Skip to content
Rabin Poudyal
Rabin Poudyal
← All models

Wisdom and philosophy

Steelmanning

Before you argue against a view, state it in its strongest form, well enough that the people who hold it would agree with your version.

The game theorist Anatol Rapoport, whose rules for criticism Daniel Dennett set out in Intuition Pumps and Other Tools for Thinking (2013). The idea goes back at least to John Stuart Mill's On Liberty (1859): "He who knows only his own side of the case, knows little of that." The word "steelman" itself is a later, informal play on "straw man".

A straw man is a weak version of an opposing view, built to be knocked down. A steelman is the strongest version of the view you disagree with. Rapoport's rules, as Dennett gives them, fix the order: restate the other position so clearly and fairly that your opponent says, "Thanks, I wish I'd thought of putting it that way"; list the points where you agree; mention anything you learned from them; and only then offer criticism.

It works because you can only test your own view against the best opposition. Beating a weak version tells you nothing. Mill argued that you have to know the other side "in their most plausible and persuasive form", or you have no real grounds for preferring your own. It also changes the conversation: people who feel understood are far more willing to hear what you think is wrong.

Steelmanning a view does not mean adopting it, and it takes real work, so save it for disagreements that matter. The common misuse is to build a better argument than anyone actually made, defeat it, and claim to have answered the people you disagree with. Another is to run through the ritual as a polite preface to a rebuttal you had already written. The check is to show your restatement to someone who holds the view.

Example

In On the Origin of Species (1859), Charles Darwin gave a whole chapter, "Difficulties on Theory", to the strongest objections to his own idea. On the eye he wrote that the claim it "could have been formed by natural selection, seems, I freely confess, absurd in the highest possible degree", before answering it with the range of simple to complex eyes found in living animals. He also said what would refute him: "If it could be demonstrated that any complex organ existed, which could not possibly have been formed by numerous, successive, slight modifications, my theory would absolutely break down."

Where it applies

In a design document, write the best case for the option you rejected, in terms its supporters would accept. In code review, restate what the author was trying to do before suggesting a change. When you ask a large language model (LLM) to argue against your design, check that the objections it produces are ones a real expert would raise and not a convenient straw man.

Think about it

Can you state the view you most disagree with so well that someone who holds it would thank you for it?

Applied it to something real? Log it.

Test yourselfUnder Rapoport's rules, what must you do before criticising someone's position, and why does that order matter?

Restate their view so well they would endorse it, note where you agree and what you learned, then criticise. It makes sure you are answering the real argument, and it makes the other person willing to listen.

Related models

Sources and further reading