1.8 — Being Your Own Devil's Advocate

For nearly four hundred years the Catholic Church employed a man whose entire job was to argue against sainthood.

When a candidate for canonisation was proposed, one official — formally the Promotor Fidei, the Promoter of the Faith, universally known as the advocatus diaboli or devil's advocate — was appointed to attack the case. He hunted for gaps in the miracles, character flaws in the candidate, and holes in the testimony. He was not a cynic and he was not opposed to saints. He was a safeguard, and the office existed because the Church had worked out something important: a case that only its supporters examine will pass, whether or not it is true.

The post was effectively abolished in 1983. In the decades afterwards, canonisations increased by an order of magnitude. Make of that what you will; the mechanism is not subtle.

This page turns that office into something you can run inside your own head, before a decision rather than after it.

Why you need an external procedure

You already know, from 1.4, that you search for confirmation rather than refutation. Here is the part that makes it worse: being told about the bias does not fix it. Studies of debiasing find that simply warning people, or asking them to "be objective", produces almost nothing. What does work is a procedure with a specific instruction that forces the search in the other direction.

The best-supported single technique is called consider the opposite, and it is embarrassingly simple. Instead of asking "is this right?", you ask: "suppose this turned out to be wrong — what would the reason be?" Then you have to actually produce reasons, in writing. Studies going back to the 1980s find this reduces overconfidence and improves judgement in a way that generic instructions to be careful do not.

The reason it works is worth understanding, because it tells you how to build variations of your own. Your mind answers whatever question it is given, and it answers it by searching memory for supporting material. Give it "is this right?" and it retrieves reasons the thing is right, finds several, and reports back high confidence. Give it "why did this fail?" and the same search machinery runs in the opposite direction and returns a different set of memories. You are not being more honest. You are querying a different part of the database.

The four techniques, in order of cost

One: the pre-mortem, for any plan

Before you commit, do this. It takes ten minutes and it is the highest-return item on this page.

Imagine it is a year from now and the plan has failed completely. Not might fail — has failed. It is over, everyone knows, and you are writing the account of why.

Now write the history of the failure, in detail, as though it already happened.

The trick is entirely in the tense. Asked "what could go wrong?", people produce a short polite list and stop. Told "it failed — explain why", the question is no longer whether to be pessimistic, so the objection is no longer disloyal, and the search returns things the first question would never have surfaced. Research on this prospective hindsight framing finds it substantially increases both the number and the specificity of the reasons people generate.

In a group it does something extra. The junior person who could see the problem and would never have raised it under "any concerns?" will write it down under "it failed, why?", because the framing has removed the social cost of saying it. Have everyone write privately first, then read them out, or the first senior voice will anchor the room.

Two: steelmanning, for any disagreement

The opposite of a straw man. You restate the other position in its strongest, most reasonable form — strong enough that a thoughtful person who holds it would read your version and say yes, that is what I mean, you have put it better than I do.

Three rules make it real rather than performative.

No adverbs of contempt. If your version contains "they naively believe", you are still arguing.

Include the best evidence for it, not the weakest example you can find.

Name the value underneath, because most durable disagreements are not about facts. Someone who wants a decision made quickly and someone who wants it made carefully are not disagreeing about information; they are weighting error types differently. Say which value is doing the work and the argument usually becomes tractable — you can then negotiate over the trade-off, which is a real conversation, instead of over the facts, which is not.

There is a specific test for whether you have done it. Could you argue their side, in front of them, and have them nod? This is sometimes called passing the ideological Turing test, and almost nobody passes it the first time on a topic they care about.

Three: the outside view, for any estimate

You are estimating how long something will take, or how likely a plan is to succeed. Your instinct is to imagine the steps and add them up. This is the inside view, and it is reliably, dramatically optimistic — the well-documented planning fallacy that makes essentially every construction project and software release late.

The outside view asks a different question: forget the specifics. What happened to the last twenty things like this?

Kitchen renovations overrun by roughly half. Software features take about twice the estimate. Most startups fail. Your project feels different to you because you can see its details and cannot see the details of the other twenty, which had their own reasons for being special.

The practical rule: start from the base rate, then adjust for genuinely unusual features of your case — and adjust less than you want to. If you find yourself explaining why the base rate does not apply to you, notice that everyone in the base rate said that too.

Four: the pause, for anything you are certain about

Certainty is a feeling, produced by your brain, and it correlates only loosely with being right. People report identical confidence for memories that are accurate and memories that are demonstrably invented (2.4).

So certainty is worth treating as a signal to slow down rather than to proceed. The specific check, which takes twenty seconds: what is the strongest single piece of evidence against this, and when did I last look at it? If the answer is that you have never looked, your confidence has no source outside your own head.

Doing this to yourself, which is the hard version

Arguing against a position in a meeting is easy. Arguing against something you have publicly committed to, or built, or based two years on, is a different order of difficulty, because now the cost of being wrong includes looking wrong.

Four moves that make it survivable.

Separate the belief from your identity, in advance and in language. "I think X" is much easier to abandon than "I am an X person". People who describe their positions as current best guesses change their minds more easily than people who describe them as who they are — not because they care less, but because there is less at stake in the change.

Say the strongest counter-argument out loud first, before anyone else does. This is counter-intuitive and it works remarkably well in practice. It costs you nothing in credibility — it gains you a great deal, because it demonstrates you have looked — and it removes the defensive crouch you would otherwise be in when someone else raises it.

Keep a record of what you got wrong. A short list, revisited occasionally. Not for self-punishment; for calibration. Most people cannot tell you their error rate on any kind of judgement, which is why almost everyone believes they are an above-average judge of character (10.1). A written record is the only way to find out what your actual hit rate is.

Set the exit condition before you start. "I will abandon this if X has not happened by June." Written down, dated, ideally shown to someone. This is the single best defence against sunk cost — the pull to continue with something because of what you have already spent, which is the one factor that logically should not affect the decision at all and is the one that most often does. The condition set in advance by a version of you who had not yet invested anything is worth more than any judgement you can make once you are in.

When not to do this

Being your own devil's advocate is a tool with a proper time, and using it constantly is its own dysfunction.

Not during execution. Deciding and doing are different phases. A surgeon does not reconsider the operation mid-incision, and a person who reopens the decision every day never finishes anything. Argue hard before, then commit, with a pre-agreed review point.

Not on questions of value. You cannot argue yourself out of what matters to you, and trying produces paralysis rather than insight. The technique is for claims about the world.

Not in the middle of an emotional conversation. Steelmanning your partner's position while they are still upset reads as lawyering, because in that moment the content is not what they are asking you to respond to. Part 12.7 covers the ordering: acknowledge first, reason later.

And not as a personality. The person who reflexively takes the contrary side of everything is not doing this. They are running one automatic move, which is precisely what the technique is meant to defeat, and after a while nobody tells them anything.

What to do with this

Adopt one habit this week: the ten-minute pre-mortem on your next real decision. Job, purchase, difficult conversation, project. Set a timer. Write the sentence "it is a year later and this went badly" at the top of a page, and fill the page. Then look at the list and ask which of these you can cheaply prevent, and which one would be fatal. Most decisions have exactly one fatal failure mode, and finding it is worth the ten minutes several hundred times over.

Add a second habit for arguments: write their side first. Before replying to anything that has annoyed you — an email, a comment, a relative — write two sentences putting their case as well as you can. You will send a different message. Sometimes you will not send one at all, and that is the technique working, not failing.

And once a quarter, run one on a belief rather than a plan. Pick something you are sure about. Ask what would change your mind, then go and look at the best available version of the other side for twenty minutes. Most of the time you will come back with the same conclusion, held more accurately and defended better. Occasionally you will not, and those occasions are how a person's thinking actually improves over a lifetime.

Next: 1.9 closes Part 1 with the question underneath all of it — what intelligence actually is, what an IQ score does and does not tell you, and which of the things people do to get smarter actually work.