The Case for Curmudgeons
An academic conference I recently attended assembled a panel to discuss the “challenges and opportunities” of agentic AI in the field of computer science. The panelists (three professors and a senior scientist from industry) and a few outspoken audience members offered their reasonable views on AI’s impact on research and education—reasonable, though, I should add, somewhat tepid.
The discussion had undertones of caution and restraint. No one seemed keen to say anything definitive, for example, by taking a clear stance for or against the use of AI agents in coursework or in PhD research. Instead, large portions of the discussion sounded speculative and hesitant, brimming with rhetorical questions in place of answers, almost as if every spoken statement had been designed to maintain plausible deniability.
The speakers seemed to imply that AI’s impact on the field remains inconclusive, and that we–as researchers and educators–are in a state of transition whose endpoint is not fully known. Nor is there much, really, that we could do about it. But all in all, they seemed to suggest, there is no cause for panic.
Normally, you wouldn’t fault academics for being nuanced and avoiding hyperbole; it’s the least you’d expect from them. But the situation we’re in is far from normal. I for one wasn’t aware of the extent of challenges the field is facing, e.g., I hadn’t known that entire scientific workflows (literature review, experiment design, coding, validation, and manuscript writing) are being delegated to AI, or that AI-powered peer-review is now a prospect that’s receiving serious consideration by serious and influential people.
And thus my frustration with the excess of nuance and middle-ground rhetoric. It left me wondering whose interests are being served if the answer to every other question on such serious issues is “It depends”, or its more in-vogue cousin, “Well, yes and no.”
In response to such serious challenges, I find myself craving answers that aren’t technically more correct or rigorous–necessarily–than those provided by the panelists, but ones that reflect a stronger moral conviction, a scholarly attitude and ethos that, when I was young and naive, I imagined academics would have.
Lest this be mistaken for an endorsement of demagoguery and alarmism, I will give an example of what this moral certitude looks like, courtesy of Edsger W. Dijkstra, a Turing Award-winning computer scientist.
Prof. Dijkstra was known for his longstanding practice of writing essays on varied technical and non-technical topics, which he wrote by hand, copied, and distributed to colleagues. Those documents became known as EWDs, in reference to his initials. In EWD1179, he expresses his unfiltered opinion on the use of overhead projectors in university lectures. Projectors, he writes, are "a crutch for incompetent speakers who could not deliver a good lecture on their own" and are "profoundly inappropriate for lectures by scientists." He then describes candidate lecturers who, in his view, lack proper lecturing skills, as having "never been taught to speak in complete, unambiguous sentences." Finally, he laments the impact this would have on future generations of students with an emphatic "God help our students!"
As far as their impact on education, there is no comparison between overhead projectors and generative AI (the modern equivalent of the former, of course, being PowerPoint slides). But it’s not for reasons of analogy that this rant on projectors is relevant, rather that it demonstrates clearly an attitude that can only arise from strong scholarly–and moral–convictions.
Aside I should mention that EWD1179’s rant on projectors and incompetent speakers would only need a few word substitutions to become a fitting description of the six-minute PowerPoint presentations that took place at the same conference I'm referencing. For this, you must blame not the speakers but whoever thought a slew of six-minute talks with no breaks for discussion in between was an appropriate vehicle to communicate scientific research.
In other EWDs, Dijkstra addresses the impact of computers on education more directly, and he even had things to say about AI. I’ve included some excerpts and references in the footnotes. (End of aside)
Dijkstra’s skepticism of new technologies would be anathema in many modern academic circles. His belief in what a lecturer ought to be, by today's standards, is almost essentialist: lecturers are those who give good lectures; he pays no heed (How dare he!) to their resumes or any unrelated accolades they might possess. Despite the seemingly trivial nature of the problem, that of overhead projectors, his convictions seem strong and unambiguous, because to the likes of Edsger W. Dijkstra, scholarship is more than a means to an end; it’s a mode of conduct and a way of being.
I would certainly be stressed if the people looking over my shoulder harbored such immutable notions of what it means to be an academic, lecturer, or computer scientist. (In other writings, Dijkstra expresses stringent demands for programming which very few programmers today would be able to meet.) I can imagine the many ways in which I would fail to meet the standards of a hypothetical, modern-day Dijkstra—an elitist academic curmudgeon. (Whether failing to meet those standards is more or less dignified than failing to meet the far more opaque and arbitrary standards of modern academia is another question.)
Nonetheless, a curmudgeon is precisely what was missing from that panel discussion I attended. Someone who wouldn’t mince their words when asked about a technology that’s threatening the way science and education are conducted. Someone who believes these institutions deserve protection and regards it as their duty to protect them.
To avoid falling into the same trap of ambiguity that I've criticized others for, I'd like to end with a more concrete example, one which the panelists tiptoed around but never addressed directly:
A principal investigator supervises a PhD student. The PI realizes that the student is relying heavily on generative AI and LLMs to meet the demands of their project (imagine for yourself how heavily). Is it wrong for the PI to allow this to continue, look the other way (which, in my estimate, is a common scenario), or even encourage it? If the answer doesn't strike you as self-evident, or if you think that it depends on how reliable the models are or how often they hallucinate, then indulge me for a moment longer:
Mr. Miyagi instructs his new student, Daniel, to wash and wax his cars; supposedly, this is Daniel’s first karate lesson. The student is baffled; he cannot imagine what washing and waxing cars has to do with martial arts, but the teacher is relentless. He even shows Daniel the exact hand movement he must use: Wax on, Wax off. Later, it’s revealed (in the movie The Karate Kid, in case you’re not familiar with this reference) that the Wax on, Wax off movement was training Daniel’s muscle memory and preparing him for fighting.
Now imagine that instead of listening to his teacher, Daniel returns the following day with a motorized car-waxing device and waxes all his teacher’s cars better–and much faster–than he could have done by hand. How would the wise teacher react?
The hypothetical PI who’s ambivalent about their PhD student’s use of AI needs to ask themself whether there’s a deeper purpose to doctoral programs beyond–you know–waxing the PI’s cars.
"Inadequate reactions are only to be expected in the face of drastic novelty such as technology can inflict upon us" - EWD867
"The main criterion by which to judge our academic research is how it improves our teachable material" - EWD1209
"The question [of whether machines can think] is just as relevant and just as meaningful as the question whether submarines can swim" - EWD867