How (Not) to Meet the AI Moment

Artificial intelligence moves quickly, and the discussion around its role in education has barely kept up.

While a few departments have implemented across-the-board bans, most students are navigating a completely new landscape on a per-syllabus basis. Decentralization, normally a virtue of academia, has so far failed to meet the moment, and academics have been left responding to the rise of large language models as individuals.

Cataloguing these responses is surprisingly productive, since it forces us to be explicit about the purpose of a university. I will argue that the best proposals share a common trait: namely, solidarity, which is my loose term for the commitment to common interests and a shared future.

While solidarity leaves much unspecified, it is a useful litmus test for professors insofar as they have obligations to students.

Many prominent viewpoints do not pass this test. One example is what I will tendentiously call “myopic enthusiasm”: Because we can use AI to solve problems more quickly, we may be able to stop training people to do theoretical science.

This was best expounded in an Anthropic blog post earlier this year by Harvard Physics professor Matthew D. Schwartz. Under this model, graduate students are simply labor, with limitations: we eat, sleep, and learn slowly. LLMs can generate the same outputs for less. Schwartz helpfully makes this comparison explicit, and suggests that graduate students transition into more experimental fields.

What this correctly avoids is the temptation to explain away the successes of LLMs, by disingenuously moving the goalposts on what counts as real intellectual work. But myopic enthusiasts elide an important distinction between the best-case use for LLMs (someone with decades of expertise, weeks to comb through every output, and the right incentives) and typical-case use (everyone else using AI without understanding the details).

The reality is that only a handful of people alive could check Claude for particle physics as well as Schwartz. I am no more able to usefully contribute to quantum field theory today than I was in 2021. Whether or not this gap will close in the near future is an open question, but betting everything on the answer seems ill-advised.

Another take one often sees is the “tenured eulogy.” On this view, students are badly situated actors in the global and terminal decline of intellectual work, and professors have nothing left to offer them.

This is tenure-inflected because the people writing eulogies have little exposure to the downsides of this future. For example, they often raise the possibility of pursuing science as a hobby, once we have instituted a Sam-Altman-backed universal basic income.

However, the prospect that anything remotely like science thrives in an AI-overlord monoculture such as this one seems vanishingly remote. This is because science benefits enormously from the idiosyncratic and path-dependent perspectives of limited human beings, who are trained in a staggering diversity of contexts.

An alternative position is what I described above as “solidarity.” At its core is the premise that young people deserve better than to have the terms of our education dictated by Silicon Valley. There are professors who have made this case: Michael Harris at Columbia University, most incisively, and Jill Lepore in these pages.

Solidarity recognizes that human beings will only bring about a future worth wanting through mutual support and interdependence. Evidence for its currency is starting to trickle in.

Many have pointed out the risks (moral, technical, psychological) we incur by offloading our thinking onto LLMs. The leaders and backers of these companies are not motivated by intellectual curiosity or the careful development of students. But it takes solidarity to believe that we can act to mitigate these effects, rather than accepting them as inevitable consequences of the market economy. It bears repeating that young people are, crucially, much more exposed to these downsides than established professors, who can now direct LLMs effectively after decades of traditional work.

The solidarist should learn from the myopic enthusiast and embrace LLMs when they can accelerate research and enhance learning. However, the solidarist needs to go further, defending the claim that it will always be worthwhile to think, read, write, and understand. Promises about the reduced economic value of these activities simply have no bearing on this commitment.

Professors sympathetic to solidarity could begin by affirming that the years required by students to develop are not wasted, a premise we can no longer take for granted. In this way, LLMs have simply accelerated an identity crisis for universities that has been decades in the making. As some MIT professors have noticed, AI agents can now replace undergraduate researchers on coding and theory projects. But it is hard to justify why the professor so inclined should work in a university setting.

What is the heart of this “university setting”? Many of us are here because we want to understand nature through the sciences. Others grapple with our place in the world through the humanities. Some of us want to teach younger people to do the same.

The best version of Harvard is simply one where we pursue these goals in solidarity with each other. This has always been a privilege; perhaps now it has become a responsibility.

Juan Carlos Fernandez del Castillo is a fourth year Ph.D. student in the Biophysics Graduate Program.