Four techniques genuinely counter confirmation bias: considering the opposite (list reasons your view could be wrong), steelmanning (state the strongest opposing case), the assigned devil's advocate (someone argues against), and the premortem (assume failure, explain it). They differ in cost and in what they fix — the comparison table below shows which to use when.
Awareness is not one of them. Knowing that confirmation bias exists does almost nothing to stop it: the bias operates on what you notice and what you bother to look up, upstream of the deliberate reasoning where "being aware" lives. What changes outcomes is not insight but procedure — a step you run at a fixed point in your process, whether or not you feel biased that day. This piece compares the four procedures with the best claim to working, including where each quietly fails.
What exactly is confirmation bias — and what makes it hard to fix?
Confirmation bias is the tendency to seek, notice, and remember evidence that supports what you already believe, while under-weighting evidence that would cut against it. The classic demonstrations come from the psychologist Peter Wason, whose rule-discovery experiments in the 1960s showed people testing their hypotheses almost exclusively by looking for confirming cases — even when only a disconfirming case could settle the question. The full guide to cognitive biases puts it alongside the other systematic tilts in judgment.
It resists fixing because it has three separable moving parts, and most "solutions" touch only one:
- Search. You google "is this stock a buy" rather than "this stock problems". The evidence you see is pre-filtered.
- Interpretation. Ambiguous evidence reads as supporting your side. Both parties to a dispute can leave the same data more confident than they arrived.
- Memory. Confirming instances are easier to recall, so your sense of "everything I know" is itself skewed.
A technique that only forces you to hear a counterargument fixes search but leaves interpretation intact — you listen, and dismiss. That distinction separates the four methods below.
How do the four debiasing methods compare?
| Method | What you actually do | Which failure it targets | Cost | Works best when | Main failure mode |
|---|---|---|---|---|---|
| Consider the opposite | Write down concrete reasons your judgement could be wrong, before deciding | Search + interpretation | ~5 minutes, solo | You already hold a view and want it stress-tested fast | Produces token, easily-dismissed reasons if rushed |
| Steelmanning | State the opposing case in its strongest form, well enough that its holder would endorse your version | Interpretation | 15–30 minutes, effortful | The disagreement is with real people whose position you have been caricaturing | You steelman a weak version and mistake it for the strong one |
| Assigned devil's advocate | Nominate someone to argue against the group's emerging view | Group-level search | Cheap in a meeting | A team is converging suspiciously fast | Role-play dissent is discounted; everyone knows it isn't sincere |
| Premortem | Assume the decision failed; each person writes why, independently, before discussing | Search + social suppression | 10–15 minutes, whole group | A concrete plan exists and is about to be committed to | Only surfaces failure modes people already suspected |
Read the table as a division of labour, not a ranking: considering the opposite is the cheap solo default, steelmanning is the expensive one that actually moves beliefs, the devil's advocate is the weakest of the four and the most popular, and the premortem is built for teams and plans.
Why is "considering the opposite" the best default?
Because it is nearly free and attacks the search failure directly. The instruction — before you commit, write down specific reasons your judgement might be wrong — comes from research by Charles Lord, Mark Lepper and Elizabeth Preston in the 1980s on countering biased assimilation of evidence. The notable part: vague instructions to "be objective" did not help; what worked was the specific version that made people generate the opposing reasons themselves.
Two implementation details do most of the work. First, write it down — held in your head, the exercise collapses into a shrug. Second, demand specificity: "the market could turn" is a token; "if our two biggest customers renew at last year's volume instead of growing, the plan is underwater" is a testable claim. The best upgrade is to name in advance what evidence would change your mind, then go looking for exactly that — a falsification condition stated before you search turns an open-ended trawl into a real test.
Its limit is honest: it is solo, and a motivated reasoner can produce three weak counterarguments in thirty seconds and call the box ticked. It reduces the tilt; it does not remove it.
Is steelmanning better than playing devil's advocate?
For changing what you actually believe, yes — clearly.
Steelmanning means reconstructing the opposing position in its strongest, most charitable form — the version its most thoughtful holder would endorse. The term is internet-native folk vocabulary, but the discipline is older and better specified in Rapoport's rules, proposed by the game theorist Anatol Rapoport and later popularised by the philosopher Daniel Dennett: re-express your target's position so clearly they say "thanks, I wish I'd put it that way", list what you agree on and what you learned, and only then argue. That sequence is hard to fake — you cannot restate a position to its holder's satisfaction while still caricaturing it, which is why it reaches the interpretation failure other methods miss.
Playing devil's advocate looks like the same thing and is not. The advocate is assigned, so everyone in the room knows the objections are a performance. Research on group dissent by the psychologist Charlan Nemeth has consistently found that authentic minority dissent stimulates broader, more original thinking in a group, while role-played devil's advocacy is far weaker — and can even harden the majority, who now feel they have "heard the other side" and dealt with it. The ritual delivers the feeling of rigour at a discount.
If you keep the role anyway, two fixes help: assign it before the group's view forms, and require a written case with evidence rather than loose objections. Better still, find someone who genuinely disagrees and protect them — real dissent is the active ingredient, and organisations mostly spend their energy suppressing it.
Steelmanning has its own subtle trap: you build a decent version of the opposing case, defeat it, and conclude the matter is settled — never noticing the strongest argument was one you never met. The test is whether someone who holds the view would sign your summary. If you cannot find such a person, you are probably arguing with a straw man wearing armour.
When should you run a premortem instead?
When the unit of analysis is a plan and a team, not a belief and a person.
The premortem, developed by the research psychologist Gary Klein, works like this: before committing, tell the group to imagine it is a year later and the project has failed badly, then have each person write down independently why. The independence matters — writing before discussion stops the first speaker anchoring everyone else. So does the framing: "what could go wrong?" invites polite hedging, while "it failed, explain how" legitimises doubts that optimism and hierarchy suppress. The decision-making guide treats it as one of several lenses worth running before any hard-to-reverse commitment.
Think of it as confirmation bias's team-scale counterpart: individuals filter evidence, groups do it collectively and then reinforce each other, which is why a room can be more confident than any of its members. It pairs naturally with second-order thinking — many surfaced failures are not "the plan is wrong" but "the plan works, and then what it causes undoes it".
Its limit: it retrieves failure modes people can already imagine. Genuinely novel risks stay invisible, so treat it as a way of unlocking known concerns, not generating unknown ones.
What should you use, and when?
A workable default policy:
- Everyday judgements, solo. Consider the opposite. Two minutes, written, specific.
- A belief you hold strongly, on a contested question. Steelman it. Find a real person who disagrees and restate their view until they endorse the restatement.
- A team converging fast on a plan. Premortem, with independent writing first. Skip the assigned devil's advocate, or repair it with evidence requirements.
- A quantitative forecast. None of the four is the first move — start with the outside view and reason from base rates before applying any of these to your inside-view story.
- A problem where every option feels wrong. Step back from evidence-testing entirely and rebuild the question from first principles; confirmation bias often defends not the answer but the framing that produced it.
None of these makes you unbiased. Each shifts the odds a little, at a cost, on one specific failure. People who reason well are not the ones with better introspection — they are the ones who put a cheap procedure between themselves and their conclusions, and run it on the days they feel certain. Especially then.
Frequently asked questions
Does simply knowing about confirmation bias help? Very little on its own. Awareness operates downstream of the filtering, and people readily see the bias in others while missing it in themselves. Procedures you run at a fixed point in a decision are what change outcomes.
What is the difference between steelmanning and playing devil's advocate? Steelmanning means you personally reconstruct the opposing case in its strongest form, well enough that its holder would endorse your version. Devil's advocacy is an assigned role, so the objections are understood as performance — which is exactly why groups discount them.
How is a premortem different from a risk assessment? A risk assessment asks what might go wrong, which invites hedged, low-commitment answers. A premortem asserts that it did go wrong and asks for the explanation, which makes voicing doubt socially safe and produces far more specific failure modes.
Can confirmation bias ever be useful? Sticking with a hypothesis long enough to test it properly has real value; abandoning views on weak counter-evidence is its own error. It turns costly when stakes are high, evidence is ambiguous, and you never specified what would change your mind.
Which technique should a beginner start with? Considering the opposite — it costs minutes, needs nobody else, and targets the search failure that starts the chain. Add steelmanning once you have a specific disagreement worth resolving.
Each technique here has a full entry on the encyclopedia, with its definition, a worked example, and the failure mode that tells you when to put it down. Read the full breakdown of confirmation bias on Build Mind.
