The problem of other minds

What entitles you to believe that anyone besides yourself has an inner life at all?

You are as certain that your friend is in pain as you are of almost anything. Ask what that certainty rests on, though, and something odd surfaces: you have never encountered anyone's pain but your own. What you encountered was a face, a sound, a hand pulled back.

So the belief is not an observation. It is an inference from something else that stands in for the thing you actually care about. And once you notice that, the interesting question is not the sceptic's — could you be wrong? — but a working one: what makes a stand-in good, and how would you ever check this particular one?

Begin with the shape of the inference, because it is a familiar shape. We routinely measure what we cannot reach by measuring something we can. A star's temperature comes from its colour, a patient's blood loss from a falling pressure, an economy's activity from tax receipts. In each case the reading is not the quantity; it is a proxy, trusted because we have some independent grip on how tightly the two are joined.

Now ask what joins wincing to hurting. The classical answer, put crisply by John Stuart Mill in 1865, is analogy: in my own case I find that damage produces a feeling, and the feeling produces the wincing. I observe the same damage and the same wincing in you. The most economical supposition is the same middle term.

Notice the peculiarity, though. Every proxy we ordinarily trust was calibrated by cases where we could see both sides at once — thermometer against boiling water, tax receipts against surveyed output. Here the number of cases in which you have ever observed both the behaviour and the feeling is exactly one. Yourself. That is the standing objection to Mill's argument: it generalises the whole population from a sample of one, which in any other domain we would refuse.

Does the proxy at least behave the way a good one should? Test it the way you would test any indicator, by hunting for the two failure modes.

The first is a false positive — the sign without the state. This is easy to produce. An actor weeps without grief; a spinal reflex withdraws a limb that no longer reports to the brain at all; a plausible enough machine can be built to say "ouch" on cue. The behaviour is evidently detachable from the feeling.

The second is a false negative — the state without the sign — and this is the one that has real consequences. Locked-in syndrome leaves a fully conscious person able to move almost nothing. More unsettling, patients diagnosed as vegetative, whose bedside behaviour showed no awareness whatever, have been asked inside a scanner to imagine playing tennis, and some produced the same distinctive motor-planning activation a healthy volunteer produces — the finding Adrian Owen's group reported in 2006 and extended in a larger multi-patient study in 2010. A minority of such patients, not most. What matters here is the structure: the proxy read empty and the mind was there, and the error was caught only by switching to a different proxy, not by looking harder at the first.

That gives the honest position. The behavioural evidence is not proof and never becomes proof; it is an indicator with known failure modes in both directions, which we improve by adding independent channels — behaviour, physiology, neural imaging, evolutionary and anatomical kinship, and, uniquely among proxies, testimony from the system itself. None of them opens the door. All of them together make the hypothesis of a mind the overwhelmingly better explanation of the whole pattern.

And there is a deeper move worth flagging, since it rejects the framing rather than answering it. Wittgenstein argued in the Philosophical Investigations (1953) that pain-behaviour is not mere evidence for pain in the way smoke is evidence for fire, but part of what gives the word "pain" its meaning at all — that a concept whose application were entirely private could not be a concept we successfully share. Whether that dissolves the problem or changes the subject is still argued about.

Two clarifications keep this from collapsing into either dogmatism or theatre. First, the problem is not a reason to doubt in practice. An inference can be non-demonstrative and still be the only rational one available; we treat the whole of empirical science that way. What the problem shows is the kind of belief this is — inferential, defeasible, and grounded in explanation rather than in acquaintance.

Second, there are really two problems wearing one name, and mixing them causes confusion. The epistemic problem asks how you know others have minds. The conceptual problem asks how you could even form the thought that another's experience is like yours, given that you have only your own to point at. Mill's argument addresses the first; Wittgenstein's beetle-in-a-box remark is aimed at the second.

Where this matters practically rather than philosophically: the false-negative case. Because the proxy silently fails in exactly the direction that hurts — minds we cannot see the signs of — the reasonable policy in medicine, in animal welfare, and increasingly in arguments about machines, is to weight an absent signal far less heavily than a present one.

• The private language argument, and whether it dissolves the problem or replaces it.
• How the same proxy reasoning is applied to animals, and where the analogy weakens as the anatomy diverges.
• What would count as evidence of experience in a system that was designed to produce the signs.

Every term the collection defines is gathered in the glossary.