the hidden question behind every confident read of another person

It Had Been Hot All Day

Hot enough that the first walk I could reasonably give the dog was eleven at night.

Rocco is a husky. In July he does very little between ten in the morning and ten at night, and by then neither do I, so we go out when it cools.

Canterbury at that hour has a handful of pubs still going, and it was graduation week, so the ones still going were full. One of them is round the corner from my house, The Lady Luck, which is one of my favourite pubs, but unfortunately it is not dog friendly after about seven. So I stood outside on the pavement, where I ran into Ed, who I had not seen in a while, and who went in and got me a pint.

That is the thing about a husky. You do not get to choose who you talk to. He is a social sponge, he collects people, and by the time Ed came back out I was already deep in conversation with strangers who wanted to know what he was.

I had not had a good day. Bumping into an old friend, a pavement full of new people and a dog doing all the work was a considerable improvement on it. I spoke to twenty or thirty people that night, most of them for ninety seconds about huskies.

Then I got talking to Abby.

The conversation moved fast. Five minutes in we were three subjects deep and she was keeping up with jumps that most people ask me to explain.

Then somebody very drunk went past and knocked most of my pint onto the ground.

She had about half of hers left. She held it out and told me to have it, to make up for the one I had lost. She had no idea who the bloke was.

I was grateful and I took it. And somewhere in the same second I decided she was offloading a drink she no longer wanted, that she had had enough and was sensibly regulating herself.

I said so later, describing the evening to someone else. Not as a guess. As a small piece of what I had seen.

The next morning it came apart in my hands. Everything about her pointed the other way. She was measured where the rest of the group was gone. She was tracking a difficult conversation without dropping a thread. She was holding a limit in a place where nobody else was holding anything at all. On the available evidence she was the most sober person on that pavement.

She was also a student. Students have no money, whatever their parents have, and half a pint at pub prices is not nothing. It was a straightforwardly generous thing to do for a stranger, and I had quietly turned it into a logistics decision.

So where did that come from?

Where It Came From

I did not observe anything. I supplied a reason.

Giving away half a drink at eleven at night is not something I would do, so the version of the act that made immediate sense to me was the one where the drink was surplus. I ran her behaviour through myself, found the reading that fitted, and reported the output as a fact about her. The inference never touched her at all.

This is social projection, and it is not a lapse. It is the default machinery of social inference, and plenty of people have been over this ground before me. Ross and colleagues named the false consensus effect. Epley showed it runs on anchoring and adjustment: we start from our own position, adjust outward towards the other person, and stop adjusting far too early. Balcetis and Dunning showed how far motivated reasoning reaches down into what we perceive in the first place.

The uncomfortable part is that projection is usually a good bet. If you know nothing else about someone, you are a sample of the population, so your own value on a trait is a respectable estimate of theirs. It is not irrational. It is a serviceable prior.

It fails in one specific place. It fails where the other person is least like you. Which is exactly where you most need to be right, and exactly where the answer feels most obvious, because the simulation runs clean and produces no sense of having guessed.

Why Abby Stood Out

Abby is an extremely attractive and charismatic woman.

It was graduation week and the pavement was full of people in their early twenties in their glad rags, so attractive on its own explains nothing. What I could not walk past was the wit. She was quick, and she was quick in a way I recognised.

There is a second thing.

It was her graduation day, and most of that half hour was me giving her advice.

I have mentored hundreds of people over my career. Some of it as a volunteer, some of it paid, and a good deal of it like this: someone interesting at a turning point, and an instinct that does not sit on its hands. I have had a hand in building a lot of careers, and it is the thing I am most straightforwardly proud of. I gave a shorter version of the same conversation to two or three other people that night.

What to think about next. Which of the choices in front of her mattered, and which only looked like they did. All of it rested on a premise: that she was built like me, and that what had worked for me would work for her.

So by the time the pint went over, I was not a bystander forming an impression. I had half an hour invested in her being a particular kind of person.

Projection is not flat or uniform. It scales with how much you want the other person to be like you, and the arrow runs in the awkward direction. We do not simply like people because they resemble us, we perceive resemblance because we like them.

Which gives you the thing worth taking away. The reads you most want to be true are the ones your mind is least likely to check.

The Devil Knows the Devil

My friend Polly said that to me once, and she meant it kindly. Polly is another remarkable individual, wired much the same way I am, which is why the phrase was hers and not mine. Her point was that you can look someone in the eye and see yourself reflected back, and she is right that the recognition is real.

That complicates the story rather than confirming it.

When the other person genuinely is built like you, projection stops being a rough prior and becomes a good one. Same mechanism, better hit rate. The autism literature supports it: the double empathy problem predicts that information moves between autistic people as efficiently as it moves between non-autistic people, and that mixed pairs do worse than either. The equivalent work on ADHD is thin, so treat this next part as my extension rather than settled ground.

I have watched it happen. A few weeks ago I ended up, entirely by accident, in a group of ADHD people in Brighton. No translation layer, full speed from the first minute, nobody asking anybody to explain the jumps. It is a specific and recognisable feeling, and it is why I clocked Abby so fast. She raised it herself, unprompted, and said she was fairly confident. She has no diagnosis, I am not qualified to hand her one, and I am not doing so here. There is a separate piece coming on why ADHD looks to me less like a disorder and more like an advantage the modern world is actively selecting for.

Here is the trap, and I think it is the most useful thing in this piece.

The mirror premium is real, and it is paid back as overconfidence. Being right about same-wired people, often, at speed, builds a habit of not checking. The reflex never develops, because for years it was not needed. Then you meet someone who reads as same-wired on the surface and diverges underneath, and you are wrong with total conviction and no early-warning system at all.

Which is precisely what happened. She read as built like me. That is exactly why I projected hard, and exactly why it never occurred to me to ask.

The Expensive Version

Reads like mine are cheap. Drunk or sober is checkable, and the correction cost me nothing but a little pride. The expensive version is motive.

You cannot observe a motive. There is no behavioural signature for jealousy, or guilt, or spite, or an ulterior agenda. So when you conclude that someone did something out of jealousy, you have done what I did with the pint. You have found the version of yourself for whom that behaviour would be jealousy, and experienced the result as something you noticed about them.

Then it hardens, from a passing state into a standing feature of who they are. That is the correspondence bias. And none of it feels like inference, because naive realism means we experience our reads of other people as perception rather than as construction.

I have made this move about people who mattered a great deal more to me than a spilt drink, in the middle of arguments where the motive I assigned them determined everything I did next. Every one of those reads felt like seeing.

One Instrument

The obvious fix does not work. Perspective-taking is the standard advice and has been for thirty years: put yourself in their shoes, imagine how it looks from where they are standing, and you will see them more clearly. Tal Eyal and her colleagues tested it twenty-five different ways, which is enough goes to rule out a fluke, and it failed every time. In some of the experiments it made people worse. The reason is obvious once you see it, because imagining another mind is still your mind doing the imagining, so what you end up with is more confidence in a picture you built yourself.

The answer, as with a lot of these things, is the one every engineer or producer should already know: ‘keep it simple stupid!’ just ask, that’s all.

It would have taken four words and saved me hours of navel-gazing today (although then you would not be getting this fascinating article out of it!), and where you cannot ask, the next best thing is to force a second answer into the space (any second answer, however unlikely) because two explanations cannot be mistaken for perception and one always can.

There is a tell, too, and it runs against instinct. A read that arrives instantly and feels beyond doubt is a read that met no resistance, which means it ran on you and stopped early. Speed and certainty are the warning, not the reassurance, and the reads that felt like hard work are the ones where you actually reached for somebody else.

So this is not really a bias to fix. You have one instrument for reading other people, you cannot put it down, and it was calibrated against itself. Concentrating harder does not help. You just learn where it reads wrong, and take a second measurement when the answer matters.

This next bit is mine, not the field’s. If reading someone really is the brain guessing from thin evidence, then projection is not a fault, it is the guess it starts with, and a first guess counts for more when the evidence is thin, when you are tired, and when you want a particular answer. Which would mean the moments we feel most certain about other people are the moments we are mostly reading ourselves. Still working that one out, but it fits what happened on that pavement uncomfortably well.

I still do not know how much she had drunk, and it was never the interesting question. What I know is that when I described a stranger, I was not describing a stranger at all. I was describing the shortest route from her behaviour to a version of myself I recognised, and calling it observation.