The skill I'd teach students for the AI era

Resolved: You should learn how to debate.

This essay is a bit lighter, inspired by an activity I loved growing up. It might be of special interest to educators or parents of teens; consider sharing it with one you know.

---

At OpenAI, I studied “AGI readiness”: how the world might prepare for AI that, one day,¹ matches human abilities at basically everything.

This would be a remarkable shift, and so people have questions. I’m often asked: What should students be studying, to be ready for that world? Where should schools focus?

The honest answer is that I don’t know. I expect the job market to be quite tough; AI continues to get cheaper and more capable.² We face other issues, too, like the extreme dangers of AI systems we don’t know how to control, as the top-cited AI scientists of all-time have warned. The AI future is heavy stuff.³

But I’m not content with “there’s nothing to be done,” so I’ve kept racking my brain, and I’ve found myself returning to the education that shaped me. If I were a teacher, a principal, or just a concerned parent, what would I lean into today?

My most important education came from competitive debate: learning to think on my feet, to build the agency to accomplish hard goals, and to relate to other people.

I expect these skills to become even more important as AI improves; we should be teaching more teens to debate.

Intro to debate

The core of debate is to argue, competitively. Back in my high school days, some debate topics were near and dear to students; should standardized exams be required for graduation? Other topics—like what would justify nuclear war—we hoped never to confront for real, but made us grapple with important ideas.

A judge watches the competitors debate, then votes for a winner. To convince the judge of your position, your tools are an initial argumentative speech; a dramatic cross-examination where you get to posture like you’re on Law & Order; and a series of rebuttals in which you respond to your opponent’s arguments.

People have this idea that debate is about flowery language and gravitas—perhaps because ex-debaters emphasize the Law & Order-ly cross examination period. There’s some of that, but really competitive debate is more like a puzzle—one that requires learning about the world, understanding how arguments fit together, and devising a strategy to ultimately prove that you’re right.

If you already buy my argument about the importance of debate, I’ve footnoted some ways to build more debating into students’ education.⁴ But if you’re not yet persuaded, let me go through why debate is so helpful, followed (of course) by some counter-arguments.

Learning to think on-the-fly

Most homework assignments are ChatGPT-able in one form or another. Debate isn’t; you have to think in real time, without the crutch of AI.

Imagine walking into a debate round, and you’ll be speaking second, rebutting your opponent’s claims. You know the topic—that they’ll argue for a more lenient US drug policy—but you’re not sure exactly how they’ll make their case.

You encounter a new argument and need to diagnose it: What is it claiming? What’s the basis for it being true? Why might it be wrong? To win, you’ll need to do this well in just a few minutes of prep time—you and some sheets of paper, deconstructing their position and deciding your response.⁵

Now, I’m not naive; I’m sure some competitors will rely on AI for writing their initial positions. But hopefully you then get to mop the floor with them, drawing on the advantages of actually engaging with the topic in-depth: being able to explain your arguments more clearly, knowing them well enough to answer counter-arguments, understanding how they interact with other positions. People often mention the importance of “critical thinking skills” as AI becomes more capable, and the in-round debate experience is exactly what I picture; I can’t imagine a better training ground.⁶

Depending on the type of debate, you might also learn to talk really fast—like, 300 words-per-minute fast. This pushes you to think really fast, and makes for a decent party trick at a friend’s talent show 15 years later. (Especially if your alternative is a poorly played piano rendition of “When The Saints Go Marching In”—not that I’m speaking from experience.)

Competition is good

AI will eventually outperform people at many intellectual tasks, probably even at debating.⁷ But I still see merit in choosing an intellectual activity and pursuing it hard. People didn’t give up studying chess when Deep Blue defeated Garry Kasparov back in 1997; there’s so much to gain from dedication.

Something about the adversarial setting of debate especially sharpens your thinking: When your arguments need to stand up to an opponent, you learn to think deeper, to push yourself, to get more creative and strategic. It’s hard for me to imagine learning as much discipline and goal-directedness without this push. Maybe debate felt especially intense because I wasn’t much of an athlete, and so that sense of “trying to be the absolute best at something” was pretty rare for me.⁸

Once I started debating, I viewed my education more actively: I supplemented my classes with other useful-seeming topics; I got comfortable reading journal articles and interpreting experimental results; I picked up niche philosophy books that other students hadn’t read. Debate rewarded people who built unique expertise—and I expect that to remain true even as AI takes on more intellectual work. There will be unfilled niches out there, and competitive debate is good practice for finding them.

Understanding other people’s perspectives

Debating well isn’t just a question of being right, but actually convincing people of your rightness. To win the judge’s ballot, you need to understand how they’re thinking and land your arguments for them specifically. Basically, you need to learn to be attuned to other people.

The quick feedback from a debate round—whether you won or lost, and why—is gratifying: You get to explore how someone interpreted your arguments, why they weren’t persuaded, or what would have made your arguments stronger. This kind of exploration can be rude in some social settings; not so in debate, as long as you do it kindly. If you don’t get to really dig into someone else’s understanding, how will you learn to be clearer?⁹

A useful quirk of debate, too, is that it is switch-sides: across multiple rounds, you’ll argue both for and against the topic. This turns out to be hugely valuable for understanding the debate topic in detail, and for pressure-testing your beliefs in general.¹⁰

Why not debate

Of course, debate isn’t without downsides. For one, if you’re constantly in debate mode, you might become easily annoyed, become kind of annoying, or both. My dad still ribs me by mis-describing logical fallacies, and I’m only now mature enough to (mostly) let it slide.¹¹

There’s also a reasonable concern that debate teaches you to view arguments as “soldiers”—that arguments are pieces to be strategically advanced (and attacked), regardless of their truth. In debate, for instance, you might be incentivized to stifle even reasonable points made by your opponent, because the round can ultimately have only one winner. (Though there are interesting ideas about collaborative formats that address this.¹²)

Merits of debate aside, debate can be a consuming activity, and some people—like yours truly—find it hard to jump in without then giving it their all. Not every student will want to do this, nor should they. That said, even a casual approach can work: My freshman year, I think I competed in the exact minimum required of our debate class—two tournaments—and then spent many afterschool hours playing Halo 3 instead. Eventually, I still got hooked.

Given how much I came to love debating, there’s a further twist to the story: that I ended up in debate class by accident. When we moved to Seattle over the summer, enrollment had already happened, and the only two electives left were debate and study hall. I wanted study hall, but my parents chose otherwise. I wonder often how different my life might be if not for that decision, and whether I might be working in some entirely different field today.¹³

And so, to any high schoolers reading this because a parent sent it to you: I apologize if I’ve gotten you conscripted into debating. My bad; I hope with time you’ll come to forgive me. But also, maybe you should give it a try?

---

Acknowledgements: Thank you to Dan Alessandro, Maureen McLaughlin, Michael Adler, Michelle Goldberg, and Sam Chase for helpful comments and discussion. The views expressed here are my own and do not imply endorsement by any other party.

If you enjoyed the article, please give it a Like and share it around; it makes a big difference. For any inquiries, you can get in touch with me here.

1. I don’t spend much time thinking about exactly when I expect this to happen, but I agree with Helen Toner’s observation that even skeptics express much shorter timeframes than they used to, making claims like, “Reaching human-level AI will take several years if not a decade.”

2. As an illustration, in one year of progress, OpenAI recently produced a model that is ~400x cheaper at solving certain problems than OpenAI’s previous best results, from a cost of about $4,500 per problem down to about $12. As I noted on Twitter, “Human labor doesn’t generally become 400x cheaper in a single year.” (Another comparable reduction in cost would bring this down to about $0.03.)

For an exploration of AI’s impact on jobs for entry-level workers in particular, see “Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence”.

3. By “heavy,” I should be clear: I don’t necessarily mean negative, just ‘very consequential and important to get right.’ I am overdue on writing about a fuller positive vision of AI that I’m excited by, though I did talk about it on a (forthcoming) podcast episode recently; stay tuned!

For one explanation of the extreme dangers of AI—the argument why much-smarter-than-human AI might result in human extinction—see my book review of If Anyone Builds It, Everyone Dies.

4. The National Speech & Debate Association is the main organizer of competitive debate events in the United States. They have resources here on how to start a program at a school, and here on learning to coach. They also offer office hours for those looking to figure out next steps, including becoming an official NSDA coach.

A typical competitive tournament is a weekend event held at a local high school, with students participating in six or so rounds of debate, though there are also shorter and longer variants. Some competitions are also held at places like universities and might draw participants from outside the local area.

For a simpler introduction to debating for students, you might consider holding some in-class debates as an assignment, perhaps in place of (or as a supplement to) writing an essay. For example topics you might consider, see the NSDA’s “Past Topics” section here. (The Public Forum or Lincoln-Douglas topics are probably best suited to an in-class debate experience.)

If you want a more bite-sized approach—rather than a full debate’s worth of back-and-forths—you could even have an assignment related to just writing a counter-argument to some material, presenting it to the class, and doing a Q&A on it. You could even have students write a counter-argument to this piece, if you are so inclined (asking them about what skills they think are especially important to learn for the AI era).

5. Depending on the form of debate, “just a few minutes of prep time” might be a bit of an overstatement; for instance, you might have anticipated common arguments and prepared your own research brief to use when responding, rather than thinking of arguments anew in those few moments. Or, you might be using a computer for quickly typing our your response, rather than making notes on actual paper. Still, you can’t rely on an AI tool like ChatGPT in-round, which would be cheating. (Internet use of any kind is usually disallowed.)

6. I emphasize the in-round experience of debating, in part because I expect ChatGPT is an excellent out-of-round supplement for debaters: making it quicker to understand the core arguments in the literature, to find sources, to understand concepts that aren’t well-explained, and so on. I wouldn’t want debaters to have to go without this as a resource.

7. Back in the 2018–19 era, IBM ran a handful of debate competitions between its Project Debater system and top humans; my understanding is that IBM’s system persuaded some audience members in its direction, but mostly still lost.

I’m not aware of major efforts since then—despite quite significant jumps in AI’s abilities—though this recent paper presents a system called DeepDebater, “a novel autonomous system capable of participating in and winning a full, unmodified, two-team competitive policy debate.”

8. Technically I could have gotten a letterman jacket with varsity patches on it for debate and nothing else, but also, come on.

9. After-round feedback—including “disclosure” of who won or lost—was still somewhat controversial when I was starting out in debate. Thankfully most judges moved in favor of this feedback, and you got to learn in the moment, rather than relying on handwritten, often-illegible explanations received after the tournament.

10. In both debate and real life, if you can’t make a strong case for a position that its supporters would endorse, maybe there’s something you’ve missed. I think of this in the context of AI, with a concern of people deferring to AI on what is right or correct without being able to justify it themselves.

11. When I showed my dad this essay, he conceded to finally understanding the is-ought fallacy: “Just because I mis-describe logical fallacies does not mean I ought to.”

12. If I were implementing a debate curriculum from scratch, I’d go for the “collaborative tournament” idea Brian Christian wrote about in his book The Most Human Human (which also happens to be one of my favorite books about AI).

Instead of teaching debate as an adversarial activity—a round with a single winner—consider making it cooperative, where debaters are rewarded for finding win-win proposals with their ‘competitor,’ who is now more of a collaborator.

For instance, instead of debating a proposition like “The US should fully disarm its nuclear weapons”—one for and one against—the two debaters would be given different maxims to achieve (individual freedom; safety and security; democracy; etc.), and made to craft a proposition—a piece of legislation they’d each sign onto—that achieved both their maxims.

The round wouldn’t need to have a winner, per se; the judge could give a joint score based on how well the two competitors achieved their maxims together. But over the course of the tournament, you’d see how skilled each competitor was at working together with a wide range of others, and building something stronger than just achieving their own interests.

13. Debate was part of my introduction to AI as an important policy topic, which ultimately led me to working at OpenAI. A wild detail: There was a time when my entire OpenAI reporting line was former competitive debaters, I believe with the exception of one person.