Philosophy & Critical Thinking: Devil's advocate

What you'll be able to do: Defend a position you actually hold against an opponent trying, in good faith, to take it apart, a different and harder test than steelmanning someone else's view.

Not the same skill as steelmanning

Steelmanning (article 02) builds the other side's best case. Devil's advocate reverses the direction: the AI attacks your side, in real time, while you hold your ground. Steelmanning is a production skill, can you generate the strongest counterargument. Devil's advocate is a resistance skill, can you survive one being generated against you, live, without retreating silently to something weaker.

Worked example: the private language argument

Wittgenstein's target in Philosophical Investigations (§243-315): a language for sensations that refers, at root, to purely private inner states with no possible external check on correct use. His argument is that such a language is incoherent, because there would be no difference between correctly applying a word and merely seeming to yourself to apply it correctly , "whatever is going to seem right to me is right.. and that only means that here we can't talk about 'right'" (§258).

A student defending this position gets pressed: "But I clearly know what my own toothache feels like better than anyone else does, doesn't that privileged first-person access show sensation-language IS private, at least at its root?"

A good defence has to separate two claims the attack runs together. Privileged epistemic access to your own sensations is plausibly true and not what Wittgenstein denies. A private language , naming sensations via a criterion-free "private ostensive definition," with no possible external check on whether you're using the name consistently, is what's under attack. Conceding the first while holding the second is the actual position, and the live exchange is what forces a student to notice they'd drifted into defending a stronger, less defensible claim than the argument requires.

Worked example: biting the bullet on relativism

Paired with article 02's steelmanned relativist, devil's advocate presses from the other side: "If morality is only true relative to a culture, you have no standing to condemn practices in other cultures that you find horrifying, are you actually willing to bite that bullet?", the classic objection from James Rachels (The Elements of Moral Philosophy, ch. 2). A defended relativist has to either bite it explicitly and say so, or retreat to a narrower "relativism about some things, not others" and name exactly which things are exempt and why. Either answer is real progress. A vague deflection is not.

How to run it

1. State the position you actually hold, not one you're merely describing on someone else's behalf.
2. Instruct the AI to attack it as hard as it genuinely can, in good faith, without pulling punches because you're a student.
3. When you feel the urge to say "well, I don't really mean the strong version" , stop, and name explicitly which weaker position you're retreating to.
4. Ask afterwards whether your original statement of the position was accurate, or whether the exercise revealed you'd been holding a stronger claim than you could actually defend.

I hold this position: [position]. Play devil's advocate: attack it as hard as you genuinely can, in good faith, looking for the actual weaknesses, not performative harshness. Don't let me retreat to a weaker version without naming the retreat out loud. Keep pressing the same line of attack if my answer doesn't actually address it.

The tell

Every attack gets the same prepared response. If your answers don't get more specific as the questioning continues, you're replaying a script you brought into the room, not defending a position under live pressure.

Try this today

Take a position you hold with real conviction, not an assigned essay position, one you actually believe. Have it attacked properly, and notice the exact point where you first feel the pull to retreat.