Fred Adams, Ken Aizawa (2010), “Defending the Bounds of Cognition”, in Richard Menary (ed.), The Extended Mind, Cambridge, MA: MIT Press, pp. 67–80 — DOI-10.7551/mitpress/9780262014038.003.0004. Read from a draft that predates it (see the history note).
Key takeaways
- The fallacy, named. “When Clark makes an object cognitive when it is connected to a cognitive agent, he is committing an instance of a ‘coupling-constitution fallacy.’” The pattern: point to a case where an object or process is coupled to a cognitive agent, then “slide to the conclusion that the object or process constitutes part of the agent’s cognitive apparatus”. Coupling is not constitution: the neurons into a neuromuscular junction are coupled to the muscle but are not part of it. The authors call it “the most common mistake that extended mind theorists make” and charge Van Gelder and Port, Clark and Chalmers (LIT-097), Clark 2001, Gibbs and Haugeland with it.
- What is needed instead is a mark of the cognitive. The 2001 proposal is restated: cognition is “constituted by certain sorts of causal processes that involve non-derived content”, and as a contingent empirical fact some processes that are recognisably cognitive occur only in the brain.
- The non-derived-content condition is not all-or-nothing. Cognitive states must involve non-derived content, not consist of it throughout. A process with no intrinsic content is ruled non-cognitive.
- Euler circles. Clark’s case of an imagined Venn or Euler diagram, whose meaning is conventional, confuses how a diagram on paper gets its meaning with how a mental image of it does. That a convention had to exist first is a historical dependence, like a mental image of a car depending on the car’s invention, not a fact about what constitutes the image’s content. And conventions need access to the syntactic items they attach meanings to, which nobody has for brain states.
- No brain-tool science. Even if information processing crosses the skull, brains process information by different principles from consumer electronics. Tools are artifacts, an open-ended set, so a science of brain-plus-tool systems would have to cover a “motley”, not just a multiplicity. Intracranial processes may be many, but the brain’s development bounds them.
Standing in the record
Filed on 2026-10-02 at the owner’s request, from a text fetched while the critics of LIT-097 were filed. NOTE-380 is the close reading, and placed it: Active, with the caveat that what was read is a draft.
Which text this is. The chapter in Menary’s volume, in a draft written for it (“Clark does not refer to this argument in this volume”, n. 4). It replies to a manuscript of Clark’s “Memento’s Revenge: The Extended Mind, Extended”, which the volume prints as the chapter before it (pp. 43–66, DOI-10.7551/mitpress/9780262014038.003.0003, per Crossref; not filed here). The draft is not a reply to Clark 2005 (LIT-403), though the two Clark texts share material: LIT-403 imagines “the partially overlapping circles of a certain Venn diagram” and Martians who store bit-mapped text, and the manuscript quoted here has both. But the draft’s only Clark reference of the period is “Memento’s Revenge” (forthcoming), and it quotes from it three things LIT-403 does not contain: the three coupling conditions (reliably available, automatically endorsed, easily accessible), a passage on complexity theory and power laws, and the Euler-circle sentence in different words (“imaging a set of Venn Diagrams/ Euler Circles in our mind’s eye”). Its reference list cites “Defending Non-Derived Content” as under review; that paper appeared in Philosophical Psychology 18(6) in December 2005 (DOI-10.1080/09515080500355186, per Crossref), so the draft is earlier than that.
The “coupling-constitution fallacy”. This is where the record first meets the name. LIT-438 (2001) argues the point, with the thermostat’s bimetallic strip, but does not name it, and its status note says the name “comes from the authors’ later work”. This text is that later work, or the earliest piece of it the record holds: it introduces the name as something “we might call” the fallacy and defers “a less quick and dirty treatment” to a paper then “in progress”. NOTE-180 and Chalmers’s 2019 chapter (LIT-415) attribute the name to the authors’ 2008 book. The book is the earlier print appearance (the chapter came out in 2010), but this draft shows the name in circulation before the end of 2005. Whether any earlier published text uses it is not settled here.
How it bears on LIT-097 and LIT-403. It is the second round of the exchange that LIT-438 opened and LIT-403 answered. Against LIT-403’s Venn-diagram argument it gives the record’s one direct reply: the conventional content belongs to the diagram on paper, and nothing Clark says shows the mental image gets its content the same way. It does not answer LIT-403’s other moves (silicon could carry intrinsic content; it would not matter). Its new charge against LIT-097 is the one NOTE-180 found unargued in the 1998 paper: the step from coupling to parthood.
Mereology. The fallacy is a claim about parthood: being coupled to Y
does not make X a part of Y, for objects or for processes. That is why
mereology is tagged. The text gives no criterion of parthood in return;
it relocates the question to the nature of X (“The nature of X, of
course”).
It carries no instruction for machine-learning practice.