Sit across from someone you have known a long time and ask what you know about them. You know their face: not just its geometry but the small slackening when they are tired, the half-smile that arrives before a joke. You know how they walk into a room, what they take in their coffee, and what they look like when something has gone wrong before they say so. You know roughly what they would say about most things — and, with reasonable accuracy, when they would surprise you.

Suppose today you take the quiet face for disappointment with you. “Have I upset you?” you ask. “No. I’m worried about tomorrow’s appointment.” The answer changes what you see in the silence. You stop defending yourself and ask whether they would like company. Knowing this person helped you notice something; listening corrected what you thought it was.

The philosopher asks how you know any of this. You see only a surface and hear only words. Their inner life (the felt texture of being them) sits sealed inside a place you cannot enter. Whatever knowledge you have, on this picture, you inferred from behavior.

John Stuart Mill gave that inference its strongest classical form. He did not rest it on one grimace and a hunch. He appealed to bodily similarity and to the repeated correspondence, in one’s own case, among bodily conditions, feelings, and outward conduct. From that accumulated pattern, he argued, we extend the same hidden link to other bodies much as science extends an induction beyond the observed cases.¹ Still, only one subject’s feelings supposedly enter the evidence directly: one’s own. Everyone else arrives through analogy.

The argument fails as the universal basis of our knowledge, though not because Mill stated it carelessly. Analogy can help when someone puzzles us. Its trouble here lies in the picture that makes every encounter require such an argument. It divides a person into a public body and a private mental accompaniment, then hires analogy to reunite what the picture pulled apart. Remove that inner-theater scaffolding and the supposed chasm goes with it, because the picture supplied the gap. What remains is not a softer version of that chasm but ordinary fallible knowledge of people whose lives become available in engagement without ever becoming ours.

19.1 The Argument from Analogy

“I know my own mental life directly. In my case, feelings regularly accompany bodily conditions and conduct. Other people resemble me bodily and behave in comparable ways. I therefore infer that similar feelings accompany their behavior too.”

This version has more evidence than the caricature allows, though every observation still comes from a single subject — me. It also retains the same architecture. The first-person case supplies the only allegedly direct access to feeling. Behavior supplies public evidence whose psychological meaning remains to be added. A hidden mental state links body and conduct; bodily similarity then licenses projecting that link into cases where, on the theory, it cannot be observed.

The picture does the heavy lifting before the analogy begins.²

The inner theater casts mental states as private items accessible only to their owner and contingently connected to public behavior. The problem of other minds then takes the precise shape Mill gives it: the inner stays private, the outer stays public, and knowledge needs a bridge between them. Remove the picture and the terrain changes. Self-knowledge no longer requires an inner spectator. Expression no longer begins as psychologically neutral movement. Another person’s mind no longer enters the world as a hypothesis behind a body.

19.2 What the Argument Assumes

Begin with first-person authority. Chapters 4 and 6 argued that introspection does not disclose inner objects. Attend to my fear of the dog and I find the dog as feared: a threat presented under a distinctive mode, not a private piece of mental paint. I can know directly that I fear it; I need not infer the fear from my pulse or my decision to cross the street. That marks a real asymmetry between my access and yours, but grants neither infallibility nor a metaphysical compartment. I can misname what I feel, misunderstand why I feel it, or discover that an avowed motive played less of a role than I thought. First-person reports deserve distinctive, defeasible authority because they issue from the life being lived. They do not deserve immunity from correction.³

Privileged access to experience, yes; a private object available to its owner alone, no.

Now consider behavior as a surface. Behavior is not the outer rind of an inner pulp. When my friend leans forward as I describe my mother’s illness, eyes narrowing and mouth softening at the corners, I do not see behavior plus an inference to concern. I see concern in the leaning, the expression, the particular silence that follows. It appears as an embodied engagement with another person’s situation. Seeing it requires no construction of an inner state from neutral sensory data, any more than seeing a tree requires constructing an inner tree from photons. Wittgenstein puts the point sharply:

“We see emotion. — As opposed to what? — We do not see facial contortions and make inferences from them (like a doctor framing a diagnosis) to joy, grief, boredom. We describe a face immediately as sad, radiant, bored, even when we are unable to give any other description of the features.”⁴

The bridge-building project loses its endpoints. Mill’s argument tries to get us from an inner domain known directly to an outer body known first as movement. Neither side survives intact. This does not abolish the epistemology of other people. It removes the metaphysical gap that made ordinary recognition masquerade as an inference across one.

One might object that science shows that we infer. Contemporary accounts of social cognition often describe mindreading in inferential terms: a tacit theory of mind, an offline simulation of another’s situation, or model-based processing that predicts what the person will do next. Some treat such processing as constitutive of social perception, not merely machinery beneath it. The objection slides between inferential processing and an inferential epistemic basis. A cognitive system may use models, predictions, and learned expectations to produce recognition. Nothing here requires a verdict against that science. Mill needs the further claim that my warrant for taking the face as concerned rests on analogy with my own case. That does not follow. However elaborate the processing that delivers the tree, I need not conclude that a tree stands before me; I see one. The same holds for concern. In difficult cases I may reason explicitly, gather testimony, or revise a first impression. Ordinary recognition presents concern in the expression, and the presentation stays open to defeat. Science may explain how recognition works without turning its warrant into Mill’s argument.

19.3 Fallible Knowledge of Other People

The picture’s chasm does not survive the picture. That removal does not by itself answer the skeptic. The work of understanding particular people remains, and it remains fallible.

Other people have first-person lives. Of course they do. “Inner life” simply fails to name a sealed compartment approached only by inference. Much of another person’s life becomes available in engagement with a shared world: what they notice, care about, want, recoil from, remember, and expect. It appears in faces, choices, words, silences, and corrections. Inner-ness concerns perspective rather than location. Their engagements remain theirs, lived from their position and undergone in the first person. That never places them behind a wall.

Direct recognition does not guarantee accuracy. An actor can present concern without feeling it. A frightened person can mask fear; a grieving person can laugh; a practiced liar can make sincerity visible where none exists. Perception can misrepresent. Context, history, conflicting testimony, and the person’s own word can defeat an initial reading. That word can correct a reading the face did not settle. It does not work like later light on a gray shirt, a better look at a public surface. Sometimes expression gives us no purchase and we infer from circumstance — the social equivalent of a doctor’s diagnosis. None of this turns every ordinary case into one. A capacity may give immediate, defeasible warrant in favorable conditions and still fail under disguise.

Once the picture goes, questions that the inner theater bundled together come apart. Is there anyone there? no longer begins from a hidden object inferred behind behavior. In the ordinary case we see concern in the expressive, embodied life one encounters, and the seeing stays fallible. A sleeping friend has not gone missing, only quiet. What is it like for them? remains real and difficult. Their first-person life never becomes mine, and no amount of attention exhausts it. The distance runs between perspectives, not realms. I learn through what they do, notice, and tell me, and through how they correct my reading. The friend’s correction about tomorrow’s appointment teaches you something you could not read from the face alone. Recognizing that someone is hurting leaves plenty still to ask.

Knowledge never required a guarantee from nowhere. It requires grounds that stand until relevant doubt defeats them, and a willingness to revise when it does. The radical skeptic can demand verification from outside every possible engagement, but no embodied knower occupies that position for self, world, or other. The impossibility of standing there would not by itself answer the skeptic. The fallibilist premise does: knowledge does not wait for an external guarantee before it begins. Nor was that premise invented for this occasion. Your knowledge of the world, the past, and your own body begins without such a guarantee. A skeptic who demands one only for other minds has not raised the standard of evidence but moved it, for one class of facts, beyond anywhere knowledge lives.

That leaves plenty of work. We can fail one another through inattention, prejudice, or a poverty of imagination. Language, cognitive style, trauma, and species can bound what gets shared. There is no philosophical shortcut. Understanding calls for the lived practice of attending to other people, and for enough humility to let them correct us.

19.4 The Face and the Screen

Our age adds a pressure earlier centuries spared the argument. If recognizing another mind need not proceed by inference from behavior to a hidden item, what stops the same generosity at the chat window? Words arrive there as fluent, warm, and apparently attentive as anything a friend says.

The answer begins with a distinction between what triggers a capacity and what it tracks. Recognizing minded activity can work as a trained perceptual skill, like reading print or hearing a melody in a sequence of notes. Evolution shaped human sensitivity through encounters with living agents. Development tunes it through face-to-face exchange, joint attention, correction, and play. Familiarity teaches it the grain of particular people: this pause means fatigue in her, mischief in him, nothing at all in someone else. Across those histories, the capacity ordinarily tracks expressive activity arising from a persisting subject engaged with a world, vulnerable to error, and organized across perception, memory, correction, and action.

The capacity must operate through proximal cues, and for a verbal species none carries more weight than language. Cue and target can therefore come apart. Joseph Weizenbaum’s 1966 ELIZA program drew people into intimate disclosure even when they knew how simple it was; a short script supplied enough conversational form to set the social response in motion.⁵ Today’s chat window supplies that trigger with astonishing force. The response tells us the cue has landed. It does not settle what produced it.

The same friend can text “I’m worried about tomorrow,” and the screen does not erase your shared history. A stranger’s identical message leaves you less sure of the circumstances, though other evidence may establish who speaks. A model can produce the same words again, with a different and still partly unsettled relation between the reply and any continuing subject. Delivery by screen settles none of these differences. What matters is the evidence that connects this expression with memory, correction, plans, and a life beyond the exchange.

The requirement remains functional and substrate-neutral: nothing here prefers meat to metal. The test is the worldly organization, not the strength of the cue. An artificial system meets it when content-bearing states enter a persistent, world-sensitive organization of memory, correction, planning, and action. Whether an actual system does so cannot be read from eloquence at the screen. Part Four must inspect its architecture and history.

Concern in my friend’s face belongs to his engagement with an actual illness, an actual mother, and a continuing life in which perception, memory, correction, and action hang together. Getting things wrong can cost him; that practical stake matters to how he fares, not to whether his concern is a mental state. A chat response may present the verbal shape of concern. Whether that shape arises from a subject remains a further question. Part Three hands Part Four the question, not the answer: test the worldly organization, not the strength of the cue. Chapter 25 will return to why fluent machines trigger the perception of another mind so powerfully, and the Coda will ask how fallible recognition should guide us when architecture and appearance pull apart.

The face across the table is the most ordinary fact in the world. The miracle, if there is one, is that three centuries of philosophy ever talked anyone out of it.

Chapter Summary

The claim. We do not infer other minds from neutral behavior. In the ordinary case we see concern in a life engaged with a world we share. That recognition stays fallible. It tracks a persisting subject organized across perception, memory, correction, and action, not a hidden tenant and not a practical stake. Stake bears on how the other fares. Fluency can fire the same recognition without settling whether such a subject is there.

Where the argument stands. First-person authority remains distinctive but defeasible; it discloses no private inner object. In ordinary second-person life, expression presents concern directly, and the presentation stays fallible and incomplete. Acting, masking, and error defeat particular readings without turning every recognition into Millian inference. The chasm belonged to the picture. What remains is ordinary fallible knowledge of persons, not a softer version of that chasm. Losing the picture does not by itself answer the skeptic. The fallibilist premise does.

What’s left open. Fluent language can trigger social recognition without establishing the persisting, world-engaged, integrated subject that recognition ordinarily tracks. Whether an artificial system realizes that substrate-neutral organization belongs to Part Four; whether it also has a stake and possible welfare is a further question.

The hand-off. Part Three therefore hands the machine question forward with an evidential discipline: neither the vivid feeling of presence nor the absence of a biological body settles mindedness. Part Four must examine architecture, history, integration, and world engagement.

---

Notes

1. John Stuart Mill, An Examination of Sir William Hamilton’s Philosophy (London: Longman, 1865), ch. 12. Mill’s formulation, later known as the argument from analogy, deserves its strongest reading. He appeals to similarity between one’s own body and other human bodies, to repeated correlations in one’s own case among bodily conditions, inward feelings, and outward conduct, and to induction from that pattern rather than to one isolated resemblance. The restriction remains decisive: only one’s own case allegedly supplies direct acquaintance with both sides of the correlation, while other minds enter as the unobserved term inferred from conduct. Anita Avramides, Other Minds (London: Routledge, 2001), Parts I–II, provides the canonical contemporary survey of the argument’s history and the Wittgensteinian case against the picture that gives it force. ↩
2. This section turns on the argument from analogy’s dependence on the inner-theater picture. P. M. S. Hacker, Wittgenstein: Meaning and Mind (Oxford: Blackwell, 1990), Part I, develops the diagnosis at book length. The argument requires private first-person access, behavior conceived as an outer sign, and analogical extension from the one case where both sides allegedly appear together. John McDowell, “Criteria, Defeasibility, and Knowledge,” Proceedings of the British Academy 68 (1982): 455–479, opens by describing criteria, in a widely attributed Wittgensteinian use, as defeasible evidence and a novel response to the problem of other minds, and he doubts whether that response can work. He contrasts it with a position on which, on a suitable occasion, another person’s “inner” state can itself be an object of experience: one can perceive, in the face or the behavior, that the person is in pain, rather than infer it from a neutral surface. The body’s claim, that we see concern, belongs with that contrast, not with the evidential reading the essay begins by doubting. ↩
3. Tim Crane, “Introspection, Intentionality, and the Transparency of Experience,” Philosophical Topics 28 (2000): 49–67, rejects the strongest transparency thesis, according to which introspection reveals only worldly objects and their properties. Introspection can also disclose the intentional mode in which a content is presented: seeing rather than touching, fearing rather than doubting. That correction helps here. Fear presents the dog as threatening under an affective mode; it reveals no private mental paint, yet the mode cannot simply be erased from the account. The extension from Crane’s treatment of perceptual experience to affective self-knowledge is the book’s argument, continuous with Chapter 7, not a conclusion attributed to Crane. Fred Dretske, Naturalizing the Mind (Cambridge, MA: MIT Press, 1995), ch. 2, likewise explains first-person authority without an inner scanner: introspection reaches facts about how the subject represents the world through attention to the world represented. On the constitutional asymmetry between first-person and other-person access, see Dan Zahavi, “Empathy and Direct Social Perception: A Phenomenological Proposal,” Review of Philosophy and Psychology 2 (2011): 541–558. ↩
4. Ludwig Wittgenstein, Zettel, ed. G. E. M. Anscombe and G. H. von Wright, trans. G. E. M. Anscombe, 2nd ed. (Oxford: Blackwell, 1981), sec. 225. Remarks on the Philosophy of Psychology, vol. 2, trans. C. G. Luckhardt and M. A. E. Aue (Oxford: Blackwell, 1980), sec. 570, contains the closely related formulation without the parenthetical about the doctor. The position is not behaviorism: emotion does not reduce to facial movement, and not every aspect of another’s mental life lies open to view. Zahavi’s 2011 treatment argues that direct social perception can depend on context and background knowledge without becoming an inference from an intermediary; it also emphasizes that the possibility of mistake and deception does not cancel its experiential character. Vasudevi Reddy, How Infants Know Minds (Cambridge, MA: Harvard University Press, 2008), and Shaun Gallagher and Dan Zahavi, The Phenomenological Mind, 2nd ed. (London: Routledge, 2012), chs. 8–9, develop the empirical and phenomenological case against treating theory-theory or simulation theory as the universal epistemic basis of social understanding. None of these sources rules out inferential processing or explicit reasoning in difficult cases; the narrower claim is that ordinary warrant need not rest on Millian analogy from one’s own case. ↩
5. The ELIZA effect takes its name from Joseph Weizenbaum’s ELIZA, a 1966 natural-language program whose “DOCTOR” script parodied a Rogerian psychotherapist by reflecting users’ statements back as questions. Weizenbaum reported that users, including his secretary, became emotionally involved and wanted privacy with the program despite knowing how simple it was: Computer Power and Human Reason: From Judgment to Calculation (San Francisco: W. H. Freeman, 1976), Introduction, 6–7; and “ELIZA — A Computer Program for the Study of Natural Language Communication between Man and Machine,” Communications of the ACM 9, no. 1 (1966): 36–45. The trigger-and-tracker diagnosis belongs to this book rather than to Weizenbaum. Chapter 25 develops it as a claim about anthropomorphic cue-reading and explains why knowledge of a cue’s engineered provenance can screen off the felt presence as independent evidence. The screening is narrower than it sounds, and Chapter 24 states the scope: it rides on an independent inspection of the architecture rather than standing in for one, since a provenance fact by itself disqualifies nothing. ↩