The Coupling-Constitution Fallacy

In the AI Story

The objection has real force. Not every causal contribution to cognition is cognitive. The coffee I drank this morning contributes causally to my current reasoning, but nobody thinks the coffee is part of my mind. The objection demands that extended mind theorists specify a stronger criterion than mere causal contribution — something that distinguishes genuine cognitive constitution from mere causal coupling.

Clark's response operates on multiple levels. First, he points out that the objection applies equally to internal processes. Neural activity in the visual cortex is causally coupled to my conscious experience of seeing, but on the Adams-Aizawa logic, this alone does not make it part of my cognition. The response requires some criterion that makes internal coupling genuinely constitutive while ruling out external coupling — and the only such criterion available is substrate-based: that internal processes are cognitive because they occur in biological tissue. This is what Clark calls bioprejudice: the presupposition that cognition is skull-bound, used to dismiss evidence that it is not.

Second, Clark and collaborators have developed what he calls the glue-and-trust conditions — the additional conditions that external processes must meet to count as genuine cognitive extension. The resource must be reliably present. It must be readily available when needed. Its outputs must be automatically endorsed. There must be a history of trust. These conditions narrow the scope of extension considerably, ruling out random causal couplings while preserving the cases — like Otto's notebook — that motivated the thesis.

Third, the objection grows weaker as the external component becomes more cognitively sophisticated. The notebook case was arguable. The AI case is harder to dismiss. When the external component performs association, inference, conceptual integration, and linguistic synthesis — the paradigmatic functions of sophisticated cognition — the insistence that these functions are "mere coupling, not constitution" begins to look like what it is: the presupposition that cognition must be biological, applied to exclude evidence that it is not.

Origin

Adams and Aizawa articulated the objection most fully in their 2008 book The Bounds of Cognition, though earlier papers established the framework. The debate between their internalist position and Clark's externalist position has defined philosophy of mind for a generation.

The debate has intensified in the age of AI. Some philosophers argue that generative AI vindicates the extended mind thesis so thoroughly that the coupling-constitution objection collapses; others argue that AI's characteristic failures — confident fluency without embodied grounding — show that the objection was right all along. Clark's position is that the AI case clarifies rather than dissolves the debate: the coupling is real, the constitution is genuine, and the specific vulnerabilities of AI coupling require new disciplines rather than a retreat to internalism.

Key Ideas

Not every coupling is cognitive. The objection correctly insists that mere causal contribution is not enough; additional conditions must be met.

But the criterion must apply to both sides. Any principle that excludes external coupling from cognition must explain why internal coupling counts — and the available answers reduce to bioprejudice.

Glue-and-trust narrows the scope. Reliability, availability, endorsement, and history of trust distinguish genuine extension from accidental coupling.

AI sharpens the issue. When the external component performs sophisticated cognitive functions, the dismissal of coupling as non-constitutive becomes increasingly hard to sustain without circularity.

The debate matters for institutions. How we answer the coupling-constitution question shapes how we assign responsibility, design professional standards, and educate for the AI age.