Steelmanning

Thinking ToolsAny stepAlso known as Steel Man

Named in the rationalist blogging community and defined in writing by Luke Muehlhauser (Better Disagreement, 2011), after John Stuart Mill's case for knowing the strongest opposing view (On Liberty, 1859); Daniel C. Dennett set out Anatol Rapoport's related rules (Intuition Pumps and Other Tools for Thinking, 2013).

Builds the strongest version of an opposing argument, stronger than its own backers may have put it, and answers that version before answering anything else.

The case against a price rise, made stronger before it is answered

The sales head's objection starts weak. At each of the pricing lead's turns, pick the move that makes the case against stronger, and watch the meter; then test the plan against the strongest version, and see what that does to the plan. Illustrative.

Pricing lead

The case against

Pricing lead

The case against

Select a turn to read it. At each pricing lead turn, pick a move.As first put. A software company plans to raise every price 12 percent. The head of sales objects in one line, and the room is ready to wave it off.Better assumption. Replace the vague claim with the assumption that would make it true.Pick a move to see the reply.The case now names who is at risk and why.Stronger: a specific group with a real alternative.The room moves on.A rebuttal too early: the case was never made strong.Sharper. The objection now points at the monthly book, $4 million of the company's $10 million.Best evidence. Find the strongest support the case could have.Pick a move to see the reply.The finance file confirms it.Stronger: the company's own record, on the same customers.Someone points out it was one email.Weak: an anecdote is easy to dismiss, which is why it does not test the plan.Strongest version. If monthly churn doubles again, the extra churn could cost more in the first year than the 12 percent adds.Say it back. Restate the case to the head of sales.Pick a move to see the reply.He says it is his worry, put better than he put it.Right: the case is now his, and stronger than he made it.He bristles, and the meeting turns into a defense.A dig: the case is weaker for being framed as a mood.The test. Only now does the rebuttal start.Pick a move to see the reply.True of annual customers, and no answer to the record.This beats the first version only.At last year's response, the rise loses about $0.4 million in year one.Right: the plan fails against the strongest case.Change the view. The test went against the plan.Pick a move to see the reply.The revised plan adds $0.7 million and keeps the monthly book.Right: when the rebuttal fails, the view changes.Monthly churn doubles by the second quarter.The exercise was for show.

Right calls: 0 of 5Right calls: 1 of 5Right calls: 2 of 5Right calls: 3 of 5Right calls: 4 of 5Right calls: 5 of 5

Strength of the case against1of 7

  1. As first put (The case against). A software company plans to raise every price 12 percent. The head of sales objects in one line, and the room is ready to wave it off.
  2. Better assumption (Pricing lead). Replace the vague claim with the assumption that would make it true.
    • The 40% on monthly plans, who can switch to a cheaper rival in a day The case now names who is at risk and why. Stronger: a specific group with a real alternative.
    • Churn has always been low, so the worry is overdone The room moves on. A rebuttal too early: the case was never made strong.
  3. Sharper (The case against). The objection now points at the monthly book, $4 million of the company's $10 million.
  4. Best evidence (Pricing lead). Find the strongest support the case could have.
    • Last year's 5% rise: monthly churn doubled for two quarters The finance file confirms it. Stronger: the company's own record, on the same customers.
    • An angry email from one customer after the last rise Someone points out it was one email. Weak: an anecdote is easy to dismiss, which is why it does not test the plan.
  5. Strongest version (The case against). If monthly churn doubles again, the extra churn could cost more in the first year than the 12 percent adds.
  6. Say it back (Pricing lead). Restate the case to the head of sales.
    • Read the strongest version aloud and ask whether it is fair He says it is his worry, put better than he put it. Right: the case is now his, and stronger than he made it.
    • Restate it with a note that sales is always cautious He bristles, and the meeting turns into a defense. A dig: the case is weaker for being framed as a mood.
  7. The test (Pricing lead). Only now does the rebuttal start.
    • Customers always complain at first, then stay True of annual customers, and no answer to the record. This beats the first version only.
    • Run the plan with monthly churn doubled At last year's response, the rise loses about $0.4 million in year one. Right: the plan fails against the strongest case.
  8. Change the view (Pricing lead). The test went against the plan.
    • Raise annual plans 12%, hold monthly, add a cheaper monthly tier The revised plan adds $0.7 million and keeps the monthly book. Right: when the rebuttal fails, the view changes.
    • Go ahead anyway: the decision is already announced Monthly churn doubles by the second quarter. The exercise was for show.

An example

A software company's leadership team was ready to approve a 12 percent rise on every plan. The head of sales objected in one line, customers will hate it and leave, and the room was ready to move on, since churn had always been low.

The pricing lead stopped the vote and built the objection up instead. The customers who could really leave were the 40 percent on monthly plans, $4 million of $10 million in revenue, who could move to a cheaper rival in a day. The company's own records gave the best evidence: after last year's 5 percent rise, monthly churn had doubled for two quarters. She read the stronger case aloud and asked the head of sales whether it was fair; he said it was his worry, put better than he had put it. Then she tested the plan against it. If monthly churn doubled again, the rise lost about $0.4 million in its first year. The team changed the plan: annual prices rose 12 percent, monthly prices held, and a cheaper monthly tier was added.

Why it works

An objection usually arrives in its weakest form: quick, emotional, and easy to beat. Beating that version feels like a test of the plan, and it tests nothing. The plan meets the real risk later, without anyone having looked at it.

Steelmanning reverses the habit. Before answering, the listener repairs the objection: better assumptions, the best available evidence, and wording its holder would accept. Only that version is worth rebutting, because if the plan survives it, the plan has survived the strongest test the room could build. If it does not, the team has learned something before it cost money. Restating the case so well that its owner endorses it also changes the meeting: the objector stops defending and starts helping.

The thinking behind it

John Stuart Mill John Stuart Mill, the British philosopher and economist, argued in On Liberty (1859) that you do not know your own view until you know the strongest form of the opposing one. The steel man is a working method for meeting that standard. It asks for the opposing case at its strongest before the decision is made, while changing course is still cheap, and treats the result as a test of the view, not a formality.

Luke Muehlhauser Luke Muehlhauser, writing in the blog post Better Disagreement (2011, LessWrong), defined a steel man as an improvement of someone's position or argument, harder to defeat than the original. He placed restating an opponent's case better than they could, before refuting it, at the top of his scale of disagreement. Wiktionary gives the post as the earliest dated citation for the term.

Daniel C. Dennett Daniel C. Dennett, in Intuition Pumps and Other Tools for Thinking (2013), presented Anatol Rapoport's rules for criticism: restate the other view so clearly that its holder would say thanks, I wish I had put it that way; list where you agree and what you learned; and only then say a word of rebuttal. The rules give the steel man its order of operations.

How to run it

  1. State the opposing view fairly. Write it in the other side's words before changing anything.
  2. Repair its assumptions. Replace vague claims with the specific assumptions that would make the view true, naming who and what is at risk.
  3. Give it the best evidence. Find the strongest data the case could have, including your own records, and drop the anecdotes.
  4. Restate it to its owner. Read it back and ask whether it is fair; keep revising until they agree it is their view, put well.
  5. Say where you agree. Name the parts of the case you accept, so the rebuttal is aimed only at what remains.
  6. Test the plan against it. Run the plan, the numbers, or the argument with the strongest version assumed true.
  7. Change the view if the test fails. Revise the plan openly, and say which part of the case changed it.

Where it goes wrong

The steel man is a straw man in better clothes. The listener adds polish but keeps the version easiest to answer. Ask the view's holder whether the restatement is fair; if they hesitate, it is not strong yet.

The exercise is theater. The team builds the strongest case, rebuts it weakly, and goes ahead as planned. Decide in advance what result of the test would change the plan, and write it down.

Every objection gets the full treatment. Steelmanning takes time, and some objections really are weak after repair. Save it for decisions that are costly to reverse, and for views held by people whose judgment you respect.

When to use something else

  • Red Teaming. When the plan needs a dedicated adversary hunting for weaknesses, not one objection made strong.
  • Murder Board. When a proposal must survive a panel of hostile questions before it goes to the real decision-makers.
  • Premortem. When nobody has objected yet, and the team needs to imagine how the plan could fail.

Where it sits in SOLVE eX

  • Any step: Tools that serve every step: habits of thought and checks on the thinking itself.

Go to the source

  • On Liberty, chapter 2, 1859
  • Better Disagreement, LessWrong, 2011
  • Intuition Pumps and Other Tools for Thinking, W. W. Norton, 2013

Back to the catalog