The Chinese Room showed that arranging symbols by their shapes never adds up to meaning. This chapter presses the deeper point behind it: a physical system is not even running one program rather than another except relative to someone’s interpretation — so “the brain is a computer” was never the kind of fact a mind could be built on.

What this chapter does. The last chapter answered the computationalist who thinks the right symbol-shuffling would understand. But there is a retreat available, and a careful reader will already have reached for it: forget the symbols, the mind is not the marks but the program they implement — the abstract computational organization, which the marks merely happen to spell out. This chapter follows the retreat and finds it has no floor. It presses one question the computationalist would rather not sit with: is running a program even a fact about a physical system, or a description we read onto it? Four moves answer it. First, the triviality result — pressed from inside functionalism by Putnam, from outside it by Searle — that lets a wall, on the cheap, run any program you name. Second, the diagnosis of why the triviality keeps returning no matter how the technical screws get tightened: formal structure carries no built-in tether to the world, so it submits to every interpretation and commits to none. Third, the best repair on the market — Chalmers’s argument that genuine implementation is a real constraint, not a free lunch — granted on its own terms and shown not to touch the load-bearing point. Fourth, a confusion this chapter has to clear out of the road: the Church–Turing thesis, stated honestly, lends computationalism none of the support it is perpetually drafted to supply. (The title names the thesis in its naïve, pre-Chalmers form; what survives his repair, in the third move, is the sharper claim — that even a properly implemented program is a shape, and a shape is not a meaning.) This is the second of Part Four’s three independent defenses — syntax ≠ semantics (Chapter 20), observer-relativity (here), simulation ≠ realization (Chapter 22). The computationalist must beat all three. The book needs only one to stand.

---

§I. The Move the Last Chapter Left Open

Chapter 20 closed on a diagnosis: when we feel the pull of crediting a language model with understanding, we are responding to the fluency of its output, not to any evidence that its words are grounded in a world. Syntax, however fluent, does not suffice for semantics. And a computationalist can grant all of that and still think the argument has missed him, because — he will say — the mind was never supposed to be the symbols. The symbols are just the visible spelling. The mind is the computation they implement: the abstract pattern of states and transitions, the program, the thing that stays the same whether you run it in neurons or silicon or, as Chapter 19 allowed, beer cans and string. Attack the marks all you like; the program is untouched.

This chapter follows him into that retreat — which Chapter 19 in fact pre-announced. Setting functionalism out at full strength, it handed the rest of this Part its marching orders: each chapter ahead is a fresh assault on the one thesis functionalism could not secure, that the right organization suffices for a mind. One of those assaults, it promised, would ask “whether computation is even a determinate fact about a physical system or something an observer reads onto it.” That is this chapter, and the answer is the chapter’s title.

Here is the question in its rudest form, the one philosophers play as a parlor game. There is always someone — there is always someone — who announces, with the calm of a person about to enjoy themselves, that the radiator against the far wall is, at this very moment, running Microsoft Word. Not metaphorically. Not “could be wired to.” Running it, now. The radiator’s molecules occupy some fantastically complicated physical state, and that state grinds, moment to moment, into other states. Write down the sequence of computational states Word passes through as it runs, then pair each one, in order, with one of the radiator’s physical states. Given enough physical states to draw on, and enough freedom in how you pair them, the pairing exists. So the radiator runs Word. It also, by the identical trick, runs a chess engine, and the flight software of a 747, and the chess engine’s exact opposite, and it dreams of Provence on alternating Tuesdays.

The healthy first instinct is to wave this off as sophistry. The instinct is right about the conclusion and wrong about the difficulty. The trouble is not that the trick hides a fallacy behind a curtain; the trouble is that it does exactly what it says, by a method nobody has managed to outlaw without quietly importing the very thing the computationalist was trying to do without. The interesting question is never whether the radiator really runs Word. It is what the radiator’s running Word, on these terms, tells us about what running a program is.

§II. The Wall That Allegedly Runs WordStar

The worry is only worth answering at full strength — a version too weak to trouble a serious functionalist would be a straw man.

Chapter 19 made functionalism’s case generously, so I will not remake it here: a mental state is a functional role, a role can be filled by anything that does the causal job, and so the same mind can run on different stuff. That was the view’s great liberating promise — multiple realizability, the mind set free of its substrate. The triviality argument takes that promise and runs it off a cliff. For if a mind is the right functional organization, then having the mind is instantiating the organization, and organization is cheap. To say a physical system “implements” a given finite-state automaton is only to say there exists a mapping from the system’s physical states to the automaton’s states under which the physical transitions line up with the automaton’s transition table. Take any system with enough internal states changing fast enough — a wall warming unevenly, a bucket of water, the radiator — and, with no rule against gerrymandering the mapping, such a mapping can always be found. Putnam proved a sharp form of it: every ordinary open physical system implements every finite-state automaton.¹

And the wound is the whole engine of the chapter. He wanted minds to be organization precisely so they could be multiply realized. The triviality result says the realization relation is so unconstrained that implementing-this-automaton is no achievement at all — it is a permission slip you write yourself. So being the right automaton is something the radiator already does, along with having my mind, and yours, and the mind of a reader persuaded by this argument and the mind of one who is not, all at once, settled only by which mapping a bored interpreter lays over it. A theory on which the radiator already understands everything has not explained understanding. It has mislaid the word.

The point arrives from the other direction too, and the convergence is the strongest single thing this chapter has to show. Searle reached it not through automata but through physics. Whether a lump of matter counts as a “symbol,” a “computational state,” a case of “running a program” is not, he argued, an intrinsic feature of it the way mass and charge are intrinsic. It is a feature assigned to it by an observer who reads the computational description on. Syntax, in his flat phrase, is not intrinsic to physics.² The same wall is describable as implementing WordStar, or its negation, or nothing, depending entirely on the interpretation projected onto its molecular states — and a physical system has the causal properties it has, full stop, whether or not anybody reads them as a computation.

So two thinkers who agreed about almost nothing in the philosophy of mind walked toward the same wall from opposite sides. Putnam came at it from inside functionalism — he had, after all, built the house; Chapter 19 introduced him as the founder who would become its most distinguished defector. Searle came at it from outside, as the program’s standing critic. When a view’s inventor and its fiercest opponent reach one verdict by independent roads, the verdict stops looking like anyone’s hobbyhorse and starts looking like a result. Putnam, to his lasting credit, did not flinch from where his own argument led. Having spent the 1960s arguing that the mind is computational organization, he spent Representation and Reality arguing that the very multiple realizability which first recommended the view dissolves it once you see how unconstrained the realization relation is — and he said so under his own name, against his own most famous idea, which is the kind of courage that ought to embarrass anyone who has ever held a position because of who else held it.³

§III. Why the Threat Keeps Coming Back

The natural response, shown the radiator, reaches for new gears. The bad mappings are gerrymandered — they pair physical states with computational ones in wild, disjunctive, after-the-fact ways; surely a real implementation has to be more disciplined than that. Require the mapping to respect the system’s causal structure, so the states genuinely push each other around in the matching pattern. Demand that it support counterfactuals — that the system would have gone here had it been there. Add complexity, or transducers, or some careful algebra over state-spaces. A respectable literature has grown up around exactly this project of saving functionalism from triviality by tightening the screws, and Peter Godfrey-Smith — a philosopher of biology and of mind with a particular gift for the patient inventory, separating the arguments that work from the ones that overreach — has written the article that walks through which tightenings hold and which leak.⁴ Some of them genuinely rule out the silliest mappings. Hold that thought; the next section is about the best of them.

The shape of the situation is the tell. Tighten one screw and a subtler gerrymander slips past; tighten that and the account starts excluding implementations you wanted to keep. The recurrence is not a run of bad luck. It is a symptom, and the diagnosis has a faintly comic structure once you let it through.

The triviality arguments do not expose a fixable bug in functionalism, some loose bolt a cleverer engineer tightens next year. They expose what formal structure is. Anything characterized purely by its form — purely by what plays which role among inputs, internal swaps, and outputs — admits arbitrary interpretation, because form carries no built-in tether to anything in particular. A string of zeroes and ones describes itself with perfect indifference to whether it stands for a chess position, a tax return, a chord, or nothing whatever. The symbols do not point at the world; they sit there, and we point them. So of course the wall runs WordStar — in precisely the trivial sense in which the wall does anything you can describe with enough interpretive slack. Give the world a creative enough reader and it becomes a Rorschach blot, the thing that means whatever is brought to it and nothing on its own. We go looking, in the physics, for the fact that fixes which program the system runs — and that fact is the one thing the physics never contains. We brought it. We always bring it.

§IV. The Best Repair, Granted and Set Aside

The strongest objection here is not that the triviality argument is wrong but that it is too cheaply won — that a careful account of implementation defeats it honestly, so the wall, properly understood, does not run WordStar after all.

Chalmers has made that case as well as anyone, and fairness wants it at full strength rather than hurried past. He grants that unconstrained mappings trivialize, and then argues that genuine implementation was never unconstrained. For a physical system to implement a computation, its causal structure must mirror the formal structure of the computation: its states must divide into components, and the components must transition under counterfactual-supporting causal laws in a pattern isomorphic to the computation’s own. He formalizes this with what he calls combinatorial-state automata, and the upshot is sharp. A rock does not implement every finite-state automaton, because the rock’s physics carries none of the rich, counterfactual-supporting, component-wise structure that genuine implementation demands.⁵ Putnam’s mapping was a fraud not because it broke a rule we invented to stop it, but because it never tracked the causal joints at all — it laid a story over the system instead of finding the story in it. Implementation, on this account, is a substantive constraint. Most systems implement rather few computations. The wall is innocent of WordStar.

I think Chalmers wins this, and I want to say so plainly, because the reply that matters depends on conceding it. Implementation is not fully trivial; the gerrymander can be fenced out by demanding causal, counterfactual-supporting structure; the radiator, disciplined, does not run Word. Grant all of it.

It does not rescue the computationalist, and seeing why is the center of the chapter. Chalmers has shown that which computations a system implements is constrained by its causal structure — a real fact about the system, not a free interpretive gift. What he has not shown, because nothing could, is that fixing the computation fixes a meaning. Look again at what the constraint delivers: a system whose component states push each other around in a pattern isomorphic to some automaton’s transition table. That is a fact about the shape of the causal flow. It says nothing about what any of those states is about. The very same disciplined, counterfactual-rich, non-trivially-implemented structure can be read as computing a chess move, or the truth-value of a tax claim, or nothing of interest — because the structure is isomorphic to all of them at once; that is what it is for them to share a form. Chalmers tightened the relation between a physical system and a computation. He did not, and could not, tighten the relation between a computation and the world it is supposedly about. The first relation is causal-structural and can be made objective. The second is semantic, and it is exactly the one that formal structure, however rigorously realized, does not contain. So the constraint earns the computationalist a genuine fact — this system runs these programs, not those — and leaves him precisely where he began on the only question ever at issue: not which program the brain runs, but how anything it does comes to mean. Observer-relativity retreats one level under Chalmers’s pressure and survives there intact: running a program, even a properly implemented one, is being a certain shape — and a shape is what all the interpretations have in common, not the thing that picks one of them out.

§V. A Theorem That Was Never About Minds

There is a second prop under the computationalist’s confidence, independent of the triviality dialectic. It is a mathematical result so secure that any argument leaning on it inherits a glamour it has not earned — and the borrowing has a name. Gualtiero Piccinini, whom the last chapter introduced as the philosopher who has done most to say precisely what physical computation is, called it the Church–Turing fallacy.⁶

The public-facing argument runs like this. The Church–Turing thesis tells us that anything computable can be computed by a Turing machine; the brain is a physical system; therefore the brain is a computer, and a good enough program will eventually be a mind. It cites a famous theorem. It seems to make consciousness tractable to engineering. And it conflates two completely different claims.

The Church–Turing thesis, stated honestly, is a proposition about functions. Church and Turing, in 1936, gave precise definitions of what it is for a function to be effectively computable — workable out, in principle, by following a finite list of mechanical rules — and showed their definitions equivalent. The thesis says that this formal notion captures the intuitive one: any function a clerk with paper and pencil could in principle grind out by rule is computable by a Turing machine. That is the whole of it. It speaks of functions and formalisms. It does not speak of brains, or minds, or physical systems at all.⁷

Watch the inference try to cross that gap. The brain computes things — but if that means it manipulates discrete symbolic states by rule, neuroscience has found no such thing; that is the commitment in dispute. And if it means only that some computable function can describe the brain’s behavior, it is harmless and trivial, true in equal measure of a hurricane, a stomach, and the solar system. Therefore the brain is a Turing machine — but the thesis says nothing about which physical things are Turing machines; it fixes what functions such machines compute, not which lumps of matter qualify as one. Therefore running the program would be a mind — the most ambitious leap of all, from the brain can be modeled by a computation to running the model would constitute the thing modeled. The thesis is silent on every step that matters. It is mathematics; the questions are metaphysics, neuroscience, and conceptual analysis, and a 1936 result about computable functions answers none of them.

This is why the present defense does not even need the wall. Grant Chalmers his disciplined implementation; grant that the brain is computable down to the last ion. The computationalist still owes an argument that being a mind is running a program rather than merely being describable by one — and his most-cited witness, the Church–Turing thesis, has nothing to say on the matter. To get from there to computationalism he needs three further things the thesis does not give him: a substantive account of when a physical system genuinely computes, one that tells the brain apart from the rock; an empirical case that the brain, on that account, computes; and a philosophical argument that mental life is constituted by such computation rather than merely described by it.⁸ Each is a separate undertaking. None follows from the theorem. Stripping the Church–Turing thesis of the role it was press-ganged into does not refute computationalism; it returns the burden of proof to where it always sat — on the empirical and conceptual case for treating cognition as computation, a case with serious defenders and serious critics, to be fought on its own ground. The mathematics survives the scrutiny intact. The metaphysics it was made to carry has to walk on its own feet.

§VI. What the Wall Lacks

So the wall does not run WordStar in any sense worth wanting; the disciplined repair leaves the verdict standing; and the great theorem the computationalist leans on was never about him. What, then, does the program-running? What has the brain got that the wall has not, such that there is a fact of the matter about what the brain’s states mean and none about the wall’s?

Not the formal structure alone — that has been the whole argument. The structure has to come tethered: it has to be the structure of something that already, for some non-magical reason, picks out one interpretation from the infinite others on offer. And here the embodiment this book has been building stops being a slogan and turns load-bearing. A nervous system that evolved to track predators tracks predators — rather than sandwich crumbs, rather than nothing — because evolution selected its ancestors for getting that mapping right, generation upon generation, under penalty of being eaten. The pairing between its states and the world was not laid on by an interpreter on a Tuesday afternoon; it was earned across a long causal history in which the bearers of the right mapping left more children than the bearers of the wrong one. Part Three gave that its proper name: selection — evolutionary, developmental, learning-historical — does the pinning-down that pure form cannot, by making one interpretation the one under which the system worked.⁹ The radiator dreaming of Provence on Tuesdays cannot dream of Provence at all, because no historical fact about the radiator makes Provence, rather than Pittsburgh or pickle juice, the content of its dream. The brain has such facts. The wall has none. Content depends on grounding, grounding depends on history, and history is not a feature of formal organization. It is a feature of bodies in worlds.

This is also where Searle’s deepest version of the point comes home — though it needs care, because it can be misheard as a claim the book rejects. The brain does carve nature at its joints. Its causally efficacious processes are specific physical and chemical events, with specific causal powers, and a neuron’s response to a molecular signal has that specificity built into the chemistry, not projected onto it by an interpreter. Notice that this is a different sense of “intrinsic” from the one running through the rest of the chapter: when Searle says syntax is not intrinsic to physics, “intrinsic” means observer-independent; here it means built into the causal powers of the stuff. Keep the two apart, because the brain’s causal specificity, real as it is, does no semantic work on its own. The radiator’s molecular processes have their specificity built in too; specificity as such does not single out the brain, and it is not yet content. What makes one of the brain’s many determinate causal profiles about predators rather than crumbs is the selection history of a moment ago, not the chemistry as such — the chemistry only supplies the determinate joints there to be selected over. What computation in the formal, substrate-neutral sense gets wrong, then, is not the chemistry but the bookkeeping: it collects systems by the shape of their state-transitions and throws away exactly the causal-environmental engagement that grounds meaning. So the wall and the brain are not two implementations of one kind, one defective. They differ in kind. The wall has an imposed description and no earned content; the brain has determinate causal joints and a selection history that ties them to the world — and only the second confers meaning.

Which is why the chapter’s verdict is, at bottom, an anti-reification verdict, of a piece with everything this book has refused to do with mind, meaning, and information. “Running a program” wears the grammar of an intrinsic activity, the way “rotating” names something the radiator’s fan really does. It is not. It names a relation between a system and an interpretation — and where the interpretation comes free, the relation comes cheap. Treat computation as an intrinsic property and you have reified a description into a deed.

That hands the next chapter its opening. Grant, now, against everything above, a system that non-trivially implements the right computation, tethered and all — a perfect computational duplicate of a brain’s causal organization. Chapter 22 asks whether running that, however faithfully, amounts to being the thing it models, and answers that a simulation of a process, however exact, is not an instance of it: the modeled hurricane never wets the desk it runs on. Observer-relativity and simulation-versus-realization are different knives — one says the program is not intrinsically there, the other says that even where it is, running it is not doing the deed — and the computationalist has to get past both, with the syntax-and-semantics point of Chapter 20 behind them.

The moral reaches the present without straining for it. When a language model writes a fluent paragraph about Provence, it does not thereby know about Provence. It performs a transformation of formal structure — a transformation, to be fair, of breathtaking sophistication and real practical worth — and the structure it transforms admits the same interpretive slack the wall does. What tethers its words to a world comes from elsewhere: from the human authors whose corpus it learned, embodied creatures with selection histories who meant something when they wrote, and whose meanings the model rides without owning. The wall that allegedly runs WordStar stands to WordStar exactly as the model stands to the meanings in its training set — in a relation conferred entirely from outside, by interpreters who already mean things. Subtract the interpreters and the structure floats free, available to every reading and answerable to none. That, in the end, is what was always wrong with the picture of the mind as software running on the brain. Software runs on nothing by itself. It runs on something that, for non-software reasons, already meant — and the brain is such a thing, not because the mapping onto it is clever, but because of the long, biological, world-involving fact that some of its patterns were selected for being about things, and the things they were selected for being about are still out there, past the window, where they always were.

Chapter Summary

This chapter pressed Part Four’s second defense: “running a program” names a relation between a system and an interpreter, not an intrinsic fact about the physics — so even a flawless computational organization could not, by itself, be a mind. Putnam’s triviality result lets a wall run any program you name, and Chalmers’s best repair tightens which computation runs while leaving wholly untouched what it means. What the brain has and the wall lacks is a selection history that grounds one interpretation over the infinite others — an anti-reification verdict that hands a tethered duplicate to Chapter 22.

---

Notes

1. Hilary Putnam, Representation and Reality (Cambridge, MA: MIT Press, 1988), ch. 6 and the Appendix, esp. pp. 120–125 (the page range at which Chalmers locates the theorem): every ordinary open physical system is a realization of every abstract finite automaton. The result requires the system be open (interacting with an environment so its states do not cycle) over the relevant interval; the proof constructs the needed state-to-state mapping by exploiting the density of distinct physical states. The strength of the claim — every automaton, not merely many — is what makes it a triviality result rather than a mere restatement of multiple realizability. ↩
2. John R. Searle, “Is the Brain a Digital Computer?,” Proceedings and Addresses of the American Philosophical Association 64, no. 3 (1990): 21–37, at 26–28: “Syntax is not intrinsic to physics… computation is observer-relative. There is no fact of the matter intrinsic to a physical system about whether it is computing or not.” Searle’s 1990 paper does work distinct from the Chinese Room of Chapter 20: the Chinese Room argues that syntax does not suffice for semantics; “Is the Brain a Digital Computer?” argues that nothing in a physical system is intrinsically syntactic in the first place. The restatement in less technical register appears in The Mystery of Consciousness (New York: New York Review of Books, 1997), ch. 1. The observer-relativity of syntax is the load-bearing premise; for its extension to semantics see Teodor Negru, “Intentionality and Background: Searle and Dreyfus against Classical AI Theory,” Filosofia Unisinos 14, no. 1 (2013): 18–34. ↩
3. Putnam, Representation and Reality, esp. chs. 5–6. The reversal is genuine and unhedged: the multiple realizability that recommended functionalism in the 1960s is, by Putnam’s 1988 lights, exactly what dissolves it once the realization relation is seen to be unconstrained. On the philosophical and rhetorical weight of functionalism’s founder arriving at this verdict, see Oron Shagrir, “The Rise and Fall of Computational Functionalism,” in Hilary Putnam, ed. Yemima Ben-Menahem (Contemporary Philosophy in Focus; Cambridge: Cambridge University Press, 2005). This chapter treats the Putnam–Searle convergence as evidential, not merely rhetorical: independent derivations of one conclusion from incompatible starting points. ↩
4. Peter Godfrey-Smith, “Triviality Arguments against Functionalism,” Philosophical Studies 145, no. 2 (2009): 273–295. Godfrey-Smith’s inventory separates the versions of the triviality argument that succeed from those that overreach and identifies the minimal additional structure — causal, and ultimately selection-historical — that a functionalism hoping to survive must take on. His verdict, that the strongest versions push functionalism toward grounding in causal and teleological facts, converges with the position §VI defends. ↩
5. David J. Chalmers, “Does a Rock Implement Every Finite-State Automaton?,” Synthese 108, no. 3 (1996): 309–333; the compressed version is in The Conscious Mind: In Search of a Fundamental Theory (New York: Oxford University Press, 1996), ch. 11. Chalmers requires that a physical system implement a computation only if it possesses a combinatorial-state automaton structure: its internal states factor into components whose transitions are reliable, counterfactual-supporting, and isomorphic to the formal computation’s state-transition structure — which rules out the Putnam- and Searle-style gerrymandered mappings, precisely because those exploit the absence of such counterfactual structure. Ron Chrisley, “Why Everything Doesn’t Realize Every Computation,” Minds and Machines 4, no. 4 (1994): 403–420, presses a kindred constraint. For the subsequent debate over whether constraints like Chalmers’s suffice — and over whether implementation carries a representational requirement — see Matthias Scheutz, “When Physical Systems Realize Functions…,” Minds and Machines 9, no. 2 (1999): 161–196; Mark Sprevak, “Three Challenges to Chalmers on Computational Implementation,” Journal of Cognitive Science 13 (2012): 107–143; and Michael Rescorla, “The Computational Theory of Mind,” Stanford Encyclopedia of Philosophy. The text grants Chalmers the technical victory and locates the residual point one level up: a constraint on which computation a system realizes is not a constraint on what that computation is about — close to the worry Sprevak presses in his own terms. ↩
6. Gualtiero Piccinini, “Computationalism, the Church–Turing Thesis, and the Church–Turing Fallacy,” Synthese 154, no. 1 (2007): 97–120. Piccinini distinguishes computationalism proper — the empirical thesis that cognition consists in the brain’s computation of certain Turing-computable functions — from the far weaker claim that brain functions are Turing-computable (which follows from physicalism plus the Church–Turing–Deutsch principle). The inference from the latter to the former he names the Church–Turing fallacy; the paper shows several historically influential versions of it (Newell, Pylyshyn, and others) to be unsound. ↩
7. The fallacy label is Piccinini’s (n. 6); Copeland is the authority on the careful statement of the thesis itself and on the narrow/wide-mechanism distinction the bad inference trades on — see B. Jack Copeland, “The Church–Turing Thesis,” Stanford Encyclopedia of Philosophy, and “Narrow Versus Wide Mechanism,” Journal of Philosophy 97 (2000): 5–32. Church and Turing published independently in 1936 (Church using recursive functions, Turing the abstract machines now named for him); the formalisms proved equivalent. The thesis has the form: for any function f, if f is intuitively computable, then some Turing machine computes f. It does not say that every physical system is a computer, that every physical process is a computation, or that a computable mind is a program — each a further claim requiring its own argument. ↩
8. The three requirements track Piccinini’s account of physical computation; see Piccinini, “Computational Modelling vs. Computational Explanation: Is Everything a Turing Machine, and Does It Matter to the Philosophy of Mind?,” Australasian Journal of Philosophy 85 (2007): 93–115, and Physical Computation: A Mechanistic Account (Oxford: Oxford University Press, 2015). Piccinini’s mechanistic account — a system computes only if it possesses a mechanism functionally defined over discrete state-types that it manipulates by rule — is the closest thing in the current literature to a substantive criterion, and it is precisely what makes computationalism a falsifiable empirical hypothesis rather than a trivial truth. On his view brains may turn out to be computers; the point is that the question is empirical and substantive, not settled from the armchair by a theorem about computable functions. ↩
9. The teleosemantic grounding gestured at here is developed in Part Three; see Chapter 16 (teleosemantics) and Chapter 17 (semantic externalism), resting on Ruth Garrett Millikan, Language, Thought, and Other Biological Categories (Cambridge, MA: MIT Press, 1984), chs. 1–2 on proper functions, and Fred Dretske, Explaining Behavior: Reasons in a World of Causes (Cambridge, MA: MIT Press, 1988), for the parallel project through learning history. Whether artificial systems trained on human corpora could inherit learning-historical proper functions in Millikan’s sense is a live frontier; see Jumbly Grindrod, “Large Language Models and Linguistic Intentionality,” Synthese 204 (2024): article 71, https://doi.org/10.1007/s11229-024-04723-8. Millikan’s own commitments suggest she would resist the extension, since the histories she has in mind run through natural selection and individual learning rather than gradient descent over a corpus. ↩