AI Is No Siren’s Call

Economists ought to be more careful in their use of classical analogies.

In a recent email to undergraduates, Harvard College Dean David J. Deming suggested that, given that reliable detection of artificial intelligence is nearly impossible, instructors ought to encourage the use of AI on take-home, writing-intensive assignments.

It probably did not surprise Deming that this view received some pushback.

Deming makes two mistakes in his argument: One, he errs in presuming that fairness ought to be the primary imperative of the College, and two, he misappropriates a Homeric analogy to argue for the Claudeification of the long-form written assignment.

Deming got off to a strong start in his September 2 communiqué. He rightly endorsed increased use of “AI-proof” assignments: in-class quizzes, blue book finals, oral examinations, and the like.

And on problem sets, the Dean seems to agree with my own view that capitulation on easily Chat-able problem sets of the sort assigned in introductory math and science courses is reasonable. Through a three hour final examination, we can make an at least reasonable assessment of a student’s command of linear algebra.

It is on long-form, writing-intensive assignments that I break with the Dean.

Deming argues that such take-home assignments create a problem for honest students: Since their unprincipled, cheating classmates are unlikely to be caught for using AI, the honest students fear falling behind, and are thus tempted to adopt AI cheating themselves.

Deming’s solution is to go full-AI — to not just allow, but actively encourage the use of AI on these assignments.

But there is much value in the sans-AI 20-page paper, the sort of assignment that requires meditating on difficult concepts over long periods of time, the parsing and synthesis of sources into a complex, multi-part argument, and many rounds of feedback and revision.

This kind of cognition, quite obviously, can neither be performed nor assessed within the confines of a classroom. Nor is it practical (though it would certainly be worthwhile) to lock undergraduates in the library for a few days without their technological appendages.

Deming is right that sorting out who exactly used AI on these assignments is an unpleasant, knotty business. Professors really shouldn’t be running papers through Pangram or using their questionable AI-detecting “spidey-sense.”

But is fairness all that universities should value?

Universities have a duty to preserve the intellectual capacities of mankind in general, and were we to do as Deming proposes, we might take a step closer towards “human enfeeblement,” as Jill Lepore has suggested.

One of the functions of a university must be to preserve important cognitive functions. And losing that capacity does not just mean an overreliance on AI models, but also the loss of the very fruits that often emerge from such extended cognitive labors.

Certainly, temptation will arise, but it is not, as the Dean suggests, an irresistible temptation. Some students will actually abstain from AI if you tell them to.

Deming likens generative AI models to the sirens that menace Odysseus in Homer’s “The Odyssey.” The half-woman, half-bird sirens sing a song of such temptation that they doom any sailor within earshot — without exception — to approach their rocky islands, a fate of certain death.

Deming argues that, much as Odysseus roped himself to the mast of his vessel so that he might hear the siren’s song and live to tell the tale, we must impose certain inflexible pedagogical restraints to avoid the “temptation” of improperly using AI.

Deming prefigured his email with a January Substack post that close-reads The Odyssey, stressing that the temptation of the sirens is not carnal, but intellectual: “unlimited knowledge and wisdom without effort,” as Deming puts it.

“This is the perfect metaphor for learning in the age of superintelligence,” he writes. This writer is not so certain about the fidelity of Deming’s conclusion.

Is it really impossible for students to resist the temptation to use AI? Even their own Dean says that they cannot be blamed for cheating, for some supernatural force has summoned their minds to misuse AI!

Indeed, it is unfair and demoralizing that some students should not use AI and others should, and that the latter group of immoral anti-intellectuals might receive higher grades than they truly deserve.

But there are ways to ameliorate this unfairness: namely, by increasing the weight of AI-proof assignments and conducting oral examinations for research papers.

We mustn’t fully nix the 20-page take-home paper. It is worthwhile to keep for the students who don’t live in Deming’s faux-Homeric world, for whom AI does not constitute an irresistible siren’s call.

Is it not worth it to have a bit of unfairness to prevent the cognitive enfeeblement of all of humanity? The students that will cheat will cheat, wasting their own time as well as everyone else’s.

The rest of us, presumably, will find the will to avoid swimming to Anthropic island from the ship of academia and consorting with Claude beneath the olive tree. And we will be better off for it.

Andrew W. Shlomchik ’29, a Crimson Editorial editor, lives in Quincy House.