematicians is just impossible as a concept, unless you are so totally bureaucratically indoctrinated that you define math as a bunch of technocratic goals.

Comments are moderated. Read our comment policy.

Add to the discussion

New posts by email.

Prefer a feed reader? Subscribe by RSS.

Also on Mathstodon.

Latest comments across the site.

  1. The entire concept of ‘computers replacing humans’ in mathematics comes from a simple distortion of natural language. Namely: there’s no...

  2. 1. Does that mean that mathematics is reserved only for winners or likely winners of the various prizes that litter...