Looking up

We are living through unprecedented trying times for research, times that increasingly look like a crepuscule, a moment of high existential danger for research communities and academic freedoms. Now is the time for academics to take a stand or go to hell after making a Faustian pact. This is my public plea to colleagues consuming genAI products, for whatever stated reason, to stick their head out of the sand, to grow a spine, and to please, please stop destroying the soul of research and of their peers who do not use it.

Yesterday, openAI released a batch of 722 mathematical manuscripts. What main bit of information does this announcement convey ? Definitely not that mathematics were done, important theorems proven and frontiers broken. The subliminal message, rather, is textbook, agressive fear-of-missing-out (FOMO) marketing brilliantly (and viciously) targeted at a professional community plagued by egos, competitive incentives, production metrics, and individual glorification with prizes and PI funding : “It has become impossible to do state-of-the-art standard research without our tools. Use them now or disappear in the competition. We are now essential research infrastructure. All resistance is futile. Buy licenses, come work with us, glory and recognition [and in some cases, money, see e.g. acknowledgements section of this paper] will be yours”.

Over the last few days, several younger colleagues of mine, in hard STEM fields, spoke or wrote to me to confide how sick they feel about the takeover of AI in their research departments. One of them looked genuinely and worryingly thrashed psychologically by his conversations on the subject with colleagues, of hearing from the mouth of colleagues who do not actually understand the work he is doing how “useful” AI is and how it s going to “transform” our research and “open new horizons”.

Over the past week, I myself had about 5 times the same real-life conversation with several old respectable colleagues in different places, all in physics/math/astro, who I thought should know better. All these discussions gravitated around FOMO, the risk of disappearing, how AI is changing their life for the better by freeing them of “tedious” technical things (like: coding, writing papers, proposal, evaluation reports), how refusing to use it is impossible to understand, because it would be “like refusing to use a computer” (whatever that means, but critical thinking is not apparently a base skill anymore in academia, and reflexivity and introspection even less). Interspersed in most of these conversations, some assertions that reveal that they are partly ill-informed (about for instance what a proof assistant is, about the grey zones of plagiarism and snooping behind recent math and biology results announcements), partly brainwashed by marketing and the institutional bandwagon dynamics that has inevitably set-off, but also — and this is the most important problem by far — deeply disconnected from the real world. Not our socially-constructed professional world, which their attitude towards AI and its “performance-enhancing” capacities shows that they are finely atuned to. Here I’m talking about the one and only Real world. That of soil, clouds, oceans, forests and all our evolutionary tree, the world that (still) sustains our very lives and whole societies. That world also includes the human condition itself in a broad sense, the sharing of a common destiny on a finite, precarious piece of rock.

Three years ago, a good share of academics (which I still believe, perhaps wrongly, has a non-negligible intersection with those consuming genAI today, because of course in academia we are all left-leaning people with good intentions aren’t we) were struggling with the idea that politicians didn’t listen to “us”, that we served no purpose, had no social value anymore, because the world’s effort to stick to the Paris agreement by reducing our dependence on fossil fuels were failing, because carbon credits were a scam, because pesticides were constantly reintroduced against scientific evidence and the huge rise in pediatric (and other) cancers. Lobbies were winning against “us”, “Science” (capital S cynically indented). Three years later, there’s a new Frankenstein creature in the ongoing movie of existence-threatening technological developements enabled by Science itself (suggested reading: “The Shock of the Anthropocene and “The joyous apocalypse, a history of technological risk” by my historian colleagues Christophe Bonneuil and Jean-Baptiste Fressoz). That new creature of our own making is genAI, and for some (essentially sociological, sheepish) reasons, we as a global community of “intellectuals” are rushing to adopt it, despite the fact that it makes the climate and ecological crises we like to blame our politicians for even worse on a blink-of-the-eye timescale. Incidentally, that new Frankenstein creature also creates a whole new suite of existential problems for society, democracies, and for freedom that our academic community usually proudly brandish like a totem or badge of honor.

All of this, colleagues, is a major problem and crisis that we must address face on, and by face on I do not mean by creating “independent” committees to bargain with the leopard-eating-people-face AI companies. I am not going to follow the usual academic ways and provide you with a deluge of factoïds to convince you why. At this stage, we should all already know what the facts are. You can read the critical press and learn about the facts everyday if only you are willing to do it, and are not sticking your head in the sand about the already huge negative externalities of AI for mediocre performance, which I have already posted about. I’ll just give you a few links to recent publications here to save you the effort and environmental impact of a chatGPT prompt though. Paul Krugman on AI vs. everything else, and its implications in terms of lack of developments of, well, everything else (schools included, currently a hot topic for our own children in France). Ketan Joshi on Nvidia’s climate hoax report and real impact. For French readers, reporting in Le Monde about the broad environmental impacts of data centers if their deployment continues at the planned pace until 2030 (which is necessary for your slop vibe-code to eventually become less worse – perhaps) and about the worrying increase of CO2 emissions associated with AI development. And so on, and so on.

Indifferent or incredulous of that ? Here’s another problem : consider the mental and psychological toll AI is taking on all of us, you genAI consumer included. First, by delegating your cognition to an automat, you are not becoming smarter by publishing more and being more “efficient”, you are actually becoming dumber and reducing your intellectual lifespan, just like doped athlets end up having grave health problems (Laurent Fignon, Greg Lemond, Lance Armstrong, the south african rugby team of 1995 anyone ?). I am not going to mince my words here : if you don’t think that’s true, read the f* literature (I have a folder full of it thanks to some colleagues showing a healthy interest in these issues) and perhaps also reconsider your definition of “smart”. Then, by pressuring your own colleagues by transmitting your FOMO and own professional insecurities on others, presenting AI as an evidence, unavoidable fate and survival requirement (or, even worse, as an evil you are supposedly cleverly manipulating to give the middle finger to the technocratic system), you are hurting and negatively shaping the thought process of these colleagues, whether you like it or not — and that is especially true of young researchers in their formative years. In practice, it amounts to telling your kids that school is not important and that they should not put too much effort into learning and thinking anymore. This is turning into my biggest short-term fear of academic impact following my recent conversations, way beyond the industrial production of mediocre slop masquerading as groundbreaking science. At the current pace, there is soon going to be a Big Quit event of geological proportions for the academic world — and those quitting will precisely be those who have refused to delegate their cognition and have kept it functional by rejecting the Faustian bargain. Not a good way to prepare for a better future, if you ask me. Do you want to be contributing to this erosion of skills and talent ? Will you be happy to “survive” the “competition” in this context by bulldozering your critical colleagues out ?

I hear a counter-argument coming at this stage. I know because I have heard it so many times now after making the above arguments in front of colleagues : “Yes we all dislike it, it is not perfect, but we have no choice and what we really need are small AIs performing well on small specific tasks, and we need sovereign, ethical tools instead”, lalala, lalala. Do you really believe you are pushing in that direction by currently remaining, or becoming new consumers of big tech AI corporations (how many fraudster colleagues in French labs trying to use Claude or copilot while it is institutionally forbidden, just because they perform better than the hilariously bad slop-generators our institutions have provided us with at a prohibitive price tag for them ? Do you seriously believe, at this stage, that this is the path we are taking ? And do you know for sure whether the scenario you are calling for is itself technically credible and sustainable ? I call either bluff or delusion. This narrative only serves as an excuse as it stands. It is obvious nobody is considering this alternative seriously at the moment, or even working on it as a priority at the scale it would require to become quickly operational. It is also crystal clear that by consuming bigtech “products” “for the lack of a better solution” we are only consolidating their monopolies and contributing to instilling more FOMO within our own labs.

Finally, if you have not given up at this stage despite my bitter, disilusioned snark, consider the following. You are an academic, and some of your peers and neighbour research departments are working hard to mitigate today’s existential world’s climate, ecological or social problems (or all of the above jointly) already present before AI appeared in the landscape. By consuming AI products (which is optional, nobody – nobody – is forcing you to do it as an intellectually–independent academic, which you are, aren’t you), you are fueling these crises and are therefore actively sabotaging and annihilating the hard efforts and work of many of your own peers in environmental or social sciences (you know, precisely the kind of science that extreme-right governments are usually happy to get rid of). Congratulations, you are now explicitly part of team “aggravating problems” instead of team “finding solutions”. Maybe many of you working in fields going all-in on AI (which, as far as I can see, ecology itself is not as yet a proud member of the club, for how much time I don’t know) do not have much idea how hard it is for conservation or social scientists to find new ecological or social, democratic resilience pathways that don’t make the problems worse, and how hard it is going to be to scale up all their little incremental findings and local solutions to the planet as a whole (because there is no global silver tech bullet to stop the mess we have put ourselves in). It is very ,very, very, very hard. This results in a simple bathtub problem. Environmental and social research are trying to pour small drops of resilience in a bathtub, while the bathtub drain you are contributing to shape by consuming genAI like popcorn at the movies increasingly looks like the Niagara falls. There’s just no way we can refill the biodiversity, climate, or democratic resilience bathtub or even just reach steady state this way. It’s essentially Brandolini’s law applied to the destruction of the world (and a perfectly credible hypothesis to solve Fermi’s paradox). By consuming genAI , like it or not you are on the most short-term, impactful, yet toxic and wrong side or the argument, and of the fight for a livable future.

By the way, I know you know (do you really, though, if you keep consuming genAI today ?), but AI is not going to solve the climate crisis either, quite the opposite, see e.g this analysis and this analysis and a myriad of others.

So now is the time to answer this question, my fellow senior academics in a safe employment position. We simply cannot afford genAI on this planet. Will you knowingly collaborate with the borg and be assimilated by the FOMO playing on your own professional insecurities, or will you resist this urge and actually show the intellectual backbone, thought independence, and creativity academics usually pride themselves on ? Or is this independence and critical spirit all a big lie, a story we made up as a community to feel good and superior ? An increasing number of people who see the impact of AI playing out in front of their eyes are now starting to oppose it. In our comfortable ivory towers, sitting in front of computers displaying seemingly innocuous prompt windows or agentic-whatever software clusterfucks, far away from the damage (but for how long ?), do we as a privileged and shielded community officially aiming at improving knowledge want to be among the last to realize the scale of the problem and the harm we are contributing to do ? Look deep in the eyes of your own children, look at the natural world around you, at the toll our unchecked thirst for new technology is taking on the planet, on communities, societies, democracies, the further inequality and racism it creates or amplifies through its biases, consider what it is that you want to contribute to this world, not just in terms of publications in big magazines and splashy hyped press releases for once, but morally and ethically. What trace are we as a community going to leave on the tiny piece of rock we inhabit when our short time here is done ? If you decide to continue consuming genAI today knowingly, you are willfully providing the fuel for bigtech AI operators to thrash the world and human societies even further and faster. There’s a myriad of well-documented research about this now, and scientific facts are supposed to be the gold standard for us : will you consider them duly and revise your position accordingly, or “skillfully” continue to ignore them to preserve your short term technical and career interests, as any unethical scientist would do by rejecting those annoying data points that do not serve their story in order to publish big claims at all costs ?

FOMO is a powerful driver of our formatted competitive academic existences, but I know by experience that there is a lot of joy and promise in rediscovering the perks of feeling insignificant and academically impotent, and to start anew. Falling into academic oblivion because you reject genAI may also be more benefitial than you think for your daily life. By the way, this won’t happen either. Because by stopping to consume genAI you will factually personally contribute much, much more to make the world a better place in both the short and longer terms, and think and act more powerfully than you would have ever done by breaking the arxiv monthly limit of automatically-generated mediocre submissions, by data-mining a 5 Terabyte astrophysical database with agentic gremlins, or even by solving the Riemann hypothesis, mind you. Deal ?