Group Agency is a pathbreaking achievement. Earlier writers in the literature on joint intentionality often dismissed with little argument the realist view that groups can be unitary agents in a fully literal sense. Christian List and Philip Pettit (L&P) put realism back on the map as a serious, unmysterious view. By drawing attention to a richer range of phenomena than earlier writers on joint intentionality often addressed, they make it plausible that individuals can compose enduring collectives with all the functional hallmarks of unitary agency. Given appealing functionalist background views, they convincingly argue that such collectives can literally be agents. And with the help of their landmark formal work on judgment aggregation, they challenge the residual inclination to see group agents as redundant beings that are easily reduced to sums or more intricate patterns in the individualistic mosaic.
And this is only to scratch the surface of the first three chapters. They are my target. Notably, I partly agree with L&P. Their revival of realism about group agents is, I think, successful: groups sometimes literally are agents. I merely remain unconvinced of their ‘non-redundant’ realist claims that (i) facts about group agents are not readily reducible to patterns in the individualistic mosaic, and that (ii) the dependence of group agents on individuals is holistic to the point of being inscrutable. L&P fail, I will argue, to establish these claims. This is because they do not consider the most promising options for those who deny (i) and (ii), and see group agents as real but redundant. Indeed, I will argue that a better explanation of the data on which L&P rely in defending (i) and (ii) is given by a sophisticated ‘redundant realism’ that draws on ideas from earlier work on joint attitudes not addressed in the book.
With these aims in mind, here is how I will proceed. In §1, I place L&P's view in a spectrum of views, rehearse their core argument for it, and explain in outline my objection. Focusing in §2 on the hard case of group agents' beliefs, I suggest a promising style of redundant realism ignored by L&P with the help of ideas from earlier writers on joint belief. I show how it can explain the key data L&P use to defend non-redundant realism, and how this blocks their argument for (i). In §3, I turn to some negative upshots for (ii), and undermine L&P's other reasons for (ii). I turn to some take-home messages in §4.
1. SOME PRELIMINARIES AND THE CASE FOR NON-REDUNDANT REALISM
L&P fix the target of their theory by separating collections like the people now on 59th Street from groups like nations, and both from group agents. Groups have an ‘an identity that can survive changes in membership’, whereas collections do not (Reference List and Pettit2011: 3). If they exist, group agents are unique in displaying the core marks of agency. For L&P, there are three. Agents have
(a) representational states (e.g., beliefs) that depict how things are in the world,
(b) motivational states (e.g., intentions) that aim to alter the world, and
(c) the capacity to alter the world on the basis of states of types (a) and (b). (20)
Because they have representational and motivational states, group agents can be held up to requirements of rationality. In L&P's current terminology (24), these include
• Attitude-to-Attitude requirements of coherence,
• Attitude-to-Action norms that ban (e.g.) weak-willed and akratic action,
• ‘Attitude-to-Fact’ norms that demand fit between doxastic attitudes and evidence.Footnote 1
L&P see these as constitutive principles (in the limit): systematic nonconformity with them in favorable conditions (and within feasible limits) simply precludes agency. Since collections and mere groups need not conform to such norms at all, let alone in point of constitutive fact, there is a big difference.
So understood, there are several views about group agents. Eliminativists deny that there are group agents. They may tolerate superficial talk, but stress that it is shorthand for summative talk about what most group members do or think. On the other hand, redundant realists (L&P's term) think that group agents exist, but that facts about them readily reduce to individualistic ones – just not merely summatively. As I explain in §2, redundant realists can profit from writers on joint intentionality like Tuomela, Gilbert, Bratman and Searle. All four reject mere summative views. With the possible exception of Gilbert, they also deny that there literally are group minds.Footnote 2 But one can exploit some of their ideas and be a redundant realist instead. One can agree with L&P that groups can be real agents: just redundant ones, since talk of them translates readily, albeit not in merely summative terms. The tack is familiar: a reduction of X need not be an elimination of X if it preserves or explains X's core features.
L&P's non-redundant realism sits between redundant realism and an emergentism on which group agents are animating forces literally beyond individuals. L&P agree with redundant realists in thinking that fundamental social ontology is individualistic and embrace a supervenience thesis. But they think the supervenience pattern is often deeply opaque. As they say:
(I: Strong Nonredundancy) ‘[t]hese agents are related to their members in such a complex way that talk of them is not readily reducible to talk about their members’ (5)
and
(II: Opacity) ‘knowledge [of group agents] is unavailable in practice – even the most idealised practice – on the basis of observing individual agents alone’ (76)
(I) and (II) are the core of L&P's non-redundant realism. L&P sometimes vacillate between (I) and the weaker claim that it is useful and important to conceptualise group agents as autonomous beings. But this is an uncontroversial claim with which redundant realists can agree. Even eliminativists about some phenomena embrace similar claims. Compare early instrumentalists in the philosophy of science. So I will set aside this weaker claim.
Now, I think L&P convincingly dismiss emergentism and eliminativism. They rightly argue that groups can display all the marks of agency. They show that one can take this at face value without going emergentist. And they illustrate their contention in richer ways than has been standard in the literature on joint intentionality. Since it is familiar from that literature that mere summative views fail, it is plausible that we ought to be realists.
But why prefer non-redundant realism? L&P's answer flows from their pioneering work on judgment aggregation.Footnote 3 To bring it out, consider why systematic majoritarian aggregation rules fail – i.e. rules that have as output a doxastic attitude toward P iff a majority has that very attitude toward P. These rules yield collective inconsistency. Taking a schematic example, consider the profile in Table 1, where Yes/No express personal belief/disbelief.
Profile
While each individual's attitudes are consistent, a systematic majority rule inconsistently suggests group belief in P and P→Q but disbelief in Q. L&P's Impossibility Theorem extends this point, telling us that no aggregation rule meets these conditions:
Collective Rationality: The rule yields as output consistent and complete group attitudes to propositions on the agenda.
Universal Domain: It admits as input any profile of consistent and complete individual attitudes to the propositions on the agenda.
Anonymity: All individuals' attitudes are given equal weight in determining group attitudes.
Systematicity: The group attitude on each proposition depends only on individuals' attitudes on it, and the pattern of dependence is the same for all propositions.
L&P say that rejecting Systematicity – particularly its first conjunct – is the best general escape route for groups that aspire to genuine agency (58). And they suggest that the best general way to do this is with premise-based aggregation rules. An example: take majority verdicts on atomic propositions in profiles like that in Table 1, and decide group verdicts on molecular propositions via truth-tables for the connectives. This eliminates collective inconsistency. But as L&P stress, it yields an apparent autonomy of the group's attitude from the members'. If this rule is applied to Table 1, the result is Yes for Q. Yet the majority disbelieves Q! This points to why L&P accept Strong Nonredundancy, at least for group beliefs. While the group belief here depends on individualistic facts, it also looks free-floating and hard to translate.
This takes us to L&P's case for non-redundant realism (focusing on group agents' beliefs). L&P clearly show that the following often fails for groups consistent enough to be agents:
Naïve Proposition-Wise Dependence for Belief (PWD-B). For any proposition P, whether the group agent believes P depends just on patterns of personal belief in P among members.
PWD-B is often deeply false. If a group uses premise-based rules and displays patterns of individual beliefs like in Table 1, its beliefs in some conclusions will be underpinned by no majority of personal beliefs in these conclusions. That seems to make them autonomous. Having seen this, L&P's core argument can be put as follows (with focus on group belief):
(a) PWD-B is false: a group agent's belief in any P cannot in general depend on the patterns of personal belief among members in P.
(b) If (A), nonredundant realism holds of group agents' beliefs if any realist view does.
(c) So, nonredundant realism is true, given that realism is plausible on other grounds.
But I doubt (B). Let me explain in outline how we can reject (B) consistently with L&P's approval of premise-based rules. I will develop these thoughts at more length in §§2–3.
In cases like the one represented in Table 1, it is true that collective rationality demands the group agent to believe conclusions few members personally believe. Premise-based rules are needed, and mandate group beliefs in conclusions that are unmatched by a majority of personal beliefs. But it is compatible with this that there must be majority attitudes to these conclusions that are potentially distinct from personal beliefs. Indeed, this is plausible. Premise-based rules must be implemented with respect to given conclusions like Q. So members must make it the case that the group believes Q on the basis of distinct majority beliefs in P and P→Q, even if each sticks to personal disbelief in Q. So what must they do? Their merely antecedently agreeing on a rule that will end up recommending Q is insufficient. Even given such agreement, no member may now be willing to use Q as a working assumption for the purposes of joint action, or to treat Q as true in functioning qua member. If so, it is very implausible that the group believes Q.
As I argue further in §2, this does not imply that most members must personally believe Q after all. Treating Q as true for some purposes (e.g. joint pursuit of some goals) does not require personal belief in Q. The member-level attitude here is often called ‘acceptance’. Many think it can diverge from personal belief. Yet if that is right, the following is left untouched:
Interesting Proposition-Wise Constitution: While a group belief in any given proposition P may often be caused and rationalised by majority beliefs in distinct propositions, it will always be constituted by certain patterns of member attitudes to P potentially distinct from personal beliefs.
Moreover, this is compatible with everything L&P actually succeed in showing. To see this, it helps to note an ambiguity in L&P's talk of cases where premise-based rules are used, and in Systematicity itself – an ambiguity easily concealed by less fine-grained talk of supervenience. L&P say that group agents' beliefs in some conclusions ‘depend on’ or are ‘determined by’ majority beliefs in distinct propositions. But they could mean two things:
Non-Proposition-Wise Causal/Rational Dependence: Group beliefs in many conclusions must causally and rationally depend on majority attitudes to distinct propositions.
Non-Proposition-Wise Constitutive Dependence: Group beliefs in many conclusions constitutively depend on majority attitudes to distinct propositions, by being (partially) constituted by them.
If L&P intend the first, they are right. Indeed, the following reading of Systematicity fails:
Causal/Rational Dependence Systematicity. Group attitudes to each proposition P causally and rationally depend only on patterns of member attitudes toward P, where the pattern of causal/rational dependence is the same for all propositions.
But the status of non-redundant realism about group belief turns on claims about the constitution of group beliefs, not their causal/rational ancestry. So the fact that this reading fails is irrelevant. Yet Non-Proposition-Wise Constitutive Dependence also fails. Group beliefs in conclusions cannot often consist in majority beliefs in distinct premises. If G believes Q because of distinct majority beliefs in P and P→Q, G's belief in Q will not consist in these.
To support non-redundant realism, L&P must say that group beliefs in some conclusions cannot consist in a neat pattern of member attitudes to those very conclusions. But only the following data about group belief are obvious from L&P's discussion of Table 1-like cases:
a. Group beliefs in some conclusions C cannot consist in majority personal beliefs in C.
b. For some conclusions C, group beliefs in C must be causally and rationally dependent on majority beliefs in certain distinct propositions.
Yet (a) and (b) do not rule out Interesting Proposition-Wise Constitution. Indeed, it is compatible with the falsity of the version of Systematicity that L&P actually refute. As I will now argue, we can and should believe Interesting Proposition-Wise Constitution. We need it to make sense of how groups believe conclusions personally believed by few members. Yet if it is true, redundant realism about group beliefs is true. So L&P's central case fails.
2. THE CONSTITUTIVE IMPORTANCE OF ACCEPTANCE QUA MEMBER
2.1. How Interesting Proposition-Wise Constitution could be true
Earlier theorists who rejected mere summative accounts of group belief already recognised that groups can believe things when few or no members personally believe them. We can make this vivid by considering a point from Gilbert's ‘Modeling Collective Belief’. She noted that there is a significant difference between voicing a certain view as a group member and personally believing it. People who participate in genuinely joint belief will, as she said, often
view themselves as speaking ‘in their capacity as group members,’ ‘as a member of this body,’ and so on. ‘I'm afraid that you did not meet our needs’, says the department chairman to one of the unlucky candidates. Conscious of his role as a representative of the department, he speaks as such. He may personally think that this candidate was the best. Or he may have no personal opinion on the matter.Footnote 4
As Gilbert brings out, it is common to accept certain views as a group member when one does not personally believe them. Not even most members must personally believe the group view:
The fact that we have this way of speaking suggests that, in general, participating in believing that p as a body does not require personally believing that p. This accords, of course, with what was said earlier about the idea of several persons ‘X-ing as a body’: this does not entail, for every X, that each of the parties individually X.
That is perhaps a surprising result, since it suggests that . . . [i]t is not even a necessary condition of our believing that P that each or most of us believe that P personally.Footnote 5
By my lights, Gilbert's observations recommend a style of view about group belief L&P do not consider: a group agent's believing P partly consists in the members' individually accepting P qua members or for group purposes, often because P is the output of a rule they agreed to use.
Here the individual attitude of acceptance qua member can diverge from personal belief.Footnote 6 I take the distinction from Bratman (Reference Bratman1992), though it is also in Stalnaker (Reference Stalnaker1984) and Cohen (Reference Cohen1992). As he noted, mere acceptance differs from belief in being limitedly purpose-relative: when acceptance of P does not constitute belief, it is because one accepts P only for limited purposes. While he did not there focus on collective intentionality, he did note that acceptance helps in cooperative contexts: ‘[W]e [may] facilitate our group deliberations and decisions by agreeing on a common framework of assumptions. We each accept these assumptions in this context, the context of our group's deliberations, even though it may well be that none of us believes these assumptions or accepts them in other, more individualistic contexts.’Footnote 7
Given this notion, the view of group belief I have in mind is like one in Tuomela's work:
VIEW 1. For a group agent G to believe P just is for (i) operative members to accept P qua members because P is the output of an aggregation rule members agreed (maybe implicitly) to use, where (ii) because of (i), non-operative members at least tacitly accept P qua members, and (iii) the truth of (i–ii) is common knowledge among members, in the sense of Lewis (Reference Lewis1969).Footnote 8
Here ‘qua members’ is equivalent to ‘for group purposes’. The latter means ‘for the purposes of joint intentional actions with other members, or actions other members jointly authorize as representative’.Footnote 9Joint intentional actions can then be viewed – as L&P agree on p. 33 – in a neat reductive way via (say) Bratman (Reference Bratman1999)'s account.Footnote 10 Or they can be viewed as proceeding from individual, isolable ‘we-intentions’ à la Searle (Reference Searle, Cohen, Morgan and Pollack1990, Reference Searle1995) or ‘we-mode’ intentions à la recent Tuomela. While this view is not as economical as Bratman's and I find it less attractive, it does not imply non-redundant realism about group agents, which Searle and Tuomela reject. While the attitudes are of a novel type, they belong to isolable individuals.
Why does VIEW 1 matter here? Because it shows how Interesting Proposition-Wise Constitution could plausibly be true. If VIEW 1 is true, the following claim is true:
Interesting Proposition-Wise Constitution via VIEW 1. A group agent's believing P is constituted by (A) operative members accepting that P qua members, and (B) other members at least tacitly accepting P qua members because of (A), where (A) and (B) are common knowledge.
But if this claim is true, we should accept redundant realism about group belief. After all, on VIEW 1, group belief plausibly boils down to neat patterns of individualistic facts.
On VIEW 1, distinct majority beliefs can still be used to arrive at group conclusions few members personally believe via some agreed-on premise-based rule. But the group's believing a conclusion is constituted by a pattern of members accepting it qua members. So while the group's belief ‘depends on’ other factors in some sense, it is a causal/rational sense. This resolves L&P's ambiguity. While they say that a group belief in a conclusion can ‘depend on’ majority beliefs in distinct propositions, it is usually false that the group belief (partly) consists in them. So what does it consist in? I fail to see how L&P can give us an answer that makes sense of group beliefs in relevant cases without the aid of my patterns of acceptance.
For, again, recall what leads L&P to see group attitudes as swinging free from individual ones – viz., cases like Table 1. Clearly, the group agent does not believe Q simply in virtue of the fact that this is rationally required, with ‘in virtue of' read constitutively. This is clear even if members had agreed to use a rule that will recommend Q. For most members now may still reject Q qua members, and refuse to rely on Q as a premise in any joint activity. If so, it is implausible that the group believes Q. So when it does believe Q and the rule is implemented with respect to Q, we need an explanation of what makes the difference. Yet the best explanation is the one given by views like VIEW 1. Plausibly, the group believes Q only if clauses like (i) and (ii) in VIEW 1 are met. It is unclear why group belief requires more than the satisfaction of (i)–(iii). So the need to explain the difference supports views like VIEW 1.
Hence, L&P cannot establish non-redundant realism by appealing to the earlier cases. One cannot establish a view on the basis of data that are best explained by a contradictory view.
2.2. Why is this group belief?
We must address a worry. If acceptance qua member and personal belief are so different, but ‘group belief’ depends on patterns of acceptance by members qua members, why is the group-level attitude I call ‘group belief’ belief at all? Why is it not mere acceptance?
The question is readily answered.Footnote 11 Remember how mere acceptance and belief differ. One merely accepts P if one is willing to treat P as true only for some limited purposes; one believes P if one is willing to treat it as true for arbitrary purposes. Belief is, as Bratman (Reference Bratman1992) said, stable across purposes. Now, G's members will often treat certain conclusions as true only for group purposes. But that implies that when the group agent exists, it will treat these as true for all its purposes. So its acceptances will be stable. So it plausibly believes after all, even though most members do not. Belief's distinctive stability appears straightforwardly at the group level.
An analogy helps. Suppose certain friends sing iff they are together, and are rarely together. The group always sings, even though each person only rarely sings. Group belief stands to the group's singing whenever it exists as members' acceptance for group purposes stands to the friends' singing only when in the group. Just as singing is a stable property of the group but not of the members, so acceptance is a stable property of the group but not of the members. So the group's attitude plausibly is belief, though it is constituted by mere member acceptances.
A final tweak is needed. In general, one may accept P merely for practical reasons. In a war, one may act on the clearly unlikely assumption that one will survive just to do one's best. This is fine, and a common feature of mere acceptance. How can the group attitude be belief if the individual ones are of a type that may be so evidence-insensitive? To answer, we can add an evidential constraint to VIEW 1. Let VIEW 2 be VIEW 1 except that we trade (i) for:
(i*) operative members accept P for group purposes because P is outputted by some agreed-on (maybe implicitly) aggregation rule that aims to be sensitive to the group's total evidence
This resolves the worry: at the group level, a stable acceptance aimed at fitting the group's evidence deserves to be called belief, even if the individual-level attitudes are mere acceptances. Even on VIEW 2, most members need not personally believe P. Suppose G's operative members are tasked with merging evidence from all members and voicing verdicts for the group that fit it. G's larger set of non-operative members may authorise this without ever seeing all the evidence. So they may rationally personally believe otherwise. Even operative members may form no relevant personal beliefs. Their job description does not demand it.
3. NO INTERESTING OPACITY
Given §2, group-level doxastic attitudes plausibly reduce to neat proposition-wise patterns of individualistic facts. Strictly speaking, this only undermines L&P's case for Strong Nonredundancy. But L&P do make the further Opacity claim – viz., that facts about group agents are practically unknowable on the basis of individualistic facts. But I worry that Opacity is either false or uncontroversial even for redundant realists. To see this, consider two claims:
(WEAK) We can know that when and only when a precise type of individualistic pattern holds (e.g. the truth of suitable instances of (i*)–(iii) in VIEW 2), group belief exists.
(STRONG) In ordinary life, we can often know whether specific group beliefs exist because we are often in a position to know whether certain specific individualistic patterns obtain.
L&P's claim can be read to imply ~WEAK, or only ~STRONG. Suppose the former. Then if any reductive theory of the kind in §2 is true, L&P's claim must be false. For by knowing the truth of such a theory, we can know in a general way that, whenever a certain neat individualistic pattern exists, a group belief exists. Indeed, we could know that this neat type of pattern constitutes group belief in general. So WEAK would be true and their claim false.
Suppose instead that L&P's claim only implies ~STRONG. If so, then it is uncontroversial even for redundant realists. Analogies help again. Imagine for a moment that we made a big advance in the metaphysics of mind: we arrived at a true neural type identity theory that closes the explanatory gap. The type identities issued by this theory would reductively explain the mental just like ‘heat = mean molecular kinetic energy’ reductively explains heat. Clearly, even if we had such a theory, it would be onerous to run token translations in workaday life. Indeed, we could rarely if ever run such translations in ordinary life. Would that show that mental phenomena are in any interesting sense non-redundant?
Not if we really had a gap-closing neural type physicalism. Consider again the case of heat. Because heat = mean molecular kinetic energy and this identity closes the explanatory gap for heat, heat is plausibly a redundant affair. Still, we can rarely know the specific molecular facts constituting specific cases of heat. This is hard! So talk of heat remains useful.
Redundant realists agree. Indeed, if they did not, their view would be so obviously false as not to be worth discussing. What they deny is an uncloseable explanatory gap. This reveals why one of L&P's driving analogies is inapt. On p. 75 and elsewhere, they say that we can see group agents ‘[i]n almost exact parallel’ to how contemporary nonreductive physicalists see consciousness. The cases are not nearly parallel. Notwithstanding my thought experiment, there is an explanatory gap for consciousness. But there is no explanatory gap for group agency, especially given the promising reductive theories of §2. And note that these theories were type identity theories (of group belief). So L&P's analogy fails for another reason: type identity theories of individual mental phenomena are now quite unpopular. Once these disanalogies are seen, group agency looks more like heat than consciousness. And there is no reason to be a non-redundant realist about heat – at least not of any interesting kind.
L&P do suggest what would amount to a disanalogy with case of heat. They think facts about group agency are multiply realisable in a way inimical to redundant realism:
[T]he multiple realizability of group attitudes in terms of individual ones and the resulting lack of an easy reduction are amplified in the case where the group attitude on a given proposition is determined, not by individual attitudes on it, but by sets of individual attitudes across multiple connected propositions. (77)
But this is wrong if we have type reductions like the ones from §2. Once again, contrast two senses in which premise attitudes can ‘determine’ the group's conclusion attitudes: (i) they causally and rationally determine the group's attitudes, versus (ii) they (partly) constitute the group's attitudes. There is not often, if ever, determination of type (ii). Yet that conclusion attitudes can be multiply determined in the sense of (i) does not give rise to multiple realisability of the relevant sort. For given the views in §2, we can still identify the group's conclusion beliefs with clear patterns of individual conclusion acceptances, held by the members qua members. Yes, these can be multiply causally produced by different premise attitudes. But that is irrelevant. We can cause heaps of rocks to exist in many ways. We can push rocks into heaps using different rocks as tools. Those heaps are still just arrangements of rocks.
There is another weak sense in which group attitudes are multiply realisable that L&P mention first on p. 77. Group attitudes can be sustained by different majority patterns in different cases. But the analogue of this is not the multiple realisability that undercuts type physicalism. Rather, it is like how different patterns of molecular motion can give rise to the same level of mean molecular kinetic energy that heat simply is. And that cannot be enough for non-redundant realism. Heat clearly remains a redundant phenomenon in spite of that.
4. CONCLUDING REMARKS
I conclude that we should not buy L&P's non-redundant realism. Strong Nonredundancy fails because they ignore some promising ideas from an earlier literature, and because they trade on an equivocation on ‘depends on’. While groups often must believe what few members personally believe, and these group beliefs may causally and rationally depend on distinct majority beliefs, groups cannot believe what no members even accept qua members. Such acceptance is needed in a majority, and a certain pattern of it plausibly constitutes group belief. This fits with L&P's rejection of Systematicity. They are right that one reading is false – Causal/Rational Dependence Systematicity from §1. But this does not help them. And finally, as I just argued, Opacity is either false or acceptable even to redundant realists.
Work remains for both sides. I have only illustrated in outline (a) how plausible forms of redundant realism can be constructed using ideas from earlier writers, and (b) how they can explain the key data L&P try to use to argue for non-redundant realism, consistently with their rejection of the only clearly false reading of Systematicity. The theories call for further development. I do not claim to have delivered a knockdown case for any specific theory. I have just shown the promise that a style of theory has for supporting redundant realism.
What matters here is this. L&P ignored points from earlier writers that aid redundant realism and neatly explain L&P's key data. This may be the greatest oversight in the book. Tuomela, Searle and Gilbert are not discussed at any length. Each gets no more than a few incidental nods. Only Bratman's views about joint intention are laid out in detail, and only on one page (p. 33). Tuomela and Gilbert's classic papers on group belief are not even cited. For these to get no attention whatsoever is striking. The omissions are all the more striking because we can use the ideas of these theorists to defend redundant realism.
In spite of this, Group Agency remains a major contribution that deserves much further discussion, and will surely get it. I addressed only one strand in the first third of the text. It is interwoven with many ideas that are appealing and novel. Indeed, as I said, L&P's case for realism is convincing. It will, I hope, end the residual tendency to claim that group agents do not literally or really exist – a tendency visible even in writers who reject merely summative accounts of group attitudes. They should instead regard group agents as we regard heat. While heat reduces readily to mean molecular kinetic energy, the sun is literally, really hot.
