Adapting to the Future of AI

In her September 5 article for The Atlantic, “Why I Want More AI at Dartmouth,” Dartmouth President Sian Beilock argues that colleges should teach students to use AI critically and productively rather than ban it. Otherwise, she says, universities risk becoming irrelevant to the working world their graduates will enter.
Many, of course, disagree with her. The first comment on the article says:
“I’m teaching a course this semester on Modern Political Thought – Machiavelli, Locke, Rousseau, Burke, Marx, Mill. The last thing my students need is AI-summaries or AI-quiz problems or what have you. They need to develop the ability to read canonical texts closely and carefully. The world outside my classroom has more than enough AI. And my students are excellent at picking up new technology. But they are not close readers in the way they need to be. The author has no idea about what they are talking about.”
I disagree with both Beilock and the commenter. They share the same flawed premise: that the central function of college is to endow students with knowledge and skills they will later use at work. The weakness of that premise is especially clear in the professor’s comment. Students who, by the professor’s own admission, cannot “read closely” are supposed to study Machiavelli, Locke, Marx, and other classical writers, then somehow abstract from them a skill useful in jobs that have almost nothing to do with political thought. I find that leap hard to believe.
The real role of college was never endowing students with directly applicable skills. It was to certify intelligence and conscientiousness and, at a place like Dartmouth, to endow students with social capital. If directly preparing students for work were the goal, Dartmouth’s Social Analysis distributive requirement would be difficult to explain. The chance that any STEM major will use what they learn in such a course in their professional life is infinitesimally small, yet they still have to take one.
As a test of conscientiousness, however, the requirement makes perfect sense. Much of professional life involves doing things you have no interest in doing, and doing them well anyway. A class you would never have chosen might be the perfect test of your pain tolerance.
It is a nice goal to give students skills relevant to the job market. Unfortunately, a college classroom cannot achieve this goal due to the nature of jobs. Anyone who has held a job for more than a month knows that a task can sound simple at a high level, then reveal a cascade of consequential details once you actually try to do it. John Salvatier describes this in his essay “Reality Has a Surprising Amount of Detail.” The fiddliness of reality is universal, and no professor can endow to his students in four years what a month on the job would not, even if they did not attend college! A student can learn how every single instrument in a biology lab works, yet will be clueless once he starts his first job at a big pharmaceutical company, as the actual skills are the developed mental shortcuts employees develop with years of experience. Liberal Arts colleges do recognise the feebleness of endowing students with “applicable skills,” and have shifted to more abstract teachings.
At its current stage, AI threatens the core function of college, but not because it stops students from learning properly. There was little to learn in the first place. What students had to do was prove themselves. AI makes the barriers to that proof easier to pass because much of the curriculum tests work that AI can now do better than most students. If colleges cannot tell whether a student completed the work, they cannot credibly certify the qualities that work was meant to reveal.
Until colleges develop new assessments that can withstand AI use, they should hold back on its use in graded work. Otherwise, they risk losing the one thing employers have long relied on them to provide: a credible filter.