NOTE-180: Clark & Chalmers — The Extended Mind

NOTE-xxx: Clark & Chalmers — The Extended Mind

Contribution

The paper separates “active” externalism from the Putnam–Burge “passive” kind. It then argues that cognitive processes, and at least one class of mental state (standing, dispositional beliefs), can be partly located in or constituted by external resources that are coupled to the organism. Before it, externalism in the philosophy of mind was mainly about content fixed by distal and historical facts. After it, there is a named, case-based thesis that the vehicles of cognition can extend. That thesis comes with a paradigm case (Otto), a heuristic (the parity sentence of §2) and a list of features that mark clear cases (§5). It closes with a brief argument that the self extends as well (§5).

Key insight

Belief, like cognition generally, is individuated by the role it plays: “What makes some information count as a belief is the role it plays, and there is no reason why the relevant role can be played only from inside the body” (§4). Grant that, and ask what deep difference separates Otto’s notebook from Inga’s memory other than the skin, which would beg the question. On the paper’s argument none is left, so the boundary of the mind is set by reliable functional coupling and not by the skull.

Assumptions

Premises and authorities. None of them is argued for at length, and several are stated, not defended.

• Role-functionalism about belief and cognition. Mental states and processes are individuated by their functional or explanatory role: “What makes some information count as a belief is the role it plays” (§4); “insofar as beliefs and desires are characterized by their explanatory roles, Otto’s and Inga’s cases seem to be on a par” (§4). This premise carries the constitutive conclusion, and the paper does not defend it.
• Dispositional belief is real belief. Non-occurrent beliefs count as beliefs (Inga “believed this even before she consulted her memory”, §4). The paper admits that denying this, by accepting only occurrent beliefs, is “the most consistent way to deny Otto’s belief”, and calls that view “extreme” (§4).
• The skin/skull boundary gets no default weight. Appealing to it “would be to beg the question” (§2, §4). The burden is put on the opponent to find “some more basic underlying difference” (§4).
• Explanatory simplicity is evidence about what exists. Explaining Otto by way of beliefs about his notebook “is pointlessly complex”, and “In an explanation, simplicity is power” (§4). A parallel argument is made for Scrabble in §3.
• Cognition need not be conscious. Many cognitive processes are non-conscious, so the fact that consciousness is not extended does not rule out extended cognition (§3). The paper allows that “some mental states, such as experiences, may be determined internally” (§4). The thesis does not claim that consciousness extends.
• Revisionary conceptual choice. “We do not intend to debate what is standard usage; our broader point is that the notion of belief ought to be used so that Otto qualifies”, because this makes “belief” “more akin to a natural kind” (§4). Part of the conclusion is therefore a recommendation about the concept and not a discovery.
• Empirical authorities cited: Kirsh & Maglio (1994) on Tetris and epistemic action, Kirsh (1995) on Scrabble tiles, Hutchins (1995), Suchman, Beer, Thelen & Smith, Ullman & Richards, Blake & Yuille, and Triantafyllou & Triantafyllou (1995). They are used as illustrations. The paper does no empirical work of its own.

Key results

What it argues, stated exactly.

• Tetris cases (§2). (1) A person mentally rotates on-screen 2-D shapes to judge whether they fit “sockets”. (2) The person can choose either to rotate the image physically by pressing a rotate button, which has some speed advantage, or to rotate it mentally. (3) In a “cyberpunk future”, a neural implant does the rotation as fast as the computer in (2), and the agent must choose between the implant and mental rotation. The claim is that “all three cases are similar”. Case (3) is “clearly … on a par with case (1)”, and case (2) “displays the same sort of computational structure as case (3)” but is distributed across agent and computer. The only thing that separates (2) from (3) is the skin/skull boundary, “precisely what is at issue”. Cases (1) and (2) are real options in Tetris. Kirsh & Maglio (1994) measure physical rotation through 90° at about 100 ms plus about 200 ms to select the button, against about 1000 ms for mental rotation. Physical rotation is often used to find out whether a shape fits, which Kirsh and Maglio call an “epistemic action”.
• The parity sentence (§2), verbatim. “If, as we confront some task, a part of the world functions as a process which, were it done in the head, we would have no hesitation in recognizing as part of the cognitive process, then that part of the world is (so we claim) part of the cognitive process.” It is introduced by “Epistemic action, we suggest, demands spread of epistemic credit” and followed by “Cognitive processes ain’t (all) in the head!” The hedge “(so we claim)” is in the original.
• Active vs passive externalism (§3). Active: “the human organism is linked with an external entity in a two-way interaction, creating a coupled system that can be seen as a cognitive system in its own right”. All components “play an active causal role” and “jointly govern behavior”. Removing the external component makes competence drop “just as it would if we removed part of its brain”. The external features are “in the loop” and act “in the here-and-now”. Passive (Putnam 1975, Burge 1979): the external features that make the difference in content are “distal and historical, at the other end of a lengthy causal chain” and play “no role in driving the cognitive process in the here-and-now”. So twins behave in physically indistinguishable ways, and if one twin is teleported to Twin Earth, his present surroundings do not change his contents. The paper takes the difference to answer the explanatory-irrelevance worry about content externalism. With active externalism, “if we retain internal structure but change the external features, behavior may change completely”. A footnote suggests the better question may be “is mind in the present?” rather than “is mind in the head?”
• Objections to extended cognition and the replies (§3). Consciousness: not all cognition is conscious. Portability, the “Naked Mind”: the paper concedes “there is something to this objection” but argues that contingent coupling does not rule out cognitive status (a plug-in memory module is “just as cognitive”). The real point of the portability intuition is that “reliable coupling is required”. A calculator or Filofax that is “always there when I need them” is “coupled with me as reliably as we need”. Occasional decoupling does not count against it, because brains also lose capacities in sleep, intoxication and emotion. Evolution and development: brains may have evolved and developed to rely on external structure. Language is the main case, and “extended cognition is a core cognitive process, not an add-on extra”. Fish analogy: “The fish and surrounding vortices together constitute a unified and remarkably efficient swimming machine”. Words and external symbols are “paramount among the cognitive vortices which help constitute human thought”.
• Otto and Inga (§4). Inga hears of a MoMA exhibition, recalls that the museum is on 53rd Street, and walks there. She believed it before she recalled it, as a non-occurrent belief. Otto has Alzheimer’s disease and carries a notebook everywhere. He writes down new information and looks up old information, so the notebook “plays the role usually played by a biological memory”. He consults it, reads 53rd Street, and walks there. The claim is that Otto believed the museum was on 53rd Street before he looked it up, because “the notebook plays for Otto the same role that memory plays for Inga”. The rival explanation (a standing belief that the museum is at the address in the notebook, plus the fact that the notebook says 53rd Street) adds “an extra term” to every explanation involving the notebook and is “one step too many”.
• Twin Otto (§4). Twin Otto is a physical duplicate of Otto “from the skin in” whose notebook wrongly says 51st Street. He “is best characterized as believing that the museum is on 51st Street”. “In these cases, a belief is simply not in the head.” Unlike the Putnam–Burge twins, Otto and Twin Otto behave differently: one walks to 53rd Street, the other to 51st. A footnote puts this in Chalmers’ notional/relational content terms: the twins differ in notional content because “the cognitive system is itself effectively extended to include the notebook”.
• Replies on the Otto–Inga differences (§4). Reliability: Otto’s access is “very reliable”, though not perfectly, and Inga’s is not perfect either (a surgeon might tamper, or she might drink too much). The paper grants that “constancy is relevant”: a one-off consultation of a guidebook would not yield a standing belief. Intermittent access (showering, darkness): “What really counts is that the information is easily available when the subject needs it.” Bandwidth: Lucy has a low-grade link to her biological memory and still believes, as long as the information is accessible. A struggle, “mixed results”, or the need for a psychotherapist’s help would make the paper more reluctant to ascribe the belief. Perception vs introspection: treating this as a difference begs the question, because from the standpoint of the Otto+notebook system the information flow “is not perceptual at all”. The only residual difference is phenomenology, which is illustrated with the Terminator’s visual recall. All of these are “shallow differences”.
• Coupled-resource features (§5). Constancy, direct availability, automatic endorsement and past conscious endorsement, quoted in the corrections above. The first three “certainly play a crucial role”, and the fourth’s status “as a criterion for belief is arguable” (subliminal perception, memory tampering). A footnote concedes that constancy and past endorsement “may suggest that history is partly constitutive of belief”. It offers two remedies: a purely dispositional reading of constancy with past endorsement dropped, or allowing history “as long as the main burden is carried by features of the present”. Applications: the Filofax counts if “I rarely take relevant action without consulting” it. The Internet “is likely to fail on multiple counts, unless I am unusually computer-reliant, facile with the technology, and trusting”. “Certain files on my computer may qualify”.
• Socially extended belief (§5). “Could my mental states be partly constituted by the states of other thinkers? We see no reason why not, in principle.” In an unusually interdependent couple, one partner’s beliefs can play the notebook’s role (the Woodens footnote, from the New York Times of 30 March 1995: Nell Wooden “served as his memory bank”). The requirement is “a high degree of trust, reliance, and accessibility”. Other examples: a waiter, which the paper suggests may be “extended desire”, a secretary, an accountant, a collaborator. A footnote on Burge’s arthritis case: deferring to a doctor through an active relationship would give something like a true belief that I have “some other disease in my thigh”. Language “is not a mirror of our inner states but a complement to them”.
• Extended self (§5), stated exactly. “Does the extended mind imply an extended self? It seems so.” The argument runs as follows. The self already “outstrips the boundaries of consciousness”, because dispositional beliefs “constitute in some deep sense part of who I am”. “If so, then these boundaries may also fall beyond the skin.” The information in Otto’s notebook “is a central part of his identity as a cognitive agent”, so “Otto himself is best regarded as an extended system, a coupling of biological organism and external resources”. Resisting this would “shrink the self into a mere bundle of occurrent states, severely threatening its deep psychological continuity”. Consequences: “in some cases interfering with someone’s environment will have the same moral significance as interfering with their person”. Some social activity may be “more akin to thought”.

Constitutive rather than causal: what makes it so, and is it argued?

The paper states the thesis in constitutive and locational terms. An external resource is “part of the cognitive process” (§2). The coupled system is “a cognitive system in its own right” (§3). Beliefs “can be constituted partly by features of the environment” (§4). Otto’s internal processes and his notebook “constitute a single cognitive system” (§4). A belief “is simply not in the head” (§4). The self is “spread into the world” (§5). The paper’s own contrast is with the causal-embedding reading, the rival explanation of §4 in which Otto has an internal belief about the notebook and the notebook is an external input. What makes the claim constitutive is that the external item is counted among the realisers of the state, as the vehicle of a standing belief, and not among the causes of an internal state.

The paper supports this step with three moves, none of them an argument from causation to parthood.

1. Role-functionalism. If belief just is occupying a role, and the notebook entry occupies it, the notebook entry realises the belief. This is where the constitutive conclusion really comes from, and it is assumed (“What makes some information count as a belief is the role it plays”).
2. Parity (§2). This is stated with “(so we claim)” and is not derived.
3. Explanatory simplicity and a revisionary concept. The internalist explanation is “pointlessly complex”. The notion of belief “ought to be used” so that Otto qualifies, because that makes it “more akin to a natural kind”. This is an inference to the simpler description and a proposal about a concept. It does not show which relation holds.

The evidence the paper gives for coupling is causal: “active causal role”, “just as causally relevant as typical internal features of the brain”, and the removal counterfactual in §3. The move from “causally coupled in a two-way loop” to “is part of” is not argued separately. It is carried by the functionalist premise and by burden-shifting (“To provide substantial resistance, an opponent has to show that Otto’s and Inga’s cases differ in some important and relevant respect”, §4). My judgement is that the constitutive step is stipulated given functionalism, not argued. The paper does not consider the reply that coupling is only causal. That reply was pressed later by Adams and Aizawa, “The bounds of cognition” (2001; LIT-438, NOTE-368). The name “coupling–constitution fallacy” comes from their later work (2008), not from that paper. The paper also offers no general account of parthood. “Part” is used informally of processes, systems and selves. Neither mereological principles nor a criterion of composition appear (see Connections).

Claims

id	claim	strength	support
C1	The three Tetris cases are cognitively on a par, and only the skin/skull boundary separates (2) from (3).	informal argument	thought experiment in §2; burden-shifting (“the legitimacy of that boundary is precisely what is at issue”)
C2	Physical rotation in Tetris takes about 100 ms plus about 200 ms to select, against about 1000 ms mentally, and is used epistemically.	cited empirical evidence	Kirsh & Maglio 1994, reported in §2 and not reproduced
C3	Parity: whatever would count as part of a cognitive process if done in the head is part of the cognitive process when done in the world.	assertion	stated in §2 with the hedge “(so we claim)”; motivated only by “epistemic action … demands spread of epistemic credit”
C4	Active externalism escapes the explanatory-irrelevance objection that faces Putnam–Burge externalism.	informal argument	§3; §4 Twin Otto case (behaviour differs)
C5	A coupled organism-plus-resource system is a cognitive system, and its processes are cognitive processes.	informal argument	§3 coupling plus the removal counterfactual; C3; explanatory simplicity (Scrabble)
C6	Contingent or intermittent coupling does not defeat cognitive status; reliable coupling is enough.	informal argument	§3 plug-in module, calculator, Filofax; comparison with sleep and intoxication
C7	Language evolved or developed partly as a means of cognitive extension, and extension is “core, not an add-on”.	speculation	§3 (“It may be that…”, “Perhaps…”); fish analogy; no evidence offered
C8	Otto believes the museum is on 53rd Street before consulting the notebook.	informal argument	§4 case comparison plus role-functionalism plus explanatory simplicity
C9	Twin Otto believes 51st Street; “a belief is simply not in the head”.	informal argument	depends on C8
C10	The Otto–Inga differences (reliability, bandwidth, perceptual access, phenomenology) are all shallow.	informal argument	§4 replies with the Lucy and Terminator cases; the perceptual-access reply assumes the extended-system view it defends
C11	Constancy, direct availability, automatic endorsement and (arguably) past endorsement are the features that make extended belief clearly applicable.	informal argument	§5 read off the single paradigm case; explicitly not necessary and sufficient; degree and indeterminacy allowed
C12	The Internet is likely to fail the features unless the user is unusually reliant, facile and trusting.	assertion	one sentence, §5
C13	Mental states can be socially extended (partners, waiters, secretaries).	assertion, with an illustrative anecdote	§5 (“We see no reason why not, in principle”); Wooden footnote
C14	The self extends: Otto “is best regarded as an extended system”.	informal argument (one paragraph)	§5; premise that dispositional beliefs constitute “part of who I am”, plus a reductio on a self of occurrent states only
C15	Interfering with someone’s environment can in some cases carry the moral significance of interfering with their person.	assertion	§5, one sentence, hedged (“It may be… in some cases”)

Concepts

• active externalism — externalism based on “the active role of the environment in driving cognitive processes” (§1). The relevant external features are coupled to the organism, act in the here and now, and make a difference to behaviour (§3).
• passive externalism — the Putnam–Burge variety. The external features that determine content are distal and historical, and have no role in present cognitive processing (§3).
• coupled system — an organism linked to an external entity “in a two-way interaction”, in which all components play “an active causal role” and “jointly govern behavior”. Removing the external part makes competence drop (§3).
• epistemic action — Kirsh and Maglio’s term for altering the world “so as to aid and augment cognitive processes such as recognition and search”, as opposed to pragmatic action (§2).
• the Naked Mind — “a package of resources and operations we can always bring to bear on a cognitive task, regardless of the local environment”. It is the target of the portability objection (§3).
• cognitive vortices — external linguistic and symbolic structures that we create and exploit, as fish exploit vortices, and which “help constitute human thought” (§3).
• extended belief — a standing, dispositional belief whose information is carried by an external resource playing the role of memory (§4–5).
• extended self — the agent as “a coupling of biological organism and external resources” (§5).

Connections

• Mereology. LIT-124 (Varzi, SEP Mereology) lays out the ladder of mereological theories and the dispute over unrestricted composition. Under an unrestricted-sum mereology (GEM), the fusion brain+notebook exists trivially, so the extended-mind thesis cannot be a thesis about whether that fusion exists. It must be the claim that the fusion falls under the kind cognitive system or believer. Read that way, Clark and Chalmers in effect propose a restricted, kind-relative composition criterion: reliable two-way functional coupling. They offer it for one kind, not as a principle of composition in general, and they never frame it in mereological terms. That is why the record’s mereology tag is justified by the paper’s subject but not by its method.
• Body parts and BCIs. LIT-138 (Register, The Depth of the Body) separates a moral “body-part” relation from ordinary parthood by level of abstraction, and applies it to cloud-hosted BCI software. Two of its themes are ones the extended-mind paper only gestures at: the moral consequence in §5 (interfering with the environment as interfering with the person) and the cyberpunk implant of case (3), which the paper treats as uncontroversially internal. Register’s question of which level the part is taken at (software or server) is one Clark and Chalmers never raise for Otto’s notebook, which could be the object, the inscription or the information. LIT-147 (Pernu, BCIs and mental causation) is a related, deferred work on neuroprosthetics.
• Personal identity. LIT-136 (Floridi, The Informational Nature of Personal Identity) individuates the self informationally through corporeal, cognitive and conscious “membranes” and treats ICTs as technologies of the self. It is a close relative of the §5 claim that Otto is “an extended system”. Floridi gives a structured account of the self where Clark and Chalmers give one paragraph. Both are presented here from their notes. LIT-136 is itself unread.
• AI tools and minds. LIT-111 (Birch, AI consciousness) and LIT-212 (LLM hallucination) are the downstream seeds named in the skim. The automatic-endorsement feature is the paper’s own lever on unreliable tools. The paper insists that consciousness is not claimed to extend (§3–4). That limits how directly it bears on AI-consciousness questions such as those in LIT-056 (Butlin et al.) and LIT-111. LIT-207 (Roberts, chatbot talk and embodied cognition) works in the neighbouring 4E tradition. LIT-046 (Pilgrim et al., collective intelligence) is a distant relative of the §5 social extension.
• Within philosophy of mind, the paper sits against Putnam (1975) and Burge (1979), and alongside Hutchins (1995), Haugeland (1995), McClamrock (1995), Varela, Thompson & Rosch (1991) and Wilson (1994), which the paper cites in footnote 4 as “of a similar spirit”.

Bearing on the record

The paper carries no instruction for machine-learning practice, and nothing in it belongs in the Anthology of the SOTA. It is a philosophy-of-mind argument whose only possible link to ML is the one the skim raises: whether a user’s LLM assistant or memory store could meet the §5 features. The paper’s own verdict on the nearest 1998 analogue, the Internet, is that it “is likely to fail on multiple counts” for all but unusually reliant, facile and trusting users. Any record document that cites LIT-097 for conditions or criteria that are sufficient for extended belief is citing it for more than it says, because the paper offers the features as marks of a clear case and allows degree and indeterminacy. The LIT-097 summary’s “provided the resource is reliably available, easily accessed and automatically endorsed” should be reworded accordingly. The record holds no THEORY documents on this subject that I could find.

Limitations

• The constitutive claim rests on premises the paper does not argue for: role-functionalism, the parity sentence, and simplicity as a guide to what exists. The step from causal coupling to parthood is not defended against a merely causal reading (see the section on constitutive vs causal).
• The perceptual-access reply is admittedly circular in form. The paper concedes that “we are in effect advocating a point of view on which Otto’s internal processes and his notebook constitute a single cognitive system”, and argues from that standpoint.
• Part of the conclusion is a proposal, not a finding. The notion of belief “ought to be used” so that Otto qualifies (§4).
• The criteria come from one case. They were read off the Otto case, are not claimed to be necessary or sufficient, and the fourth is flagged “arguable”. Constancy and past endorsement bring in history, which the paper’s own footnote says sits uneasily with its “active, present” externalism.
• The extended-self and social-extension claims are brief. Each is argued in a paragraph or less, and the moral consequence is one hedged sentence.
• No mereology. “Part” is never given a sense, and the paper does not say whether the notebook, its inscriptions or the information in them is the part.
• Consciousness is excluded. The paper allows that experiences may be internally determined, so the thesis is narrower than its title.
• The authors’ confidence is asymmetric. The authors are “listed in order of degree of belief in the central thesis”.
• Access. Only the authors’ web version was read. Agreement with the Analysis typeset text is unverified, and the page range on the web page (10–23) conflicts with Crossref (7–19).

Open questions

• Does the published Analysis text differ from the consc.net version? That needs the typeset PDF, from JSTOR 3328150 or OUP, which could not be fetched on 2026-09-26.
• What stops the “cognitive bloat” that the §5 features only gesture at, for example a phone contacts list that meets constancy, availability and endorsement? A sharpened, non-historical version of the features, which the paper’s own footnote suggests, would settle whether the thesis overgenerates.
• Is coupling enough for parthood, or only for causal dependence? Settling it needs an independent account of constitution for cognitive kinds, such as a mechanistic-constitution criterion, applied to the Otto case. The paper supplies none.
• Would an LLM assistant satisfy automatic endorsement in a way that makes a user’s false beliefs extended beliefs, as with Twin Otto, when the tool hallucinates? The paper’s framework suggests yes, if the features hold. It gives no answer on how endorsement should respond to known unreliability.
• Does the extended self carry the moral weight §5 suggests? LIT-138’s depth criterion for body parts is one way to test the “interfering with their person” claim.

Corrections to the seeded skim

• The skim’s summary (repeated in the LIT-097 summary) says beliefs can be environmentally constituted “provided the resource is reliably available, easily accessed and automatically endorsed”. That makes the features sufficient conditions. The paper refuses that framing. It says “We do not think that there are categorical answers to all of these questions, and we will not give them”. It then lists four “features of our central case that make the notion so clearly applicable there” (§5), and says applicability “gradually falls off” as cases lack them, with intermediate cases possibly “indeterminate” or context-dependent. The features are also not worded “reliably available / easily accessed”. They are (1) the notebook “is a constant in Otto’s life”, meaning he “will rarely take action without consulting it”; (2) its information “is directly available without difficulty”; (3) “upon retrieving information from the notebook he automatically endorses it”; and (4) the information “has been consciously endorsed at some point in the past, and indeed is there as a consequence of this endorsement”. The paper calls (4) “arguable”. “Reliable” belongs to the §3 discussion of reliable coupling and to the §4 reply on access, not to the list.
• The skim quotes “The Internet ‘is likely to fail on multiple counts’” and drops the qualification that follows in the same sentence: “unless I am unusually computer-reliant, facile with the technology, and trusting”. The same sentence adds that “information in certain files on my computer may qualify”. For the AI-tool question the skim raises, the omitted clause is the important part: the paper expects the Internet to fail for typical users, not in principle.
• The skim says language is “a ‘cognitive vortex’, on the analogy of a tuna”. The text says “cognitive vortices”, in the plural: “Words and external symbols are thus paramount among the cognitive vortices which help constitute human thought” (end of §3). It speaks of “the fish” and names no species. The species belongs to the cited Triantafyllou & Triantafyllou (1995) Scientific American article, and I have not checked it here.
• The skim and dossier call the §2 principle “parity-style”. That description is fair, but the paper never names a “parity principle”. The word “parity” occurs once, in the Simon footnote in §3 (“treating internal and external processing with the parity they deserve”). The name “Parity Principle” comes from later literature and should not be quoted as the paper’s own term.
• The dossier has an “Abstract, paraphrased” section. The authors’ version has no abstract. Whether the Analysis print had one is unverified, and short Analysis pieces of that period usually did not.
• Pagination: the web page says “Published in Analysis 58:10-23, 1998. Reprinted in (P. Grim, ed) The Philosopher’s Annual, vol XXI, 1998.” Crossref gives 7–19. The skim was right to prefer 7–19. The web page’s page numbers look wrong, and “10-23” should not be cited.
• The skim does not mention the first author footnote: “Authors are listed in order of degree of belief in the central thesis”. The paper states outright that the two authors do not hold the thesis with equal confidence. This matters when attributing the view to “Chalmers”.