What AI changes about 'Viewpoint Diversity'

Reviewing a book in which I appear

While the grownups are fighting, students are off widening the Overton chat window on their large language models (LLMs) in the privacy of their dorm rooms. The viewpoint diversity available to students is perhaps only tangentially related to how their professors vote and whether they have tenure.

I wrote this a month ago, after the first major review of Viewpoint Diversity: What It Is, Why We Need It and How to Get It, edited by Bernard Schweizer and John Tomasi, appeared, on March 20, in Inside Higher Ed. The reviewer, John K. Wilson, did not like the volume, calling it ideologically narrow and arguing that “[e]xpanding protections of academic freedom and tenure would do more to expand viewpoint diversity than anything proposed in this book.” Wilson doubled down in a second review, also negative, three days later, saying that focusing on political representation “betrays academic standards.” I have seen no other reviews, negative or positive, since, except for Jesse Singal’s review of Wilson’s reviews on his Substack, and John Tomasi’s letter to the editor responding to the review on March 23.

I am weighing in because most essays in the volume, including mine, entitled “Viewpoint Diversity in the era of AI,” have little to do with the shots fired by Wilson and returned by Tomasi and Singal. Many of us are writing about the student experience, not faculty politics.

“Every essay in this volume is written on the assumption that viewpoint diversity is a human project,” my essay begins. I argue that AI changes everything about what it means to have a viewpoint and to express it. I suspect the wider public is beginning to recognize this. When over 85% of students are chatting regularly with their LLMs, diverse views are no longer “rare” and may no longer need to be protected. I also suspect that most students are far less interested in their “own” views than the views open to them to try on. “Generated” diversity is a non-trivial complication to the entire viewpoint diversity movement. Academic leaders might think about clawing back the funds dedicated to “ensuring” diversity and spending the funds instead on faculty AI training.

Wilson thinks that a book about viewpoint diversity should have more viewpoint diversity and that’s what my essay is about. I would have hoped he would have noticed and mentioned it (which he did not), because my whole point is that there is a superabundance of viewpoint diversity to be had in the AI era. “Nobody in this volume wants a pluralism that is ventriloquism, but that is the reality online,” I observe. Humans are authoring a narrower and narrower slice of what is available in the marketplace of ideas.

Our Viewpoint Diversity volume should force a clarification (among my fellow contributors particularly) about the relationship between views held and views expressed. While every contributor defends open inquiry, pluralism, and the value of disagreement, there is little agreement about whether it’s important to assume or insist that a speaker “holds” the views expressed. Some do. Jon Haidt quotes John Stuart Mill: “Nor is it enough that he should hear the opinions of adversaries from his own teachers, presented as they state them, and accompanied by what they offer as refutations. He must be able to hear them from persons who actually believe them.” Tyler J. VanderWeele argues that it would be advantageous to have a faculty member who actually “holds the view” in question. Nafees Alam talks about open exchanges with students “holding” views with confidence. Jesse Singal characterizes the viewpoint diversity argument as advocating for the presence of individuals “holding a wide variety of views.” John Inazu writes about “firmly held convictions,” his own and others. Ilana Redstone focuses on the many reasons someone might hold a view. Yascha Mounk begins by observing that AI can generate viewpoints on any subject but quickly paraphrases Mill to point out that “it is crucial to hold our beliefs as living truths” and that we should be exposed to “a genuine diversity of views.” Views, it seems to these writers, must be genuine to matter.

The section on “Viewpoint Diversity in Literature” takes the opposite view: Bernard Schweizer, Richard North Patterson, and Henry Louis Gates, Jr. push back against identity-based voice to argue that authors should be able to write what they want in the voice they want. “Should empathy and imagination be allowed to cross the lines of racial identity?” Patterson asks. The assumption that you should “look like the subject to master the subject,” as Gates puts it, is a prejudice. “Any teacher, any student, any reader, any writer, sufficiently attentive and motivated, must be able to engage freely with subjects of their choice,” he adds. Literary authors should be able to bring to life voices that are not their own. The arguments in this third section suggest that viewpoint diversity is perfectly acceptable when the voice is fictional. Not all “genuine” viewpoints need to be “embodied” and “held.”

There is a vast gray area between the poles of “holding” and expressing an “authentic” view and writing fiction that involves characters who don’t look like you saying things you don’t believe. This gray area is where most students—and faculty—live their lives. My “genuine” “firmly held” political or religious views are rarely expressed. They are none of anyone’s business. (I’ve come out strongly against HxA’s product Sway as an invasion of student privacy.) I have never expressed my political views in the classroom and would never ask my students to express their views. If a class were reading Uncle Tom’s Cabin, I would expect students to read passages aloud, if asked—let’s say a monologue by Simon Legree—whether or not they agreed with the fictional Legree’s viewpoint as expressed.

The distinction between “embodied” viewpoints and expressed viewpoints matters as LLMs enter the conversation, joining humans and fictional characters. What does it mean for a viewpoint to be “authentic”? Across the essays in the volume the noun “authenticity” is used largely negatively (the basis of “cultural appropriation” and identity politics) while “authentic” and “authentically” are used positively (“engaging authentically and graciously across differences”).

Jonathan Zimmerman is absolutely correct that most universities don’t do much to differentiate great teaching from terrible teaching (I have written about this). But I am not sure I agree that great teaching requires trying to elicit diverse student voices in class. I wonder whether “drawing out” quiet students is always generous, or whether it can become a demand to perform inner struggle for the benefit of the class. Perhaps silence is the desire to be left alone with one’s own thinking. Nadine Strossen argues that institutions should not prescribe orthodoxy, and they must not demand that students disclose their beliefs as the price of participation. While Bret Stephens wrestles with whether certain views about Israel deserve a platform, I wonder whether a student sitting quietly should be valued for her silence as much as her voice.

If the goal for some of my fellow contributors is to bring more conservative voices into the classroom, on what basis does it matter that those voices be physically embodied, beyond the fact that this was Mill’s view? I oppose going around looking for “authentic” and “held” views because it requires prying into somebody’s voting life, which is unsettling and unhealthy. “None of your business” is a perfectly fine position for faculty and students to take. (I took it a lot during the “implicit bias” craze and I appreciated Jesse Singal’s essay on the overreach during these years.)

With AI, students can summon up points and counterpoints to any line of questioning, complete with supporting data, before, after, or even during class. Every student can look up what a Burkean, a Marxist, a libertarian, or an isolationist-populist has said about any subject on the planet. There’s no “wrong” view to fear. Students grasp that the LLM that generates views does not “believe” those views. Student embrace of AI may be the social and logical endpoint of a campus culture that insists upon a public airing of views.

buy this book!

I will continue to defend intellectual privacy against the demand made from advocates of both orthodoxy and heterodoxy that students make their inner convictions public. I argue, in my essay, that a student “has the right to try on an argument like a coat, to see how it fits, without anyone assuming it is now part of one’s skin.” I am hopeful that the era of the classroom performance of “held” rather than “expressed” viewpoint diversity will end. Let a thousand fictional or LLM identities bloom and disagree with one another with respectful abandon.

My essay may have flown under Wilson’s radar because it was described in the introduction as arguing that AI threatens authenticity “by providing ‘pseudo-diversity’—responses that lack true conviction, nuance, and human responsibility.” That isn’t quite right and I probably ought to have spoken up during the editing process. Rather, my point is that the demand for authenticity, the insistence that students reveal their genuine convictions, is coercive and counterproductive when all viewpoints are available to try on outside of class.

Final note: the book is excellent and my intervention is only this: if the viewpoint diversity movement cannot make room for the argument that the demand to perform one’s beliefs is itself a form of orthodoxy, that the machinery of openness can become a machinery of surveillance, then the movement has a problem larger than a bad book review.