Unpopular Opinion: My Liberal Arts Degree Is Better Than Your Law Degree

I have been telling my daughter—loudly, repeatedly, with the confidence of a man who has seen a few economic cycles come and go—that her $92,000 Wake Forest tuition is money she might regret spending. She's studying Art and Psychology, two disciplines that, by the traditional calculus of parental anxiety, rank somewhere between "podcast host" and "aspiring influencer" on the career security index. I have made the speech, probably more than once. Learn to code. Consider nursing. Have you thought about the trades?

I got particularly excited about mobile pet grooming a few months back—the woman who comes to our house charges $170 an hour and is booked six months in advance, which puts her effective hourly rate somewhere north of most attorneys I know. I offered to buy my daughter her first two trucks if she'd skip college entirely. She declined, politely, in the way children decline suggestions they find charmingly out of touch.

I may have had it exactly backwards.

Not about the pet groomers and the plumbers—they really will be the new aristocracy, and I'll die on that hill. You cannot outsource a clogged toilet to a language model, and the woman with the mobile grooming van has what economists call pricing power. But about my daughter, about what she's actually learning when she studies how humans create and how humans think, there's reason to believe I've been measuring with the wrong instruments entirely.

The Credentials That Flattened

I went to Connecticut College in the late eighties—theater, English literature, philosophy, the holy trinity of unemployable majors. My parents wrote the checks anyway, probably lying awake at night wondering what exactly one does with a deep familiarity with Platonic dialogues and the staging conventions of Jacobean revenge tragedy. For most of my career, I wondered the same thing. I fell into advertising, then data, then martech, then whatever we're calling the enterprise software racket these days, and the skills that seemed to matter were always the quantitative ones: the SQL queries, the attribution models, the ability to sit across from a CFO and discuss pipeline contribution without visibly sweating. Nobody in a performance review ever asked about my thoughts on Heraclitus.

Here's what I've started to notice, though, working with AI systems every day—and this is the part my friends in law and finance don't particularly want to hear. The skills they spent three years and $200,000 acquiring, the ones that justified the credential and the debt and the delayed entry into actual life, are now available to anyone with a ChatGPT subscription and a few hours to practice. Legal research, contract drafting, summarizing case law, the entire apparatus of what a junior associate does in their first years of practice—an AI handles it in seconds. Not perfectly, not without human oversight, but well enough that the person billing $400 an hour to do the same work starts to look less like a value-add and more like an expensive habit the client hasn't yet realized they can break.

The same pattern holds in finance, in consulting, in most of what we've called "knowledge work" for the past fifty years. The protected guilds built walls around information access and specialized procedures, and those walls made sense when the procedures were genuinely difficult to learn and the information was genuinely scarce. Now the information is everywhere and the procedures are encoded in systems that anyone can use. The moats are evaporating, and faster than most people inside them seem to realize.

But there's a category of skill that doesn't flatten in the same way—the capacity to know which question to ask in the first place, to sense when an audience is disengaging, to recognize the moment when a chain of reasoning has gone epistemically sideways despite being internally consistent, to distinguish between an answer that's technically correct and one that's actually useful. These are the fuzzy skills, the ones that don't show up on a transcript, the ones I thought were basically useless for two decades of my career. It turns out they might be the whole game now, or at least a much larger part of it than anyone anticipated.

What Theater Actually Teaches

Nobody tells you this about studying theater, but it's essentially applied psychology with a minor in benevolent manipulation. You learn to read a room—not metaphorically but literally, as a technical skill you practice until it becomes instinct. Where are people looking? When did their attention drift? What happens to the energy when you hold a pause one beat too long, or not long enough? You learn that dialogue isn't really about what characters say; it's about what they want from each other, the tactics they're deploying to get it, the subtext running beneath the surface of every exchange. Every line is a move. Every scene is a negotiation conducted through the medium of words that are rarely saying what they mean.

This is, I've come to believe, closer to what prompting actually requires than anything they teach in computer science programs. When I work with Claude or GPT-4, I'm not entering commands into a terminal and waiting for deterministic output. I'm shaping a conversation, reading the responses for signals about what's working and what isn't, adjusting my approach when something falls flat, trying different angles of approach when the direct path isn't producing what I need. I'm thinking constantly about audience—in this case, an audience of one very capable and very literal-minded system—and crafting inputs that will actually produce the outputs I'm looking for rather than the outputs that would satisfy a less careful question.

The people who struggle with prompting, in my experience, tend to be people who've never had to hold a room's attention through sheer force of communication. They've never felt an audience slipping away and had to pull them back in real time, adjusting on the fly, reading the subtle signals that tell you something isn't landing. They think you can just say what you want and the machine will figure out what you meant. You can't. Communication is a craft, and theater majors—whatever else they might lack—know this in their bones.

What Philosophy Actually Teaches

Heraclitus, one of the pre-Socratics writing around 500 BCE, observed that you cannot step into the same river twice. The water is always moving, always changing; by the time you've named what you're standing in, it's already become something else. This is, more or less, the epistemological situation of working with probabilistic AI systems, though I doubt Heraclitus had transformer architectures in mind.

Every interaction is a new river. The same prompt yields different outputs depending on context, on what came before, on factors that aren't always visible or predictable. The system that worked beautifully yesterday produces something slightly off today, for reasons that may or may not be discoverable. If you're expecting deterministic reliability—same input, same output, every time, the way a spreadsheet formula works—you're going to have a frustrating experience and probably conclude that the technology isn't ready for serious use. But if you've spent time with the Greeks, with their frameworks for thinking about change and knowledge and the limits of certainty, you might recognize this as a different kind of tool requiring a different kind of relationship.

Plato's dialogues aren't about arriving at the right answer. They're about the process—what he called dialectic—of getting closer to truth through rigorous questioning, through the willingness to discover that what you thought you knew doesn't survive scrutiny. Socrates didn't lecture. He asked, and asked, and asked again, each question refining the last, each answer revealing new gaps to probe. This is iterative prompting before anyone called it that: you're not searching for a single correct answer so much as developing a relationship with the problem, circling it from different angles, testing the boundaries of what the system knows and where its knowledge breaks down.

And epistemology—the branch of philosophy that asks how we know what we claim to know, what would count as evidence, what are the ways our reasoning might fail without being visible on the surface—might be the single most practical thing I studied in college, which would have been news to me at twenty. Because when an AI gives you an answer, the relevant question isn't whether it sounds correct or whether it's formatted nicely or whether it matches your intuitions. The question is: how would I even verify this? What would falsify it? What are the failure modes I should be watching for that wouldn't announce themselves?

The lawyers and the bankers and the consultants were trained to take inputs and produce outputs, to be excellent at execution within well-defined parameters. But they weren't necessarily trained to interrogate the foundations, to ask whether the question itself was well-formed, to notice when their confidence is socially reinforced—everyone around them believes the same thing—rather than epistemically justified. That's what philosophy is for, and it turns out to matter quite a lot when you're working with systems that can generate confident-sounding nonsense as easily as insight.

Three Classes

So here's what I've come to think about who wins in the economy that's emerging—and I want to be careful about the word "wins," because I'm not sure it's the right frame, but it's the one people seem to want.

There are the trades—the plumbers, electricians, HVAC technicians, mobile pet groomers, anyone who works with atoms instead of bits and whose skills don't commoditize because software can't fix your toilet or calm your anxious dog. They'll do fine, better than fine probably, and their pricing power will only increase as the supply of people willing to do physical work continues to shrink relative to demand.

There are the builders—the engineers and ML researchers who actually construct the AI systems, who understand the mathematics and the architectures and can make the technology do new things. They'll do well for obvious reasons, though even here the ground is shifting in ways that aren't entirely predictable.

And then there's a third group, the one nobody quite anticipated: the people who spent four years learning how to think rather than what to think, who can close-read a text and spot the gap between what's said and what's meant, who know how to trace an argument back to its foundations and test whether they hold, who understand that every communication is a kind of performance and that getting someone to see what you want them to see—whether that someone is a person or a very sophisticated language model—is a skill you can practice and improve.

We're the ones, it turns out, who know how to drive these machines. Not because we understand the technology—we mostly don't, and that's fine—but because we understand humans, and we understand the strange art of bridging the gap between what a system can do and what a person actually needs. That bridge is where most of the value gets created or lost, and it's a bridge that liberal arts education, whatever its other limitations, builds surprisingly well.

The Apology

To my daughter, finishing her first year at Wake Forest, studying Art and Psychology, learning how humans create and how humans think: I was wrong about the degree, and I'm sorry for the speeches. Art teaches you to make things that didn't exist before—the creative act itself, the skills that AI can amplify but cannot replace because they require the thing AI doesn't have, which is a genuine point of view. Psychology teaches you why people do what they do, the messy and irrational and endlessly fascinating machinery of human motivation. Together they might be better preparation for whatever's coming than law school or business school or probably anything else I could have told you to study.

The $92,000 isn't wasted. It might be the best investment I ever made.

Though I still think she should learn to fix a toilet. You never know.