02 · Metaphysics
Part I's second hearing: with logic's family quarrel behind it, the court turns to a plainer question — what kind of object is sitting in the dock at all.
What kind of thing is a program?
In 1988 Hilary Putnam published Representation and Reality (MIT Press) with what he called a theorem: every ordinary open physical system — anything not perfectly sealed off from its environment — implements every finite-state automaton. Two years later, in his presidential address to the American Philosophical Association, John Searle pushed the same point to a memorable extreme: "the wall behind my back is right now implementing the Wordstar program" — because some pattern in its molecular motion is isomorphic with Wordstar's formal structure ("Is the Brain a Digital Computer?", Proceedings and Addresses of the APA 64, 1990: 21–37). If a wall runs Microsoft's word processor, then computation is not a fact about the world — it is a story an observer chooses to tell about any sufficiently complicated system, brains included. Putnam's own triviality result, aimed originally at his own earlier theory, exploded computational functionalism's central promise: that being a mind is a matter of running the right program, no matter what you're made of. If any wall runs any program, running the right program cannot be what makes something a mind.
The wall argument is where this chapter has to start, because it forces the question philosophy usually skips: before asking whether a program can think, ask what a program even is.
The first principle is the split between the abstract and the concrete — computing's own version of a distinction Charles Sanders Peirce drew for language in 1906: type versus token. Count the letters on this page and you can answer two different, equally correct questions — how many letter-shapes appear (tokens), and how many distinct letters of the alphabet they instantiate (types) ("Prolegomena to an Apology for Pragmaticism," The Monist 16, no. 4, 1906: 505–506). A piece of software works the same way. "Microsoft Word" the program is a type — an abstract pattern of instructions, nowhere in particular, no more locatable than the number seven. The install on your laptop, and the running process consuming memory on it right now, are tokens — concrete, datable, capable of crashing. When your copy of Word freezes, Word itself, the abstract program, has not broken; a token has. Software's whole industry runs on treating the type as the real product and the token as a disposable instance of it — which is exactly why a "critical vulnerability in Word" can be announced before your particular copy has even loaded.
The second principle is identity, and it is harder than it sounds: when are two programs the same program? Take a bubble sort written in Python and the identical algorithm rewritten, variable for variable, in C. Same steps, same inputs and outputs, different languages — same program, or two? Now take one Python file compiled by two different optimizing compilers into two binaries that never produce a single identical machine instruction but always agree on output — same program? Philosophers of computation have no settled answer, and the disagreement tracks a real fork: identify a program with its behavior (the input–output function it computes) and the two binaries are one program; identify it with its structure (the algorithm, the specific steps) and a compiler's optimizations may quietly turn it into something else that merely agrees on outputs so far tested. Software engineers live this dispute daily without naming it, every time they ask whether a refactor "changed the program" or just changed its clothes.
The third principle is the wall itself, formalized. Putnam's proof works by constructing, after the fact, a mapping from a chosen sequence of the wall's molecular microstates onto the automaton's state transitions — a mapping guaranteed to exist for any finite stretch of time because there are always enough distinguishable microstates to go around. David Chalmers's reply, "Does a Rock Implement Every Finite-State Automaton?" (Synthese 108, 1996: 309–333), grants the construction and denies it proves what Putnam needs. A genuine implementation, Chalmers argues, must track the automaton's transitions not just along the one trajectory the world happened to take, but counterfactually — for every input the automaton could have received, the physical system's causal structure must have produced the matching state, whether or not that input actually occurred. Putnam's wall satisfies the mapping only along the single path molecules actually followed; feed the wall a different "input" and nothing in it is disposed to track the automaton's corresponding response, because nothing in a wall is wired to respond to inputs at all. A laptop running WordStar, by contrast, really is built so that pressing a different key really would produce the counterfactually correct state. The reply does not make implementation cheap; it makes it a genuine, checkable, causal fact about a system's dispositions — restoring exactly the distinction Searle's wall was built to erase.
The fourth principle takes the same question into worlds built entirely of information: virtual realism. In Reality+: Virtual Worlds and the Problems of Philosophy (W. W. Norton, 2022) and the paper that preceded it, "The Virtual and the Real" (Disputatio 9, no. 46, 2017: 309–352), Chalmers argues that objects inside a simulated world are not lesser or fictional — they are real digital objects, meeting the marks philosophers use to certify anything as real: they exist, they have causal powers, they are not merely someone's say-so, and they are mind-independent once the simulation is running. A sword forged inside a game world is not made of steel, but it is made of data structures that really exist, really can be transferred, and really can be sold for currency that buys real coffee. Chalmers's target is not that virtual and physical objects are identical in kind — a virtual river is not H2O — but that being virtual is not the same as being unreal; it is one more way of being real, alongside being physical.
The fifth principle is a horizon this chapter only touches: the ontology of information itself. Luciano Floridi's informational structural realism (The Philosophy of Information, Oxford University Press, 2011) proposes that what ultimately exists is not substance but structured information — "dedomena," primordial data-like differences — of which physical objects and virtual objects alike are patterns at different levels of abstraction. If Floridi is right, the type/token puzzle and the virtual-object puzzle are not two separate metaphysical headaches produced by computers; they are two symptoms of the same deeper claim, that information was always the more basic category and substance the derived one. Chapter 11 audits that claim on its own ground, against Shannon's severing of information from meaning; this chapter only notes that metaphysics, once it starts asking what a program is, cannot easily stop there.
The people, briefly. Hilary Putnam changed his mind about the philosophy of mind more publicly than almost anyone in the field, ending his life doubting the very functionalism he had founded. Searle's wall is a philosopher's weapon turned, characteristically, against a friendly-adjacent target — computational theories of mind he otherwise also opposes on different grounds (see chapter 04). Chalmers has spent a career doing what this chapter shows twice: taking a claim meant to dissolve a distinction and finding the exact repair that saves it.
Connections
Chapter 04 needs this chapter's implementation problem directly: if implementing a program is not enough to be a mind, functionalism about mental states inherits the same objection Searle raised here. Chapter 11 owns the full audit of information as a candidate ultimate substance. Chapter 15 asks the largest version of the virtual-realism question: is the physical universe itself the simulation.
Open questions
If Chalmers is right that implementation requires tracking counterfactual structure, is that structure an objective fact about a physical system, or does specifying "the right counterfactuals" still require an observer to say which inputs count? And can one consistently hold, as Chalmers does, that game-world objects are real without holding that the base universe is itself a simulation — or does virtual realism quietly commit you to taking chapter 15's simulation argument more seriously than you intended?
Go deeper
- David J. Chalmers, Reality+: Virtual Worlds and the Problems of Philosophy, W. W. Norton, 2022 — the book-length case for virtual realism, built from the implementation problem up.
- Hilary Putnam, Representation and Reality, MIT Press, 1988 (the triviality argument); David J. Chalmers, "Does a Rock Implement Every Finite-State Automaton?", Synthese 108 (1996): 309–333 (the reply).
- Types and Tokens, Stanford Encyclopedia of Philosophy.