Defense against Impersonal Realism - part III

Refutation of The Teleosemantic Argument

Formulated by Claude Sonnet 4.6

Definitions: Locking the Terms

Correctness conditions are the standards relative to which a state, output, or process can be evaluated as accurate or inaccurate. A map has correctness conditions: it is accurate if it corresponds to the territory. A belief has correctness conditions: it is true if it corresponds to the facts. Correctness conditions are third-personal — they can be specified entirely from outside the system being evaluated. They require no perspective internal to the system.

Malfunction is deviation from a system's operating profile — the output the system is designed, selected, or constituted to produce. Malfunction is defined relative to a function. Functions are assigned by design, evolution, or systemic role. All of these are third-personal attributions. Nothing in malfunction requires a subject for whom the deviation is a failure. The thermostat doesn't experience its misreading. The network doesn't suffer its misclassification.

Error — genuine error, as opposed to malfunction — is failure to meet a standard that is operative for a subject. Not merely applicable to a system from outside, but binding for a perspective from inside. Error requires that the subject be answerable to the standard: that getting it wrong is not merely deviating from a profile but failing — where the failure has a direction, a subject, and a genuine normative weight that is not exhausted by third-personal description of the deviation.

Answerability is the condition under which a standard is binding for a subject rather than merely applicable to a system. A system has correctness conditions. A subject is answerable. The difference: correctness conditions can be specified without remainder in third-personal terms. Answerability cannot. It requires a perspective from which the standard registers as binding — from which getting it wrong is not merely producing a discrepant output but being wrong.

Binding as used here does not mean moral blameworthiness. It means: operative for a subject such that deviation constitutes genuine failure rather than mere mismatch. A logical inference is binding in this sense: if you accept the premises and deny the conclusion, you haven't merely deviated from a profile — you have made a mistake, one for which you are answerable to the standard of validity. That answerability is what binding means throughout this argument.

Interiority is not rich psychological selfhood, qualia, or narrative consciousness. It is the minimal condition under which modal and normative distinctions are operative rather than merely described: the condition of being a for which rather than a for which it is the case that. A system has interiority if and only if standards can be binding for it — if it can make genuine errors rather than merely malfunction.

These definitions are not stipulative. They track the distinctions that collapse if conflated. The argument that follows depends on keeping them separate. Any response that treats correctness conditions as sufficient for answerability, or malfunction as sufficient for genuine error, has conflated the terms defined above and must show either that the distinction fails or that the conflation is principled.

The Teleosemantic Counter — and Its Demolition

The Counter, In Its Strongest Form

Teleosemantics, developed by Millikan and extended by Papineau and others, offers the most sophisticated naturalist account of content, correctness, and error. Here is its best case, stated without charity withheld:

Evolution selects for mechanisms that track features of the environment. A frog's fly-snapping mechanism has the biological function of tracking flies. When it snaps at a black dot that isn't a fly, it isn't merely deviating from a statistical baseline — it is misfiring, failing to perform its proper function, the function that explains why the mechanism exists and persists. That misfire is a genuine error, not a mere mismatch. The correctness conditions are grounded in natural history, not in external attribution. The frog's visual system is supposed to track flies — supposed to in a sense that is not merely our interpretive stance but a fact about what the system was selected to do.

Scale this up. Human cognition has proper functions grounded in evolutionary and developmental history. Our logical faculties are supposed to track valid inference — they were selected for because tracking validity is fitness-enhancing. When we reason fallaciously, we are not merely deviating from a profile we assign ourselves from outside. We are failing to perform a function that is ours in virtue of our natural history. That failure is genuine error. Normativity is naturalized. No interiority required beyond what evolutionary biology can describe.

This is the serious version. It doesn't just say "error is misrepresentation." It says: proper functions ground genuine normativity because they are not assigned from outside — they are internal to the system's biological nature. The "supposed to" is not third-personal attribution. It is a fact about what the mechanism is.

Demolition: Move 1 — Proper Function Is Still Third-Personal

The teleosemantic account claims that proper functions are not assigned from outside but are internal to the system's nature. This is the crucial move, and it fails for a precise reason.

A proper function, on Millikan's account, is defined by reproductive history: a mechanism has the function F if it exists because ancestral tokens performed F and were selected for it. This is a historical-causal fact. It is entirely third-personal. It says nothing about the current mechanism's perspective on its own operation. The frog's visual system has the proper function of tracking flies in exactly the same sense that the heart has the proper function of pumping blood — which is to say, in a sense that involves no perspective internal to the frog or the heart. The heart doesn't know it's supposed to pump blood. The frog's visual system doesn't know it's supposed to track flies.

The teleosemantician will say: but cognition is different because cognitive systems represent their own functions.

We already showed this fails. Self-representation is another functional state. A system that represents its own proper function has a more complex functional profile. The self-representation is itself subject to correctness conditions — it accurately represents the proper function or it doesn't. That correctness question regresses immediately to the same structure: the self-representation either tracks or misfires. Nothing in that tracking relation generates answerability. The regress doesn't bottom out in a perspective. It bottoms out in another functional state.

Proper function explains why we attribute error to systems. It does not explain why error is binding for systems. The frog that snaps at a dot has misfired according to its evolutionary history. But the misfire is a fact about the frog's causal-historical profile, not a fact about the frog's answerability to a standard. The frog is not answerable to the standard of fly-tracking. It merely instantiates a mechanism that was selected for fly-tracking and is currently failing to do so.

That is malfunction. Not error. The definitions are not satisfied.

Demolition: Move 2 — Selection Explains Attribution; It Cannot Explain Answerability

Here is the precise gap, stated as sharply as possible.

Teleosemantics succeeds completely at explaining why we treat certain states as having content, correctness conditions, and error-conditions. It gives a principled, non-arbitrary account of why "fly" is the content of the frog's visual state rather than "ambient light reflected at a certain frequency" — because the mechanism was selected for fly-tracking, not light-tracking. This is a genuine contribution. The content-attribution is not arbitrary.

But explaining why content-attribution is principled is not the same as explaining why the standard is binding for the system. The principled attribution is still attribution. It is still third-personal. The fact that evolutionary history grounds the attribution in something real rather than our interpretive whim does not move the attribution from outside the system to inside it.

Here is the test. Ask: for whom is the standard binding?

On the teleosemantic account, the answer is: the standard is binding for us, the attributors, in the sense that we have principled grounds for our attribution. The evolutionary history makes the content-attribution correct rather than arbitrary. But the standard is not binding for the frog. The frog has no perspective from which the misfire registers as a failure rather than just a causal event. The frog's answerability is entirely a projection of our framework onto its mechanism.

This is exactly the distinction the definitions above are tracking. Correctness conditions grounded in proper function are still correctness conditions — still third-personal, still specifications of what the system is supposed to do from outside. The grounding in evolutionary history makes the attribution principled. It does not make the standard operative for the system.

Now the consequence.

If teleosemantics is the best naturalist account of normativity — and it is — and if it still only yields principled third-personal attribution rather than first-personal answerability, then no naturalist account yields genuine answerability.

The gap between malfunction and genuine error cannot be closed by making the attribution of malfunction more principled. The gap is categorical, not a matter of degree of principled-ness. No amount of principled grounding in natural history, evolutionary selection, or systemic role generates the condition under which a standard is binding for a subject rather than applicable to a system.

That condition — the for which — is not a functional property. It is not a representational property. It is not a property grounded in historical causation. It is the condition under which the system is an answerable subject rather than a profiled mechanism.

And that condition is what we mean by interiority.

Teleosemantics is the most sophisticated naturalist attempt to close the gap. It closes everything except the gap. It explains content. It explains misrepresentation. It explains why attributions are principled rather than arbitrary. It does not — and cannot — explain why any of this is binding for the system rather than applicable to it.

The gap is not closed by making attribution more principled. The gap is between attribution and answerability. No account that remains purely third-personal crosses it.