THEORY-043: A mind of mindless agents and a mind of minded agents are one functionalist move run in two directions, and what separates them is an anti-nesting principle or an architectural criterion, neither yet principled

Source

• Schwitzgebel (2014), LIT-159, read in NOTE-131: the composition argument.
• Minsky (1977), LIT-399, read in NOTE-344: the earliest reading here of the decomposition, with K-lines (LIT-402) and Singh’s exposition (LIT-444). The Society of Mind itself (LIT-429) is unread, so this account does not lean on it.
• Block (1978), LIT-407, read in NOTE-366: the China-brain and the first critique of anti-nesting.
• List (2016), LIT-401, read in NOTE-340: group agency granted, group experience denied.
• Kammerer (2015), LIT-409, read in NOTE-362: the functional anti-nesting principle.
• Oizumi, Albantakis & Tononi (2014), LIT-445, read in NOTE-365: the exclusion postulate in formal statement.
• Levin (2019), LIT-439, read in NOTE-341: nested goal-pursuing collectives, one of the three readings that grant agency under composition. His account is stated in THEORY-583.
• Schwitzgebel (2016), LIT-459, read in NOTE-381: the reply to Kammerer, for the condition-A gap, the Leo case and the simplicity dilemma.

The two directions

Downward. Minsky’s 1977 memo, LIT-399, takes as background (its C1, asserted rather than argued) that a mind is a society of simple agents whose interactions produce intellect and affect alike. The move is older than Minsky. Selfridge’s Pandemonium (LIT-427) assembles a competence the whole has, recognising a pattern, from demons none of which has it. Brooks (LIT-436) builds and runs it: a robot with no central representation, control or goals, whose coherence, he says, is imputed by an observer (his C4). He cites Minsky as giving “a similar account”. The parts are never minded, so nothing nests.

Upward. Schwitzgebel, LIT-159, runs the same functionalist reasoning from the other end. If what makes a brain conscious is its organisation, then contiguity and the mindedness of parts are not disqualifying (NOTE-131, C2–C3), and the United States has the materialist marks in some degree (C7). Here the parts are minds, so the whole nests minds inside a candidate mind.

The functionalist premise is the same both ways: the whole’s mental properties come from how parts are organised, not from what the parts are. Hofstadter’s “Ant Fugue” in GEB (LIT-437, unread here) is the meeting point, a colony whose ants are mindless and whose organisation is a mind.

What the readings agree on: agency survives both directions

• Block grants that a homunculi-headed system could have beliefs and other non-qualitative states (NOTE-366, C6). His doubt is about qualia only.
• List grants that organised groups are agents in a thin functionalist sense and are aware in the access sense (NOTE-340, C1–C2). He says the question of group experience then depends wholly on which bridge principle is true (C3).
• Levin (LIT-439, NOTE-341) finds goal-pursuing collectives (his “Selves”) nested at every biological scale (C2), and proposes that human groups may have their own cognition (C6). Both claims are about agency; his extension to experience is labelled speculation (C8).

So the decomposition and the composition are not in dispute as accounts of agency. The dispute is over experience, and it is there that someone who accepts Minsky’s downward move must find a reason to refuse Schwitzgebel’s upward one.

The two ways to refuse the upward move

1. Anti-nesting: a conscious whole cannot have conscious parts.

• Putnam (LIT-396) introduces it by stipulation, with one stated motive: to keep a swarm of bees from counting as a single pain-feeler. Block shows it ad hoc. Each of three readings he tries fails on a case, including people the size of elementary particles who take over a body’s microphysics (NOTE-366, C4).
• IIT states it formally. Of all overlapping sets of elements, only the local maximum of Φ is conscious (NOTE-365, C2). That is a postulate, derived from an axiom that experience has definite borders, with no argument specific to the system level. Schwitzgebel’s reductio, that a single ballot could switch a voter’s consciousness off (NOTE-131, C5), is not answered in the paper read.
• Kammerer moves the line from “are the parts conscious?” to “does the whole depend on parts consciously representing the whole, in a role nothing else could fill?” That keeps Minsky’s agents, workspace modules and Block’s tiny people on the permitted side (NOTE-362, C1). But the verdict that excludes the United States rests on an asserted counterfactual about what it would lose without its citizens’ representations (C2), and the principle’s own justification is weak by its author’s account (C6). Schwitzgebel’s reply (LIT-459) opens an independent gap at condition A: it is unclear that citizens consciously represent the entity whose consciousness is in question at all, rather than one of several vague candidates for “the United States” (NOTE-381, C2). It also argues that any rule of simplicity strong enough to justify the principle either becomes the causal-exclusion problem, and so denies consciousness to brains as well, or is written for this case alone (C6).

2. Architecture: the whole is organised the wrong way.

• Block’s intuition is not about nesting at all. His doubt survives replacing every homunculus with a mindless gate (NOTE-366, C5), and rests on the system being unlike us and built to mimic us (C3), which he calls “not an overwhelmingly powerful argument”.
• IIT gives an architectural verdict too: a feed-forward system can match a conscious one’s behaviour without being conscious (NOTE-365, C7). Modular networks split into small complexes (C5).
• List argues that a typical group agent has low Φ because it is made of self-contained units (NOTE-340, C5). That is asserted, not computed.

The architectural route has a cost the anti-nesting route does not. A criterion that marks a group down for being built from loosely coupled, self-contained units marks a society of mind down for the same reason. Minsky’s agents communicate mainly without shared symbols, through K-lines and learned codes (NOTE-339, C3), and Selfridge’s design argument is that quasi-independent modules are easier to modify (NOTE-360, C2). IIT’s modular-network result (NOTE-365, C5) is the measured version of the worry. The criterion would have to separate a modular mind from a nation by degree, and nobody here computes the degree for either. Schwitzgebel’s own claim is that no architectural criterion excludes the United States without excluding admissible aliens (NOTE-131, C8); this account adds that it would also have to spare a mind built as Minsky describes.

What the Chalmers and Dennett readings add

Read after this account was drafted, they leave its verdict standing and sharpen it in five places.

• Why agency is cheap to agree on. Dennett’s intentional stance is blind to composition and to consciousness by design (LIT-440, NOTE-350): a system has beliefs if treating it as a believer pays. On the criterion of “True Believers” (LIT-398, NOTE-353), a group whose group-level stance pays would count, though Dennett never applies it to one. That is why every reading here grants agency in both directions.

• Why the downward move was licensed and the upward one was not. Dennett’s discharge of homunculi (“Artificial Intelligence as Philosophy and as Psychology”, LIT-393) is harmless only when the parts are stupider than the whole and are replaced step by step by machines. That work is filed unread and known through Stich’s summary, which Dennett accepted. Minsky’s and Selfridge’s agents are on the licensed side. Schwitzgebel’s citizens, who are as smart as the whole or smarter, are exactly what discharge never licensed. That is a reason the two moves are not symmetric for intelligence. Block’s n. 19 shows it settles nothing about experience.

• Organisational invariance narrows the dispute to architecture. Chalmers’s fading-qualia argument (LIT-392, NOTE-347), if sound, makes any fine-grained functional isomorph of a conscious being conscious, including a population of people implementing a person’s neurons. That closes Putnam’s stipulation for isomorphs and conflicts with IIT, which lets a functional equivalent lack experience. It mostly leaves Kammerer’s principle standing: the United States is not an isomorph of anything, so invariance does not reach it. The exception is Schwitzgebel’s Leo case (NOTE-381, C3). Read loosely, with “required” meaning that a member’s conscious representation is what moves them, the principle switches a group’s experience on and off with one member’s motives while group-level function is held fixed, which is the shape of case invariance arguments are built against. On the strict reading the case does not arise. What remains is the architectural route, which invariance cannot decide.

• Functionalism itself forbids no nesting. Chalmers’s implementation conditions (LIT-404, NOTE-342) do not care whether a part is a neuron or a person, and they let one system host two minds. The combination-problem paper (LIT-395, NOTE-358) reaches this account’s two routes independently, from panpsychism. Its candidate for what bounds a subject is integration, which it finds unprincipled. It adds an IIT-independent objection to exclusion: exclusion makes consciousness extrinsic. The meta-problem paper (LIT-425, NOTE-354) adds another: a zero-Φ isomorph reports alike.

• Two new candidates for a principled line, neither yet enough.

  • Dennett & Kinsbourne’s argument against a conscious central part of a brain (LIT-431, NOTE-367) has Kammerer’s shape and is motivated independently of groups, but does not reach the United States.
  • Chalmers’s 2019 condition, that consciousness needs access “not mediated by other mental states” (LIT-415, NOTE-356), would plausibly exclude a nation and spare a society of mind. That application is the reader’s, not Chalmers’s, and its motivation has the weakness the anti-nesting route already has.

  Both are the right kind of thing for promote_when. Neither has been applied by one procedure to a group model and a modular-mind model.

Two positions that refuse the framing. Searle (LIT-400, NOTE-349) rejects both directions at once: it is causal powers, not organisation, that matter. His key step against the systems reply (“the system is just a part of him”) is asserted. Dennett’s “Quining Qualia” (LIT-433, NOTE-357) denies there is a further property to dispute, so on his view the dispute never narrows to experience. Neither supplies the criterion this account asks for; each denies that one is needed.

What this does not say

• It does not say the United States is conscious. Schwitzgebel’s own conclusion is conditional and “gappy” (NOTE-131, C1 and C11), and nothing here strengthens it.
• It does not say Minsky’s society of mind is conscious. The account draws no conclusion about it either way. The 1977 memo speaks of intellect and affect as products of the society (NOTE-344, C1), and The Emotion Machine (LIT-432) treats feelings as reflective self-descriptions. Whether experience comes with them is the question this account leaves open.
• It does not say no principled line exists. Only that none of the readings supplies one. Kammerer’s is the best candidate on offer, and its weakness is its justification, not its shape.
• It does not rest on the unread works. Lycan (LIT-419), Huebner (LIT-426) and List & Pettit (LIT-391) are filed unread. Each is reported to bear on this seam, and Huebner’s test for a collective mind (integrated specialised subroutines) is the one most likely to change it. Tollefsen’s Groups as Agents (LIT-483), also unread, may take an interpretivist line on group agency, which would put it beside Dennett’s intentional stance; that is a guess from a chapter title.

Connections

• Corrections to NOTE-131 found in these readings. Each was found by a later reading, and all three are now applied in NOTE-131 v3.
  • Block’s counterexample to Putnam is elementary-particle people, not tiny people as neurons (NOTE-366).
  • Schwitzgebel’s “quotation” of Putnam is a paraphrase, and the principle’s first appearance is 1967, from a 1965 colloquium (NOTE-335).
  • The “Brooks (1986)” in Schwitzgebel’s group-mind tradition is D. H. M. Brooks, “Group minds”, not Rodney Brooks (NOTE-351).
• Schwitzgebel’s reply to Kammerer (LIT-459, read in NOTE-381) is now read, and its two arguments that bear here are in the Kammerer and invariance bullets above. Its reading also corrects NOTE-362, whose Limitations had the reply pressing condition B; it presses condition A, and NOTE-362 v2 says so. D. H. M. Brooks is the next reading on this seam.
• Which group a criterion is applied to has to be said. Two readings from social ontology bear on this, though neither discusses minds or experience; the application is the record’s.
  • Ritchie (LIT-479, NOTE-390) holds that groups with the same members can be distinct and coincide, and (LIT-486, NOTE-389) that only organised groups, not feature groups such as classes, are concrete structured wholes. “The United States” therefore names several candidates, which is a structural version of the condition-A gap Schwitzgebel’s reply opens, and only organised groups are candidates for agency or mind at all.
  • Epstein’s survey (LIT-487, NOTE-393) reports challenges to “local dependence”: a group’s facts may be grounded partly in non-members, infrastructure and membership rules. If so, an architectural criterion computed over the members alone is computed over too small a system. This adds a constraint on any criterion that would meet promote_when; it does not change what would.
  • Both readings are now stated as accounts: Ritchie’s in THEORY-549 (coincident groups and groups as realized structures), and the survey’s grounding/anchoring split in THEORY-487. Neither is a premise here.
• The agreement on agency is about thin agency. That agency survives composition holds for thin, functionalist agency, and two accounts bear on it from opposite sides: organisational minimal agency (THEORY-071) makes agency dear rather than cheap, since a whole counts only if it individuates itself, modulates its own coupling and regulates it by norms its own organisation sets, which Minsky’s and Brooks’s designed agents do not and a group is not yet shown to, while Korsgaard (THEORY-065) grants collectives agency in the same sense as individuals, but because a group can achieve the unity of agency by conformity to its constitutive norms, a reason that does not turn on architecture and says nothing about experience.
• Levin’s account of nested agents (THEORY-583) states the composition-side view of agency this account cites from LIT-439. It is the cheap end of the scale THEORY-071 makes dear: on it any homeostat is a Self.
• The cognition readings, now stated as accounts. Chalmers’s directness criterion is part of THEORY-512, the sensorimotor statement of the extended-mind thesis, which denies extended consciousness through perception. Dennett’s concession that the intentional system is an idealisation is THEORY-596. Block and Fodor’s case that the organisation that makes a mind cannot be a machine table is THEORY-471. None is a premise here.
• The readings stated as accounts. Putnam’s hypothesis with its condition (3) is THEORY-436, resting on the excerpt in NOTE-335. Dennett’s intentional strategy is THEORY-538, Searle’s causal-powers argument THEORY-638, and Chalmers’s implementation conditions THEORY-609. None is a premise here; they state on their own what this account cites them for.
• The Chalmers and Dennett readings as accounts. Organisational invariance is THEORY-507, the combination problem THEORY-644, the meta-problem challenge THEORY-577, “Quining Qualia” THEORY-415 and Frankish’s illusionism THEORY-633. None is a premise here; each is cited above from its reading.
• Neighbours: collective intelligence (LIT-046), “Real Patterns” (LIT-220), IIT’s later statement (LIT-185) and Schwitzgebel’s book version (LIT-216).