Philosophy & Critical Thinking: How AI can best tutor this subject

What you'll be able to do: Use AI for the parts of philosophy where dialogue is literally the method, argument-building, objection-raising, definitional pressure, while treating everything it tells you about what a philosopher actually held as a claim you still have to verify.

The subject the tool fits, and the trap that comes with the fit

No subject matches the interaction pattern of a chat-based AI better than this one, on paper. The discipline's actual method, state a thesis, take objections, revise, defend under questioning, is what a dialogue interface does natively. History has to fight the tool's tendency to invent facts. Mathematics has to fight its tendency to skip steps. Philosophy has almost no equivalent friction: the format of the tool and the format of the discipline are the same format.

That should make this the easiest subject in the corpus. It doesn't, because the fit hides a specific failure that's harder to catch here than almost anywhere else.

AI produces fluent, coherent, well-organised readings of philosophers that are subtly wrong. Not wrong the way a bad date is wrong, wrong the way a plausible interpretation is wrong, which is to say: it doesn't look wrong. A fabricated quotation in a history essay can be checked against an archive. A fabricated reading of Kant reads exactly like a correct reading of Kant, because both are fluent prose making confident claims about what a difficult text means, and the text doesn't refute a bad paraphrase of itself the way a primary source refutes an invented fact.

Take a concrete case. Ask most language models to summarise Nietzsche on the death of God, and you'll typically get something like: with God dead, no values have any foundation, so ultimately nothing really matters. That sentence is fluent, quotable, and sounds exactly like the intro-course one-liner students already half-believe. It is also a serious misreading. Nietzsche treats nihilism as a crisis to be diagnosed and overcome, not a conclusion he's recommending, the whole architecture of Thus Spoke Zarathustra, the revaluation of values, the will to power as a value-creating force, is built to answer nihilism, not endorse it. Whole books (Bernard Reginster's The Affirmation of Life is the standard one) exist because this exact point is still argued over by specialists. A student who takes the fluent one-liner as settled hasn't made an error a fact-check would catch. They've absorbed a live scholarly dispute as if it were closed.

So the working rule here is different in kind from history's, though it rhymes:

> Use it to build and test arguments. Treat every claim about what a named > philosopher held as contested until you've checked it against the text or a > named secondary source who'd defend that reading.

What it is genuinely good at here

Generating the strongest objection to anything, instantly, without social cost. A study partner gets bored, or embarrassed to keep pushing a friend. The tool doesn't, and philosophical maturity depends on having your position attacked by someone who won't ease off.

Reconstructing an argument into numbered premises. Mechanical, tedious by hand, and it's the single move that turns "I think this essay is wrong somewhere" into "premise 3 is false."

Definitional pressure, without fatigue. Asking what you mean by "free," or "person," or "good," is unglamorous and most humans stop after one round. The tool doesn't get tired of asking.

Sustained adversarial questioning for thesis defence, the closest thing a chat window has to a real viva, on demand, at 11pm before the essay is due.

Cross-tradition breadth. It has "read" analytic, continental, ancient and non-Western material broadly, and can put a Humean and a Buddhist account of causation in the same room, which most single courses never do.

Generating counterexamples to a generalisation on demand , useful because a definition that's never been attacked hasn't been tested, it's just been stated.

What it is bad at here, specifically

Exegesis stated as settled fact. As above. This is the central danger of the subject.

Distinguishing a philosopher's considered view from a stage they later repudiated. It will blend early and late Wittgenstein, or treat Hume's more sceptical moments and his more naturalistic ones as one continuous position, unless pressed to separate them.

Concepts that resist translation. Dasein, ren and li, dharma, sunyata, it will flatten these to a rough English gloss, which loses exactly what the original term was coined to preserve.

Knowing whether a question is a live specialist disagreement or one with a strong majority answer. It can manufacture false balance on settled questions, or false consensus on contested ones, and gives no signal for which it's doing.

Being too rhetorically fluent for its own logical rigour. It's optimised to sound persuasive, and a bad argument in confident, well-structured prose reads exactly like a good one until you've learned to check the joints.

The shape of a good session

1. Reconstruct the argument , yours or the text's, into numbered premises and a conclusion before doing anything else.
2. Force the definitions of the terms carrying the weight, and check whether the disagreement survives once both sides mean the same thing.
3. Steelman the position you're inclined to reject, checked against the principle of charity.
4. Surface your own unstated premises.
5. Defend the resulting thesis under real, sustained pushback, and hunt counterexamples against any generalisation you've made.
6. Verify every claim about what a named philosopher actually held against the primary text or a named secondary source before it enters your essay.

The tool does real work at every one of steps 1-5. Step 6 is where you stop trusting it and start checking.

The instruction to set

For this session you are tutoring me in philosophy. Rules: when you reconstruct an argument, number every premise separately and label the conclusion. Never present a reading of a philosopher as settled; if a reading is contested, say so and name a scholar or school who reads it differently.
Before we argue about a claim, force me to define the key terms, and if our disagreement disappears once we're using the same definition, tell me that instead of letting me think I won. Attack my thesis for real. Don't soften your objections because I'm a student. And when I ask what a philosopher "really meant," tell me plainly when that's still disputed rather than picking a side and asserting it as fact.

The failure that looks like success

A fluent, well-organised paragraph on Nietzsche and nihilism, or on Kant's categorical imperative, or on the private language argument, that reads better than anything you'd have written unaided, every sentence confident, the structure textbook-clean, and that a specialist would flag as a misreading in the first line.

It doesn't announce itself as wrong. There's no fabricated date to catch, no citation to chase to an empty shelf. There's just a reading, stated as if it were the only reading, of a text that supports at least two.

The tell: you can't say, for any claim you've made about what a philosopher held, where in the text that's stated or which two named scholars disagree about it. If the honest answer is "the AI told me," that claim isn't yours yet.

What it cannot replace

The actual encounter with the primary text, its resistance, the fact that the Critique of Pure Reason is hard to read because the thinking is genuinely hard, not because no one has paraphrased it well enough yet. A good paraphrase can orient you before you go in. It cannot substitute for going in.

It also can't replace holding a position under real, embodied, socially costly pressure, in a seminar room, in front of people whose opinion of you is actually at stake, which is part of what philosophical conviction is built from. And it can't replace arriving at a position through your own failed attempts. A dialogue that's too good at producing a defensible take too quickly can rob you of finding out, first-hand, exactly why the obvious take doesn't survive.

Where to go next

01 Argument reconstruction is the mechanical skill everything else in this subject depends on. 02 Steelmanning is the discipline's central obligation, not a courtesy. 03 Definition forcing is the direct fix for arguing past someone for an hour over a word.