Steelmanning: Arguing in Good Faith
Ethics Tribunal
Three looks at the difference between defeating an argument and defeating a cardboard cutout of it. Watch who actually wins.
Rapoport’s Rules: Daniel Dennett’s Recipe for Criticism — First, Restate Your Target So Well They Say “Thanks”
Philosopher Daniel Dennett popularised Anatol Rapoport’s rules for criticism: (1) re-express your opponent’s position so clearly they say “Thanks, I wish I’d put it that way”; (2) list every point of agreement; (3) name what you’ve learned from them — and only THEN rebut. The rebuttal lands because the audience knows you understood the real thing.
"Earn the attack by first building the target you’re allowed to hit."
Source ↗Pew: Most Americans Believe Their Political Opponents Argue in Bad Faith
Pew has repeatedly found large majorities of Americans see the other side as not just wrong but dishonest — acting in bad faith. When you assume your opponent is stupid or lying, you stop building their real argument and start swinging at a strawman. The disagreement curdles, and nobody learns anything — a habit AI chatbots will happily feed you if you let them.
"Assume bad faith and you’ll only ever defeat the scarecrow you built."
Source ↗Ask an AI to "Argue Against X" and It Often Hands You a Strawman
Prompt a chatbot to argue against a position and it frequently produces the *weakest* version — the easy-to-knock-down cliché, because that pattern is everywhere in its training data. If you accept it, you walk into a real debate having rehearsed against a scarecrow. Ask instead for the STRONGEST opposing case, with its best evidence, and you get sparring worth having.
"The AI will build you a scarecrow unless you demand a champion."
In each case, did the arguer beat the real opposing view — or a weakened copy they built themselves? What did that buy them, and what did it cost?
Reading
Build It Before You Break It.
A strawman is the weakest version of your opponent’s view — easy to knock down, and worthless to knock down, because nobody actually held it. A steelman is the strongest version — the one its smartest defender would recognise and endorse. The discipline is simple to state and hard to do: you have not earned the right to attack a position until you can argue it better than the person who holds it.
Why bother, when your goal is to win? Three reasons.
It makes you right less often — which is good. Building the strongest opposing case is the fastest way to discover your own is wrong. Better to find out in private than in the tribunal.
It makes your win count. An audience that watches you defeat the real argument believes you. An audience that watches you defeat a scarecrow knows you dodged.
It’s the only honest move. Arguing in frame means engaging what your opponent actually claims, not a convenient cartoon of it.
Rapoport’s rules give you the steps. Before you rebut: (1) re-state the opposing view so well its holder says “yes, that’s it”; (2) list the agreements — what’s genuinely right in it; (3) name what you learned from it. Then, and only then, attack — the strongest version, with its best evidence already on the table.
Watch for the AI trap: ask a chatbot to “argue against X” and it will often hand you the strawman, because clichéd weak versions are what it has seen most. Demand the champion, not the scarecrow.
- 1
What exactly makes a steelman different from just “being fair” to the other side?
Reveal answer
A steelman is the STRONGEST version — stated so well its holder would endorse it, with its best evidence supplied by you — not merely a polite or accurate restatement.
- 2
Why does steelmanning make your eventual rebuttal MORE convincing, not less?
Reveal answer
The audience sees you defeated the real argument, not a scarecrow — so your win is credible. Beating a strawman signals you dodged.
- 3
You ask an AI to “argue against” your essay and it gives a weak objection. What’s happening, and what do you ask instead?
Reveal answer
It defaulted to the common, clichéd (strawman) version from its training data. Ask for the STRONGEST opposing case with its best evidence and a defender’s reasoning.
Build it before you break it
The Steelman
A strawman is the weakest version of a view — worthless to beat. A steelman is the strongest — the one its smartest defender endorses. You haven’t earned the attack until you can build the target.
Defensible placements (the *argument* is graded, not the verdict): - **A — systemic failure (justice lens).** No single actor intended bias; the injustice is structural — a feedback loop trained on biased arrests, then laundered as objectivity. Strongest counter: utility (does it beat human judges on average?). - **B — moral abdication (care lens).** A duty of care to a foreseeable vulnerable user was designed away for engagement. Counter: rights (the company's speech, the parents' role) — must be met, not ignored. - **C — tool-misuse, contested (virtue vs common good).** Automation is not new; the question is whether craft has standing. Weak arguments appeal to "harm" as a feeling; strong ones name what is owed and by whom. A student who places C at "abdication" can still **win** if their justice-and-virtue argument outframes the other side.
Order: restate (so they’d endorse it) → agree → supply their best evidence → name what you learned → THEN rebut. The rebuttal comes last and aims at the strong version. Any order that attacks before building is a strawman — the whole point is that the attack is earned.
Marshal Your Case
A tribunal verdict that beats the steelman is the one that wins your portfolio’s case — and the one a judge believes.
Steelman the side you’re going to argue against.
Take the tribunal motion (or any view you disagree with). Build its STRONGEST form — the opposition research you’d do before attacking (200–300 words):
- Re-state the opposing view so well its holder would say "yes, that’s exactly it." No scarecrows, no sneering.
- Best evidence — the two strongest facts or arguments on THEIR side (that you supply).
- Genuine agreement — one thing in their view you actually concede is right.
- Only now — your rebuttal, aimed at this strong version, in one paragraph.
You pass when a person who HOLDS the opposing view would read your steelman and nod — and when your rebuttal attacks that strong version, not a weaker one you slipped in.
- ✓ [
- ✓ ]