How dare you?

Vadim Lebovici, tenured assistant professor in mathematics at Sorbonne University

“If your bottle was shaped like a torus, we could stir its content so that no point has come back to its original position!” Professor Carcan said enthusiastically to his student coming back empty-handed from the last source of drinking water of the city. She will be his first student to die from dehydration.

In: “The Mathematics of Professor Carcan”

Introduction

If you are spending some time wondering to what extent AI should be developed, spread and used within activities of the mathematical community, administrativewise, teachingwise and researchwise, this note is for you.

The purpose of this note is not to extend the Overton window. The Overton window has been dislodged and carried away from its original location. The purpose of this note is to center it back. Or, under time constraints, to make use of an overlooked inscription on the Overton window: in case of emergency, break glass.

Before stating our main result, let us recall the state of the art. All IPCC results cited in this note are taken from [IPCC23]. The following theorem is directly taken from there:

Theorem 1. (IPCC) Widespread and rapid changes in the atmosphere, ocean, cryosphere and biosphere have occurred. Human-caused climate change is already affecting many weather and climate extremes in every region across the globe. This has led to widespread adverse impacts and related losses and damages to nature and people (high confidence). Vulnerable communities who have historically contributed the least to current climate change are disproportionately affected (high confidence).

Highly alarming is the fact that this situation is not improving:

Theorem 2. (IPCC) Hazards and associated risks expected in the near term include an increase in heat-related human mortality and morbidity (high confidence), food-borne, water-borne, and vector-borne diseases (high confidence), and mental health challenges (very high confidence), flooding in coastal and other low-lying cities and regions (high confidence), biodiversity loss in land, freshwater and ocean ecosystems (medium to very high confidence, depending on ecosystem), and a decrease in food production in some regions (high confidence). Cryosphere-related changes in floods, landslides, and water availability have the potential to lead to severe consequences for people, infrastructure and the economy in most mountain regions (high confidence).

As always, IPCC results have two main advantages: they come with a scientific level of confidence and they are crystal clear. Mortality and losses. Death and desolation, if we go business as usual. This is what the statements above mean.

Remark 3. To give some order of magnitude of what needs to be done, let me give a few numbers. To keep global warming under +2°C, we need to divide the greenhouse gases emissions by 2 to 3 worldwide before 2050. This means a decrease of roughly 5% every year until 2050. A decrease of 5% is very easy, it already happened: during the COVID19 crisis. In other words, we need a COVID19 crisis effect on the greenhouse gases emissions every year until 2050. A COVID19 crisis effect. Every year. For 24 years.

But AI does not propose to keep business as usual, it proposes to increase the speed at which these consequences arise:

Theorem 4. Generative AI is, from an ecological standpoint, a disaster.

Proof. The fact that generative AI has dramatic ecological consequences which will not be compensated by its benefits is not only obvious, but also scientifically grounded; see for instance [A26]. A detailed proof is outside the scope of this note, but I still want to add an evidence which appears too rarely. The main application of generative AI is to speed up processes. The main activity of the countries having access to generative AI is a capitalistic extractivism and productivism which pollutes and destroy ecosystems by design. Speeding up these processes will do nothing but speeding up pollution and destruction. QED.

This is the state of affairs, from which any discussion on generative AI should start. Arguing the opposite is either a naive hallucination or a guilty lie.

Main result

The purpose of this note is to expose the following theorem.

Theorem 5. There exists an integer NN\in\mathbb{N} such that when our human-based activities will have killed NN humans1, mathematicians will stop wondering if using AI will help them solve Erdös problems to start taking strong political actions and putting their knowledge into fruitful directions.2

The proof of the theorem is obvious. The opposite would mean that mathematicians would prefer to let everyone die than changing the course of their activities. In spite of being a triviality, Theorem 5 naturally induces a useful question:

Question 6. What is the value of NN?

The number NN could be refined to depend on the mathematician considered. Some people already satisfy the conclusion of the theorem. Some do not. For instance, contributing to collections of mathematical problems that help big tech companies to further develop their algorithms does not satisfy the conclusion of the theorem. It is irresponsible and immoral. This means that NN has not been reach for these mathematicians. Of course, a very useful question if you are a mathematician is: what is your value of NN?

Conclusion

Thanks to Theorem 2, we know that the number of human beings dying due to ecological reasons is strictly increasing and will rise at an increasing pace if we do not change radically our activities (see also Remark 3). This brings us back to our starting point. If you are spending time wondering to what extent AI should be developed, spread and used within activities of the mathematical community, administrativewise, teachingwise and researchwise, then the above developments naturally lead to nothing but:

Question 7. How dare you?

How can you spend your energy on discussing what is a good way of using AI for mathematics, how to provide AI-based learning tools to students, what is a smart way of prompting to find new theorems or write grant applications when you know Theorems 1, 2 and 4?

Maybe you think it is easy to point fingers at people without proposing concrete solutions. I should make myself clearer. I am not asking why you do not pursue options A or B that I would have proposed. I am asking why you do not put time and effort into seriously and rigorously seeking for scientific activities which benefit humanity as a whole. Why you do not use your valuable knowledge and your ability to produce analytical and critical thinking to create an urgent and radical change in society—instead of enslaving it to the growth of a technology which has deleterious effects on humans, ecosystems and resources. The unquestioned, vague and business as usual biological application you are working on does not qualify as such. It is not up to the task. If you think it is, you should probably read again Theorems 1 and 2. The one-hour weekly discussion on AI with your colleagues is not up to the task. If you think it is, you should probably read again Theorems 1 and 2. Why don’t you stop and think about it? This is my question.

Of course, my question does not apply identically whether you are undergrad or grad student, postdoc, tenured or not. So let me be even more precise here. I want to address my question to tenured mathematicians with permanent positions who have not yet reached their number NN: what is your excuse?

References

[A26] Alpine, W., Geldner, N., Alpine, H. et al. AI-driven productivity gains enable more CO₂ emissions than they avoid in a global energy–economy model. npj Clim. Action 5, 71 (2026).

[IPCC23] IPCC, Summary for Policymakers. In: Climate Change 2023: Synthesis Report. Contribution of Working Groups I, II and III to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change [Core Writing Team, H. Lee and J. Romero (eds.)]. IPCC, Geneva, Switzerland, pp. 1-34, (2023).

  1. We mean, via the ecological disaster as stated in Theorem 2.
    ↩︎
  2. By fruitful directions, we mean fruitful for humanity as a whole, not for a small subset of individuals. ↩︎

10 responses to “How dare you?”

  1. AS Avatar
    AS

    This text looks more like a general critique of capitalism than a critique of a particular technology.

    I don’t dispute the IPCC findings. But Theorem 4 takes quite a leap in the argument offered alongside the citation:

     >> **The main application of generative AI is to speed up processes. The main activity of the countries having access to generative AI is a capitalistic extractivism and productivism which pollutes and destroy ecosystems by design. Speeding up these processes will do nothing but speeding up pollution and destruction. QED.**

    Notice that nothing in this argument is about generative AI. Substitute “science” for “generative AI” and every step still goes through. Does that imply we should close down applied science departments, since they too contribute to “capitalistic extractivism and productivism”?

    That question is not rhetorical, because Theorem 5 asks us to redirect our knowledge toward directions “fruitful for humanity as a whole, not for a small subset of individuals,” and I would like to know precisely which ones those are.

    What makes you think that all activities involving technological advancement necessarily benefit only a small subset of individuals? Practitioners in many fields, like drug discovery, utilities and energy, electrical engineering, are finding applications for AI that benefit humanity as a whole, or at least a large segment of it.

    Not everything is about big tech companies and Silicon Valley billionaires.

    Vadim, I see that you have taken steps to preempt this with “Maybe you think it is easy to point fingers at people without proposing concrete solutions.” But that is precisely what I am thinking. The argument would be much stronger with some examples of the directions you consider fruitful, and an account of why those directions should not benefit from AI.

    1. JM Avatar
      JM

      I think most people would concede that most technological advancements can benefit humanity as a whole and not only billionaires but the point is that technological innovations exist in a system which is harmful to the vast majority of people. This point generally true but is particularly glaring in the case of LLM companies which could only viable as a business in the dangerously speculative world of silicon valley and who uses the “good” uses of LLMs as a deflection from their incredibly damaging social and ecological effects.

      I don’t think it is a bad thing this post could be used to criticize capitalism more broadly. It is necessary that we are critical of how new technologies are instantiated and it is the case that LLMs are instantiated in a capitalist system which if left on its current course will lead to total ecological (and probably social) collapse.

      1. AS Avatar

        “LLM companies which could only viable as a business in the dangerously speculative world of silicon valley” – this is simply not true. Open-source models already exist, and there is every reason to believe more will follow.

        This argument overlooks a fundamental dynamic aspect of technological evolution. Usually, technologies get cheaper and democratize over time.

      2. JM Avatar
        JM

        I was referring to initial AI investments in hyperscalers which I think there is good reason to believe was only possible in an environment like Silicon Valley.

        I also think we can’t really wait for this technology to become democratized when there is a climate crisis happening right now. Especially when it seems likely that energy costs for AI will only increase for the foreseeable future and other energy intensive technologies are surely on the horizon.

    2. Paul Avatar
      Paul

      I disagree with “Substitute “science” for “generative AI” and every step still goes through” : the goal of science is not to speed up processes, but to understand the world and perhaps to build progress — but this is not necessarily speeding up anything. Even for applied and technical science, this may perfectly lead to different things than technical solutions, extractivism and productivism (random examples : new forms of medical treatment through other means than drugs, or lighter drugs, or new materials for insolation, or better recycling processes, or only pointing at issues with current technical systems)

      1. AS Avatar

        Paul, and why AI can’t help with coming up with “new materials for insolation” or “lighter drugs”?

  2. DavidS Avatar
    DavidS

    Refreshing. This environmental viewpoint was yet missing in the posts.
    Your arguments sadly rises another question for our community: given the emergency, what is the point of maths?

  3. Tobias Bisang Avatar
    Tobias Bisang

    Thank you for repeating the obvious, for talking about it, for taking it serious, for your emotional courage.

    I wish the same for all mathematicians.

  4. AS Avatar

    JM,

    Energy consumption limits on data centers are determined by the regulators after consultations with local stakeholders. If regulators cannot decide on a spot, they can enact a temporary ban: see https://www.engadget.com/2214456/new-york-kathy-hochul-data-center-ban/ , for example.

    If you care about these issues, please write to our local government and advocate within your local community. There is nothing wrong with political activism.

    But the political activism does is problematic when it becomes a part of publishing, grant, and hiring decisions. You don’t want your work to be judged based on someone else’s political preferences.

  5. SM Avatar
    SM

    The question “how dare you?” here is mainly rhetorical. I believe it would be instructive to try and properly answer it.

    The text is written in such a way that it is hard for any reader to say “I dare because…”. If this was hard because everyone gets converted, then the author would be delighted. But I suspect that, for many, the situation would be “I can’t answer but I’ll continue using AI”. So, whether we use AI or not, it could be interesting to investigate robust reasons why people want to use AI; reasons that resist being told all the above. By understanding them, maybe a more efficient course of action can be taken.

    As long as these reasons are playing a role in the psyche of people, it is worthwhile to inventory them. They do not need to be good reasons.

    By the way, regarding climate change, bad practice is very common. I also believe that this phenomenon of being aware but not acting accordingly must have been well studied; if some people are knowledgeable about this, I would be interested to know more. I know that, in French, there is “Climat : mon cerveau fait l’autruche”.

    So what could be these reasons? Here is an improvised list, not meant at all to be exhaustive:

    – Not caring about climate: either out of disbelief, or because it would change to many things to accept it, or because “I am single person, I won’t change the world by epsilon but I can change my life a lot” (dilution of responsibility), or selfishness, or because it is too abstract to get a feel for it, or simply not caring.

    – Comfort: everything becomes easy and fast. We can do more and get more free time.

    – Habit/addiction. This does not justify starting but if we start, then by induction we may continue.

    – “I only use AI a little”. Or, in combination with the previous item: start a little and it gets larger and larger.

    – Proving great theorems – this is a powerful tool. If one applies pre-AI criteria to judge people, then by using AI, anyone can become an excellent researcher overnight.

    – Personalised tutoring for students.

    – Getting boring stuff out of the way.

    – Incentives by institutions or people around, and of course by companies.

    – Competition with other teams, who use AI. “I don’t want to fall behind.”

    – “Oh anyway, our daily life already kills the planet, including AI in my routine won’t change this by a huge factor, will it?”

    – “I am permanent but I want a grant or to get Full Prof or whatever.”

    – “I am a permanent but I work with non-permanent people: for them to get a job, we have to use all tools available.”

    What else do you see?

Add to the discussion

New posts by email.

Discover more from Proofs and Prompts

Subscribe now to keep reading and get access to the full archive.

Continue reading