The Promise and Peril of Liberal Arts Education
It's "AI" proof in some ways, but critical thinking still isn't the thing we want it to be
I’m stuck proctoring an exam this morning, which on the bright side leaves me with a bit of time in which to blog. I’ve also had a few interactions recently that have had me thinking a bit about what it is that I’ve chosen to spend my career doing, so you’re going to get some rambling about the state of liberal arts education.
One of the things that’s prompted this reflection was a question from a student at one of our spring Open House events for students who have been accepted to Union. I was staffing the Physics and Astronomy table at the “academic expo” right after lunch, to answer questions about our programs from students and parents1 and try to convince them that coming here would be a good choice.
One of these questions was from a student who was deciding between studying engineering at Union or one of a couple of larger technical universities. He asked “Given the AI boom that’s happening, wouldn’t I be better off going to a more technical place, where I can really go in depth on how these things work?”
Setting aside for the moment technical quibbles about the exact nature of “AI2” in the year Reiwa 8, this is a reasonable question to be asking, and indeed, we’ve seen a lot of variants on this basic theme. But as I said to him, I think the premise of the question is exactly backwards— that is, I think the rise of “AI” is an argument for the liberal arts education model, rather than a narrower more technical education.
I say this because the transformative effect of currently-existing (and reasonably forseeable) “AI” models comes in the form of lowering the threshold needed to do somewhat technical things. You can use LLM systems to “vibe code” tools for all manner of projects, and thus accelerate analytical progress in those areas— the resulting programs won’t necessarily be the most optimal way of doing those tasks, but there’s a good chance that you’ll end up with something functional in a fraction of the time it might otherwise have taken. A greater depth of technical knowledge might help you to spot the inefficiencies in the LLM code, but for a huge amount of analyses that’s not going to represent any material advance in the quality of the results, and certainly won’t speed up any of the things that you would use this kind of code to do.
What these systems can’t currently do, and arguably may never be able to do, is to decide what problems to attack and what to prioritize. These generative models make it easier to do a wide range of things by automating various technical processes— dividing up data sets, extracting fit values, etc.— but deciding which of the nearly infinite number of possible analyses are worth doing and what actions to take based on the results are questions of a different sort. These are ultimately decided not by technical factors, but by individual and societal priorities and values.
Those questions of priorities and values demand judgement that isn’t rooted in narrow technical expertise, but draws on a broader range of knowledge.Deciding what to study with technical tools requires some knowledge of historical and cultural context, and some sense of how to assign priority to some problems over others. Deciding how to act on the results of a given study again requires knowledge of contexts, and also how changes are likely to play out in a real society, and how to communicate new ideas and information to people in an effective manner.
That is, making judgments about what to do with “AI” tools is a problem that demands liberal arts education. A deeper but narrower education in the technical underpinnings of “AI” isn’t going to help choose the best problem to study, or the best policy option once the results are in— for that, you need both the ability to understand the essential quantitative features of whatever data you’re working with and the broad context to know why it matters. You will be better served, both as an individual and as a member of society, by having a basic grounding in a broad range of topics letting you place technical results in a human context, and vice versa.
I’m not sure how effective my pitch was at convincing this particular student, but I do believe that it’s the best argument we have. Honestly, it always has been the best argument we have for the value of a liberal arts education, but the ways that these “AI” tools are being used only make the case stronger.
The other thing that’s had me thinking about Big Picture stuff regarding liberal arts education was a reminder from Facebook of a “strategic planning” process the better part of a decade ago where we were asked to come up with a compact description of our goals for our students. My bumper-sticker summary of the end result of a liberal arts education was:
Students should be able to analyze a situation, decide on a course of action, and advocate for their choice
I was pleased enough with this to tweet it out (thus putting it in the queue of “memories” for Facebook to throw at me), and still largely agree with it as a compact summary of what it is that we provide in skills-based terms. It encompasses both some level of technical expertise— analytical ability— and also the oft-cited “soft skills” of communication and persuasion, and the combination of those is one of the big selling points of the liberal arts model.
On reflection, though, there’s a problem with this as a definition that is sort of the flip side of the pro-liberal-arts case I offered above. The issue, again, is that this phrasing is primarily about capabilities, but those are downstream of values and priorities. Education gives students the tools they need to analyze a situation and advocate for a policy option, but choosing which option to advocate for is another matter. That comes down to individual and societal values and priorities, which ultimately is not a matter of practicable skills.
Which is, ultimately, the dangerous side of the liberal arts approach. The combination of technical analysis and persuasive ability can be a very powerful thing, but both of these are ultimately just tools that can be put to use in pursuit of a wide range of possible goals. Whether the end result is a Good Thing, though, comes down to whether you as an individual agree with the use to which those tools are being put. The same analytical and rhetorical skills can be used to either improve the world by driving wise policy or make things worse by tricking people with superficially convincing arguments. As I noted all the way back at the start of this Substack, “critical thinking” by itself isn’t the answer to anything.
I’ve been very reminded of this in a very visceral way recently, in a context that I shouldn’t say too much about, and it’s been a very frustrating and troubling experience. Not least because it isn’t really a problem that’s amenable to being addressed through education— the fundamental questions of values and priorities aren’t things with a clear and objective answer, or even a broadly shared consensus answer. Which is at the heart of a lot of the more intractable problems we face, in higher education, and in society as a whole.
I think probably the only thing that can be done is to encourage the practice of being as up-front as possible about the underlying values and priorities going into our decision making. That can both make the logic of the decision a little more transparent to the outsider and, if done well, can somewhat head off the possibility of rhetorical distortions of those core values. It’s a tricky balance, though, and probably doesn’t allow for a satisfying resolution.
Not the cheeriest topic, I guess, but it’s helping pass the time while students work basic problems in intro mechanics. If you like this kind of thing, here’s a button:
And if you feel so moved, the comments will be open:
And at least one person who had recently read one of my books, and just wanted to say that he’d enjoyed it. Which, you know, thanks, that made my day…
OBCatchphrase: Roiling cauldron of linear algebra being marketed as artificial intelligence.
We should fear a world where it is easy to get others to adopt our values.
That said, values can be shaped by where we spend our time, who we spend it with, and the culture of how we interact with each other.
Values of open debate, tolerance, mutual respect can be modelled and fostered in an academic setting. So can encouraging novel thinking or subject mastery. And there are certainly works which encourage empathy with people in hard situations which can help sensitivity to such situations in the readers' lives.
No recipe but all this should be carefully considered and constructed.
> I think probably the only thing that can be done is to encourage the practice of being as up-front as possible about the underlying values and priorities going into our decision making. That can both make the logic of the decision a little more transparent to the outsider and, if done well, can somewhat head off the possibility of rhetorical distortions of those core values. It’s a tricky balance, though, and probably doesn’t allow for a satisfying resolution.
I think the problem is though, that if you know the person won't agree with your values/priorities you just won't be explicit about them. I know I've been guilty of it. To use an example no one else cares about, I think names on a Wikipedia page should be romanized in the language a person speaks, but if there's some busybody replacing them with the romanization in the majority language because of the National Agenda(TM), I'm hardly going to bring up my values when I'm telling them to stop.