* This post was inspired by an email I recently received, announcing the new AI initiative at the University of Copenhagen. It’s about this initiative, but really, it’s about the broader rush to incorporate AI into higher ed with (I would argue) insufficient caution and pedagogical warrant. Not to mention environmental concern.
Max Dresow writes…
The University of Copenhagen, where I work, just made a $17,000,000 investment in artificial intelligence. According to the University Post (Uniavisen), AI will soon be integrated into everything “from research and teaching to administration”— in three short years, KU will be thrust into the AI future. “The time has come to grasp the nettle,” says newly-appointed AI czar Morten Axel Pedersen. “We can’t wait any longer.” Pedersen then rehearses a litany of familiar talking points. AI “will lead to great scientific discoveries.” It will be “the alpha and omega of the future job market.” It might spark a second Industrial Revolution, or at least a socioeconomic change of equivalent magnitude. It will even revolutionize our politics. “It’s going to do something about our productivity, our political conversations and democracy, and the whole way we perceive ourselves and relate to each other.” In short, the AI rapture is upon us— so get on board as we revolutionize not only the university’s digital infrastructure but also its core pedagogical philosophy and vision of its place in the world.
As you will have gathered from my tone, I’m skeptical of all this. I’m skeptical, first, about the world-shattering promises of AI. Take the point about major scientific discovery. Now, maybe these really are on the horizon— I suppose it’s possible that we’ll all look back on the 2020s as a significant rupture in the history of scientific progress. But we’ve been here before, haven’t we? The Human Genome Project was supposed to hasten the discovery of disease cures, inaugurate a new era of personalized medicine and satisfy the Socratic injunction to know thyself. Go back and read the hype surrounding this project. Listen to Bill Clinton’s embarrassing speech, delivered during the twilight of his presidency. Then ask yourself whether the discourse sounds familiar. Great expectations are not always met.
This is not to deny that AI will have important consequences for the world. It already has. The sheer amount of money that’s been invested is mind-boggling. Many people have profited; some have become astronomically rich. Certain firms have used AI as a pretense to cut payroll. A variety of tasks that were once performed by humans are now routinely automated. All of these are real changes, which have affected real people. But to compare them to the Industrial Revolution seems a bit much, even if we grant that it’s early days. While AI has changed the way people work, there’s little evidence that it will remake the workforce, revitalize democracy, or alter the very texture of human experience. Sometimes hype is just hype, and as desperately as AI boosters want LLMs to change the world, the most likely outcome is that AI will leave the world much as it found it, only worse. (Think the smartphone as opposed to the printing press.)
But actually, let’s not presume that things will be this benign. Because, while AI will have positive uses, it’s now clear that it will be an environmentally disastrous technology. As a recent UN report put it:
Data centres, the global infrastructure powering AI, could consume 945 terawatt-hours of electricity annually by 2030— nearly triple the combined annual electricity use of Pakistan, Bangladesh, and Nigeria, countries collectively home to more than 650 million people.
However, this is just the tip of the iceberg. On top of the carbon footprint, every unit of electricity used by data centres also carries a “water footprint” for cooling and energy production, and a “land footprint” associated with power generation and supply chains… According to a new study from UN University (UNU), AI-related water consumption could equal the basic annual domestic needs of 1.3 billion people by the end of the decade, while its land footprint may exceed 14,500 square kilometres— roughly twice the size of the Jakarta metropolitan area.
Set against these very-bad-and-scary projections, claims that LLMs will deliver major climate benefits seem utterly cynical, or else deluded. Here, AI boosterism joins hands with something darker: a promise that technology can rescue a society in thrall to perpetual growth and angrily unwilling to consider anything else.
WTF
It’s disappointing that my employer— an institution with a commitment to sustainability, seeking to halve its carbon emissions by 2030— is largely tabling these concerns.* Still, what’s perhaps more disappointing is that the university seems intent on foisting AI into virtually every degree program they offer. This strikes me as wildly irresponsible, not to mention out of step with the university’s commitment to evidence-based pedagogy. Where’s the evidence that this is a good idea? I’m willing to be convinced that there are pedagogically beneficial uses of AI. Certainly I can imagine uses that enhance the accessibility of classrooms, which is doubtless a good thing. But this new initiative doesn’t strike me as an adventure in evidence-based pedagogy. It strikes me as a bit of institutional FOMO, and a desperate gamble on the ill-defined transformative potential of a fashionable technology.
[* Pedersen does at least acknowledge environmental concerns, and has remarked that “in the longer term, I would like to explore whether we can assess how [increased AI use] contributes to our overall climate footprint.” (Adversely.)]
My PhD is in philosophy, and likewise most of my teaching experience. So— I’ve been curious to know— how does AI stand to improve philosophical education? If there is a good answer to this question, I haven’t heard it. I’ve heard of professors training chatbots to argue with their students; of chatbots made to play the role of the Socratic naif; of AI outputs used as classroom conversation starters; and sure— these are uses of AI. But are any of them… good? It feels like someone asked an AI model how to integrate AI into the classroom and it spit out a bunch of unimaginative suggestions. Until further notice, the pedagogical utility of AI for philosophy seems entirely unproven.
But in fact things are worse than this, because increasingly routine applications of AI are (I would argue) pedagogically harmful. Consider that most endangered of assignment types: the argumentative essay. Now, everyone knows that generative AI has made the assessment of argumentative essays a complete farce. But people disagree as to where exactly the problem ends. Most everyone agrees that it is bad to use generative AI models to write your paper— if a student presents writing as their own, when in fact it was generated by an LLM, that student is straightforwardly cheating. Yet opinion divides on whether it is equally bad too have AI refine an idea, “check the strength'“ of your argument, or come up with an outline for a paper.
I think it’s just as bad. The reason is that writing is a basically empathetic activity. To write a good essay, it’s not enough to have a good argument. The writer must also present the argument in a way that will be intelligible to a reader, which could be the professor, a peer, or someone else (e.g., a policymaker). Needless to say, it isn’t easy. For many students, the act of planning an essay requires a considerable mental effort. It's no small thing to project yourself into the mind of a reader, and to decide what they need to hear in order to make sense of your argument. But that’s exactly how it should be! Learning requires friction. When your thoughts roll effortlessly across a frictionless plane, you aren’t learning. (It’s the same reason people don’t learn much from glossy YouTube videos, or for that matter, university lectures. These don’t generate friction, unless you generate it yourself.) So when you outsource the planning of an essay to an LLM, you forgo the chief pedagogical value of the activity. You circumvent friction by means of an anti-friction technology, and exclude empathy from an empathetic practice.
I didn’t mean for this to become an anti-AI screed. Of course AI has useful applications. Lord knows I’ve used it to translate documents since I moved to Denmark, and sometimes to track down references and retrieve bits of information from the web. I do these things with a measure of guilt, but I’m not an ascetic and I don’t oppose the use of AI every now and then. What I oppose is the embrace of AI maximalism— the view that we should use AI everywhere, without limits, including in the classroom. Almost certainly there are pedagogically valuable applications of AI. I don’t know what they are, but I imagine they probably exist. (Right?) Still, we shouldn’t let the projections of AI maximalists convince us that AI must have an expansive role in the classroom of the future. Can’t we at least have some encouraging studies before we rush, arms outstretched, into a new pedagogical experiment? And don’t we owe it to ourselves to have a discussion about whether it’s consistent with our values to infuse AI into every fiber of the university?*
[* I want to be fair. Since I put up this post, I’ve become aware of another statement from the university: this one specifying that, under the new initiative, “degree programmes and lecturers [will] need to increasingly consider and use AI where it makes sense and opt out of it when it does not support education and learning.” This is better than a policy of outright maximalism; but in stressing the need to integrate, and in referring to non-integration as “opting out,” a maximalist stance is nonetheless implied.]
It’s understandable that people are excited about AI. The predominant theme of the past twenty years has been the futility of coordinated action. We document, with ever improving data, global biodiversity loss, but then we miss every single Aichi Target. We watch wildfire smoke descend on U.S. cities, but we keep putting unprecedented levels of carbon dioxide into the atmosphere— Paris be damned. Income equality is worse than ever, our political leaders are knaves and idiots— in short, everything is stuck, and seems unlikely to get un-stuck: so why invest in a technology that promises to disrupt the status quo, even if the promised disruption is a shapeless one? The answer, of course, is that not all disruptions are the disruptions we need. Sometimes technology really does make us worse off. Or, in the larger number of cases, leaves things exactly as they were before, while the world spins on with all its problems and latent possibilities.