NOTE-550: List & Pettit — Group Agency and Supervenience

Contribution

It turns “can a group be an agent over and above its members?” into a question about the form of a supervenience relation, and answers it with judgment-aggregation theorems. Individualism is kept: fix what the members judge and do, and the constitution, and the group’s judgments are fixed. But the dependence must be holistic. A rational group agent’s judgment on one proposition generally depends on members’ judgments on other propositions, and can depend on nobody’s judgment on that proposition.

Key insight

Majority voting on every question, applied to rational individuals, can give an irrational group (the discursive dilemma). The paper’s point is that this is not a defect of majority voting. Any rule that settles each question by looking only at what people think about that question fails in the same way, given enough interconnected questions. So a group that reliably reasons must route its judgments through some structure: premises first, then conclusions by logic, perhaps with different members settling different premises. That structure is what makes the group an agent, and it is also what makes the group’s view able to differ from what its members, one by one, think.

Assumptions

• Four conditions of agency (§2, p. 3). Representational and goal-seeking states; theoretical rationality in forming them (completeness, consistency, deductive closure); action on them as practical rationality requires; and robustness across a class of relevant possible conditions. Stated as near-consensus, not defended. Function, not structure, matters, apart from two stipulations: no control from outside, and no “pre-emptive rigging” (p. 4, citing Peacocke and Block).
• Judgments, not degrees of belief (pp. 4–5). Judgments are categorical and cover fact and value. Rationality constraints on plans are said to track those on value judgments, so little generality is lost.
• Supervenience (p. 5). B-facts supervene on A-facts iff fixing the A-facts fixes the B-facts. Group judgments supervene on individual contributions once a constitution, a set of rules for forming group judgments from members’ inputs, is in place. The constitution is itself treated as members’ shared dispositions, so it adds nothing non-individual (p. 6).
• Agenda (p. 7). The court example: P, Q, R and “R iff (P and Q)”, closed under negation. The results are stated for this agenda and said to hold for wider classes (n. 3: Proposition 1 for agendas with a minimal inconsistent subset of three or more, 2 for “minimally connected”, 3 for “strongly connected” agendas).
• Universal domain (p. 8). The constitution accepts every profile of complete, consistent, deductively closed individual judgments.

Key results

• Robust group rationality (p. 9). For every admissible profile, the group’s judgments are complete, consistent and deductively closed.
• Proposition 1 (p. 9). Robust group rationality is inconsistent with majoritarian supervenience (the group judgment on each proposition is the majority judgment on it). Table 2: individuals (T,T,T,T), (T,F,F,T), (F,T,F,T) on (P, Q, R, R↔(P∧Q)); the majority accepts P, Q and the biconditional and rejects R.
• Proposition 2 (p. 11). Inconsistent with uniform proposition-wise supervenience: the group judgment on each proposition is the same function, for all propositions, of the individual judgments on it, and depends on more than one individual. A strengthening of List & Pettit (2002), proved by Pauly & van Hees (2003).
• Proposition 3 (p. 12). Inconsistent with proposition-wise supervenience: the function may differ by proposition, provided it depends on more than one individual and respects unanimity. Proved by Dietrich & List (2005).
• Proposition 4 (p. 13). The premise-based procedure (majority vote on the premises, conclusion derived) satisfies robust group rationality and set-wise supervenience (the group’s set of judgments is a function of the members’ sets), but not proposition-wise supervenience. Tables 2 and 3 have identical R columns and give different group judgments on R. Members’ judgments on R are insufficient to fix the group’s (“weak discontinuity”) and unnecessary (“strong discontinuity”, p. 14, after Pettit 2003).
• Sequential priority procedure (p. 15). The generalisation to any agenda: propositions in a priority order, majority vote unless logically constrained by earlier accepted ones. Proposition 4 carries over.
• Homogeneous and heterogeneous bases (p. 15). A base is homogeneous if the group judgments are invariant under permutations of the members’ judgment sets.
• Proposition 5 (p. 16). The distributed premise-based procedure, where each premise is decided by majority in its own subgroup, satisfies robust group rationality and set-wise supervenience, with a heterogeneous base. Each member need contribute a single judgment on a single premise, and no member’s judgment on the conclusion is needed (p. 17). It models legislatures with subcommittees.
• Table 4 (p. 17). Proposition-wise relations, majoritarian or other, uniform or not, are inconsistent with robust group rationality; set-wise relations, homogeneous or heterogeneous, are consistent with it.
• Escape routes (n. 9). Proposition-wise supervenience can be kept by restricting the domain, relaxing completeness or closure, a dictator, a constant rule, or shrinking the agenda.
• Emergent agency (§6, pp. 19–21). The examples assume members who knowingly sustain the group through joint intentions (Tuomela, Bratman, Gilbert). Could humans form a group agent as a swarm does, with no member conceiving of the group? The authors doubt it: no constitution could guarantee robust group rationality whatever members’ personal reasons are, unless there is central direction and deference, which is itself a joint intention (Block’s China-body case). Invisible-hand processes yield results stable across contexts, but agency needs outputs tailored to circumstances, and “no existing theory makes sense of how this could happen” (p. 20).

Claims

id	claim	strength	support
C1	A system is an agent if it forms representational and goal-seeking states, is theoretically and practically rational about them, and does so robustly	moderate	stated as consensus, §2; not defended
C2	No proposition-wise aggregation rule (respecting unanimity, non-dictatorial) yields robustly rational group judgments on interconnected agendas	strong	Propositions 1–3; 2 and 3 cited, not proved here
C3	Set-wise supervenience is consistent with robust group rationality	strong	Propositions 4–5, constructive
C4	A group agent’s judgments can be discontinuous with its members’ judgments on the same proposition, while remaining individualistically grounded	strong (as a consequence of C2–C3)	§5 tables; the “weak” and “strong” discontinuity
C5	Group agency is possible only where members knowingly sustain it, or defer to a director	weak	§6, an argument from the absence of any theory; offered with stated scepticism, not proof

Concepts

• Constitution — the formal or informal rules by which members’ inputs determine group judgments (p. 5); a function from admissible profiles to group judgments (n. 4).
• Robust group rationality — complete, consistent and deductively closed group judgments for all admissible profiles (p. 9).
• Proposition-wise supervenience — the group judgment on each proposition is a function of members’ judgments on that proposition alone (p. 11).
• Set-wise supervenience — the group’s set of judgments is a function of members’ sets of judgments (p. 12).
• Weak / strong discontinuity — members’ judgments on a proposition do not suffice / are not even needed to fix the group’s judgment on it (p. 14).
• Homogeneous supervenience base — the group’s judgments are invariant under permuting the members’ judgment sets (p. 15).

Connections

• The book (LIT-391), unread. The paper’s §2 agency conditions and §§3–5 supervenience results are what the book’s Part I chapter titles (“The Conditions of Agency”, “The Aggregation of Intentional Attitudes”, “The Structure of Group Agents”) name. LIT-391’s “what a reading should check” asks whether group attitudes supervene “holistically or proposition by proposition”. On this paper’s evidence the answer is holistically, but the book may say more.
• List 2014 (LIT-700) extends this paper; the relation is declared on that LIT. It recasts proposition-wise and set-wise supervenience as the mark separating aggregate from corporate attitudes.
• List 2016 (LIT-401). That paper’s thin functionalist definition of an agent is the descendant of §2 here.
• Huebner (LIT-702). The distributed premise-based procedure is a formal miniature of Huebner’s distributed collective representation, and “strong discontinuity” is the property Huebner needs for a genuinely collective emotion: the group’s state on an object need not be any member’s state on it. Neither paper cites the other.
• Wimsatt (LIT-707), unread. Homogeneity is an invariance-under- rearrangement condition, one of Wimsatt’s four conditions for aggregativity as he summarises them. Proposition-wise supervenience fails; set-wise survives. By Wimsatt’s definition that makes group judgment non-aggregative, hence emergent, and compatible with reduction, which is List and Pettit’s individualism. The strong emergence §6 rejects is a different notion. The comparison is the record’s.
• Searle (LIT-482). The paper’s constitution is treated as members’ dispositions, not as collectively accepted status functions; Searle is not cited.

Bearing on the record

• THEORY-043 (composed and decomposed minds) says that what separates the two is an anti-nesting principle or an architectural criterion, “neither yet principled”. For agency, though not for experience, this paper supplies a principled architectural criterion. A collective of minded members can hold robustly rational attitudes only if its organisation routes judgments holistically. It says nothing about consciousness, so it does not settle the THEORY.
• Candidate THEORY (proposed, not filed): a group attitude is emergent in Wimsatt’s non-aggregative sense exactly when it fails proposition-wise supervenience on members’ attitudes, so List and Pettit’s impossibility results prove that every robustly rational group agent has emergent, reductively explicable attitudes. Sources: this paper, LIT-700, LIT-707. Promote when Wimsatt’s 1997 conditions are read and checked against an aggregation function.
• No instruction for machine-learning practice.

Limitations

• Only one necessary condition of agency, theoretical rationality of judgments, is studied. Action, desire and robustness are not modelled.
• The results are for categorical judgments under universal domain. Restricting the domain, as real groups with shared priors might, reopens proposition-wise rules (n. 9).
• Two of the three impossibility results are cited, not proved, and the agenda conditions under which they generalise are only named (n. 3).
• §6’s scepticism about emergent agency is an argument from the absence of a theory. The paper says so.

Open questions

• Do premise-based procedures describe how real courts, boards and campaigns form their positions, or are they idealisations?
• Is there a graded version: how far can a group stray from proposition-wise supervenience and still count as reflecting its members?
• Does the distributed procedure extend from judgments to emotions, so that a group’s fear could be fixed by members’ appraisals of different aspects of a threat, none of whom fears it?