Modelling a Mind That Is Wrong

Sally puts a marble in a basket and leaves the room. While she is gone, Anne moves the marble to a box. Sally comes back.

Where will Sally look for her marble?

The answer is the basket, because that is where Sally thinks it is. To produce that answer you have to hold two things at once: the world as it is, and Sally's model of the world, which you know to be wrong. Most children pass this reliably at around four. Most three-year-olds say the box: the true location, which is the answer reality supports and Sally's mind does not.

This is the false belief task, and it is diagnostic precisely because the wrong answer is the one the world endorses. You cannot pass by attending harder to the marble.

What it tests is theory of mind: the capacity to attribute mental states (beliefs, desires, intentions) to others and to use them to predict behaviour. The name comes from a 1978 paper by Premack and Woodruff asking whether chimpanzees have one, and the word "theory" is doing real work. You cannot observe a belief. Attributing one is an inference from behaviour to an unobservable cause, structurally like positing an electron.

Then in 2005 Onishi and Baillargeon reported that fifteen-month-olds show the pattern implicitly. Using looking time rather than a question, infants looked longer when an agent searched in the location inconsistent with that agent's false belief, as though they expected the agent to act on what the agent had seen.

The gap between this and the verbal task passed at four generated two decades of argument, and it is not settled. One camp reads the infant data as early mentalising that the verbal task masks behind language and executive demands. Another argues the infants are tracking a simpler behavioural regularity, where an agent last engaged with an object, that mimics belief attribution without being it. Several large-scale replication attempts have had mixed results, which is itself informative.

The claim that has had the most public reach, and has worn the least well, is about autism.

Baron-Cohen, Leslie and Frith's 1985 paper proposed that autism involves a specific theory-of-mind deficit, on the finding that autistic children in their sample failed false belief tasks at higher rates. It was enormously influential and the strong version has not held. Many autistic people pass false belief tasks; performance covaries heavily with language ability; and autistic adults frequently describe understanding others well but arriving at it differently and more effortfully.

The reframing that has gained ground is Damian Milton's double empathy problem. On this account the difficulty is a two-way mismatch rather than a one-way deficit: autistic people are better at reading other autistic people than non-autistic people are at reading them, and vice versa. Studies passing information along chains of participants find that autistic-only chains transmit information about as well as non-autistic-only chains, and mixed chains do worse than either. That is a finding about interaction, and it locates the problem between people rather than inside one of them.

Which brings up the thing that gets lost when theory of mind is taught as a milestone.

It is not a capacity that switches on at four and then works. The same computation remains effortful and unreliable for life, and adults fail it constantly in one particular direction: we cannot model a mind that lacks something we know.

Elizabeth Newton's doctoral study is the cleanest demonstration. Participants tapped out the rhythm of a well-known song on a table while a listener tried to identify it. Tappers predicted that about half their listeners would get it. The actual rate was around one in forty.

The tapper hears the tune. They cannot not hear it; it is playing in their head while their finger moves. The listener hears an irregular series of knocks. And there is no amount of careful consideration that lets the tapper hear what the listener hears, because the knowledge cannot be set aside.

This is the curse of knowledge, and it is the root cause of most bad explanation in the world. The expert writing instructions does not omit steps out of carelessness; they omit the steps that have become invisible through practice, and the invisibility is exactly what prevents them from being noticed. Rereading your own draft will not find them, because the mind doing the rereading is the one that already knows.

The only reliable fix is external. Watch one actual beginner attempt the task, in silence, without helping. Ten minutes of that reveals more than a week of careful revision, because it replaces a simulation of another mind with the thing itself.