As a writing partnership, Christian List and Philip Pettit are probably best known for a paper in Economics and Philosophy that describes and generalizes the ‘discursive dilemma’ (List and Pettit Reference List and Pettit2002). That paper is one of the main points of reference for what is now a large literature on the aggregation of judgements – a literature to which List and Pettit have continued to contribute, individually and jointly. Their new book Group Agency reviews and synthesizes that body of work, and proposes an analysis of group agency in which the aggregation of judgements plays a central role.

List and Pettit say that the aim of their analysis of the judgement aggregation problem is ‘to defend the logical possibility of group agents’, presumably by showing that there are rationally defensible procedures for aggregating judgements (p. 19). The implication is that group agency logically requires that group agents use such procedures. This is an extremely strong claim; later I will raise some questions about its validity. But much of what List and Pettit say about judgement aggregation is independent of their position on group agency.

List and Pettit open their discussion of the judgement aggregation problem with an example of a discursive dilemma, due to Lewis Kornhauser and Lawrence Sager (Reference Kornhauser and Sager1986). A court of three judges has to decide a breach-of-contract case. It has to rule on the truth or falsity of three propositions: that the defendant was contractually obliged not to do a certain action (proposition O), that he did that action (proposition A), and that he is liable for breach of contract (proposition L). According to legal doctrine, L is true if and only if O and A are both true. It is understood that the judges will resolve disagreements by majority vote. Judge i concludes that O and A are both true, and that therefore L is true. Judge j concludes that O is true and A is false, and that therefore L is false. Judge k concludes that A is true and O is false, and that therefore L is false. If the judges vote separately on each proposition, they will conclude, contrary to legal doctrine, that O and A are both true but L is false. In more general terms, the problem is that the judges collectively are expected not simply to make decisions but also to provide reasons for those decisions. Their collective judgements are expected to satisfy certain principles of valid legal reasoning. Even if each individual's judgements satisfy those principles, validity can be lost in the process of aggregation.

On one interpretation, Condorcet's paradox (which forms the core of Arrow's impossibility theorem) is an instance of the same problem. In that paradox, one might say, a democratic assembly is called on to make judgements about the relative goodness of three alternative decisions x, y and z. Using majority voting to resolve disagreements, it arrives at three judgements: that x is better than y, that y is better than z, and that z is better than x. Taken together, those collective judgements contravene what is arguably a principle of valid reasoning, namely that ‘is better than’ is transitive – even though each individual's judgements are consistent with that principle. The discursive paradox shows that transitivity is not the only principle of valid reasoning that can be lost in aggregation.

In the modern literature of social choice, the Condorcet paradox has often been interpreted as a problem in the aggregation of preferences, interpreted as subjective tastes. However, Condorcet himself thought of voting as a procedure for arriving at judgements about truth and falsity. His jury theorem shows that, under certain (extremely restrictive) assumptions, the probability that a majority decision is correct increases with the size of the group of voters, and is close to one for realistically large groups. This theorem has provided a template for a good deal of recent work on designing judgement aggregation procedures with the aim of ‘tracking truth’.

List and Pettit do an excellent job of describing the problem of judgement aggregation, explaining its significance, and summarizing what is currently known about the various forms the problem can take and the possibilities for solving it. To give a flavour of what is involved in this enterprise, here are two of List and Pettit's examples of how the design of judgement aggregation procedures can be informed by theoretical analysis.

The first example applies Condorcet's jury theorem to the problem faced by a university committee deciding whether to grant tenure to a junior academic. The committee's rules require that tenure should be given if and only if the candidate has demonstrated excellence in teaching (proposition T) and has demonstrated excellence in research (proposition R). Suppose that these propositions have objective truth values, and each committee member's judgement on each proposition has an independent probability of 0.6 of being correct. If the committee proceeds by voting separately on T and R, its decisions will track the truth. (Condorcet's theorem tells us that a large committee is almost certain to reach the right conclusion about each of T and R, and hence about the award of tenure.) But even if T and R are both true, the probability that any given member will judge this to be the case is only 0.36. So if a large committee votes on the composite proposition that the tenure criteria have been satisfied, it is almost certain to reject every candidate (pp. 92–95).

The second example is a surprising result about strategy-proofness, due to David Austen-Smith and Jeffrey Banks (Reference Austen-Smith and Banks1996). Consider an odd-numbered jury which has to decide on the proposition ‘the defendant is guilty’ (G). Suppose that each juror has an independent probability of 0.6 of reaching the correct conclusion. Condorcet's theorem tells us that, if each juror records her private judgement, a large jury which decides by majority vote is almost certain to reach the correct conclusion. But consider an individual juror who wants it to be the case that the defendant is not convicted unless guilt is established beyond reasonable doubt. Suppose she is a rational Bayesian who interprets ‘beyond reasonable doubt’ as ‘with probability greater than 0.95’. And suppose that the prior probability she assigns to G – that is, her belief prior to hearing the evidence – is 0.5. Having heard the evidence, her private judgement is that G is true. How should she vote? Suppose she believes that the other jurors will record their private judgements. Her vote affects the collective decision only in the event that the other jurors are evenly split. Conditional on this event, the posterior probability that G is true is only 0.6. (For an intuitive sense of why this is so, consider what one juror j can infer from the information that exactly half of the other jurors have judged G to be true and that the others have judged G to be false. By assumption, the prior probability of G is 0.5, and there is no systematic bias in the jurors’ judgements. It follows from the symmetries of the problem that the posterior probability of G, conditional on this information, is 0.5. So j is effectively in the same position as if she had received no information about the other jurors’ judgements. Her private judgement is that G is true, but she knows that this judgement has only a 0.6 probability of being correct.) This conclusion holds, however large the jury. So the Bayesian juror will vote ‘not guilty’, contrary to her private judgement. In other words, majority voting is not strategy-proof for voters who want collective decisions to track the truth and who recognize that their own beliefs are as fallible as those of other voters.

Clearly, the judgement aggregation problem has some bearing on issues of group agency. As the examples of judges, juries and committees show, there are many cases in which groups of individuals are expected to arrive at collective judgements about propositions that are not just expressions of personal taste, and where there is an expectation that those judgements can be supported by reasons that the group collectively endorses. Another example discussed by List and Pettit is the case of a committee of scientists which is charged with advising the public about scientific findings, for example about climate change or public health (pp. 45–46). For groups like these, it seems reasonable to claim that judgement aggregation is fundamental to group agency. But notice that these examples are all of groups whose function is to deliberate about the truth value of propositions, and to do so on behalf of some wider community. How do List and Pettit generalize from this special case to group agency in general?

They start from the observation that ‘common sense and the social sciences represent many collections of human beings as if they were unitary agents, capable of performing like individuals’; the question that the book is to address is ‘How should we interpret such talk?’ (pp. 1–2). One might have thought that this was a central topic of the existing literature on plural subjects, joint action, collective intention and team reasoning, exemplified by the work of Michael Bacharach, Michael Bratman, Margaret Gilbert, Natalie Gold, John Searle, Raimo Tuomela and myself. The work of these writers is occasionally acknowledged, usually en bloc, in the course of the book (pp. 33, 194, 215), but List and Pettit distance themselves from it on the grounds that their own concern is with ‘fully fledged agents’ which ‘establish a more enduring collective entity’ than the forms of group agency analysed by other writers (pp. 1–2, 215). Obviously I have an interest to declare here, but the implicit claim that the existing literature is only tangentially related to List and Pettit's self-declared project seems rather unconvincing. Still, if List and Pettit prefer to start from scratch, that is their privilege. And their starting point is interestingly different from that of much of the current literature.

Rather than considering groups as viewed from inside and asking what group members mean when they use ‘we’ language to speak about themselves, List and Pettit take an external view of group agency. They ask what outsiders mean when they attribute agency to groups. In particular, they are interested in the use of group-level concepts, such as the supposed preferences, beliefs, intentions or actions of a group, to explain and predict group behaviour and to assign legal or moral rights and responsibilities to groups. Drawing on the ideas of Daniel Dennett, they distinguish between two stances that one can take when thinking about an organism or system. The mechanical stance views the system as an assemblage of working parts (as in the view of the brain characteristically taken by neuroscientists). The intentional stance views the system as having some kind of autonomous capacity to represent its environment, to form beliefs and to make decisions – to be the author of its actions. To treat a group as an agent is to take the intentional stance towards it (pp. 11–16). Since, on List and Pettit's account, the intentional stance towards human groups involves treating them as if they were capable of performing like individual human beings, the kind of authorship that is attributed to groups must correspond to some degree with the kind of authorship that individual human beings typically attribute to themselves.

List and Pettit declare their ontological position about groups as a combination of non-redundant realism and methodological individualism. They are realists because they maintain that ‘group-agency talk’ – that is, propositions about group-level properties, made from the intentional stance – can state truths about the world. Their realism is non-redundant because they allow that such properties may not be ‘readily reducible’ to properties of individual group members. But they are methodological individualists because they maintain that group-level properties supervene on individual-level ones: there are no ‘psychologically mysterious forces’ that work only at the group level (pp. 2–7, 64–66). List and Pettit claim that their form of ‘group-agent realism’ is a break with the two main historical traditions of realist thought about groups – the ‘authorization theory’ of Hobbes, Locke and Rousseau and the ‘animation theory’ that they attribute to progressive thinkers of the early twentieth century (pp. 7–9). That is possibly true, but it should also be said that most contributors to the literature on plural subjects and joint intentions have treated group properties as supervenient on individual ones in much the same way that List and Pettit do.

The crucial link between group agency and judgement aggregation comes through List and Pettit's claim that rationality – essentially, rationality as that is represented in decision theory – is a defining property of agency. Thus, asking how an outside observer might attribute agency to a biological organism, they say:

And:

‘Standards of rationality’ are presented as if they were the demands of some normative authority that lays down rules or requirements about what agents may and may not think. It is never clear what this authority is, or why its demands have force; but we (the readers) seem to be expected to accept List and Pettit's claims to speak on its behalf. We are told that the relevant standards of rationality are of three kinds. ‘Attitude-to-fact’ standards ‘rule in favour of representations that fit with how things are and rule against those that don't’. ‘Attitude-to-action’ standards ‘rule in favour of actions that are required – or at least permitted – by the agent's representations and motivations and rule against those that aren't’. ‘Attitude-to-attitude’ standards (which List and Pettit describe as ‘particularly important’ for their purposes) require that all the agent's representations and attitudes are mutually consistent. So if an organism or system is to count as an agent, it must act sufficiently like the agents modelled in rational choice theory. It must hold beliefs that can be represented as mutually consistent propositions, and which correspond with the available evidence. It must have desires (or ‘goals’) that can be represented as further, mutually consistent propositions. And it must choose the actions that best serve its desires, given its beliefs (pp. 24–25). List and Pettit magnanimously permit occasional mistakes, but (in a way that is reminiscent of Maoist self-criticism) demand that agents continually monitor their thoughts and actions for deviations from rationality and strive to correct these. Thus: ‘The sensitivity for the demands of rationality displayed in the acknowledgement that criticism is appropriate may be evidence of agency, albeit of an underperforming sort’ (p. 31).

It is easy to see how this kind of argument makes the problem of judgement aggregation central to group agency. List and Pettit can (and do) tell groups what they must do before they can be admitted to the community of agents. Among their demands are:

In other words, the group must organize itself in such a way that it can solve the problem of judgement aggregation. But how does all this follow from the premise that we attribute agency to groups by recognizing that those groups perform like human beings do when they think of themselves as the authors of their individual actions?

The only way I can see of linking the conclusion to the premise is by adding another premise, namely that when human beings think of themselves as the authors of their individual actions, the actions they take are generally consistent with conventional decision theory (and that if they find that they have accidentally contravened the theory, they engage in self-criticism). But that premise is disconfirmed by a huge body of experimental evidence collected by cognitive psychologists and behavioural economists. Indeed, as Hume recognized two and a half centuries ago, each of us can disconfirm it by careful introspection (for this reading of Hume, see Sugden Reference Sugden2006). Notice also one implication of accepting that premise. If we observe some human being whose individual choices systematically contravene conventional decision theory, and if that person shows no inclination to correct those supposed mistakes, we must conclude that he is not the author of his actions. Since most human beings fall into that category, this way of thinking seems unattractive and dogmatic. One might think that decision theorists should adapt their theories to fit the evidence, not criticize their subject-matter for not conforming to their preferred hypotheses.

List and Pettit sometimes seem to suggest that unless an (individual or group) agent has beliefs and desires that are consistent in something like the way assumed in rational-choice theory, it will be unable to act at all, or will act in such a chaotic way that its actions will be unpredictable. Thus, they claim that if an agent tries to act on inconsistent representations and motivations that are ‘near the coal-face of action’, there will be a ‘straightforward breakdown: actions will be supported that cannot be realized’ (pp. 24–25). This, I suggest, is a mistake (although a common mistake among advocates of rational-choice theory). If one assumes that individuals make decisions by consulting a pre-existing set of propositions about desires and beliefs and then arriving at their chosen actions by logical deduction, it is of course true that logical inconsistencies in those propositions can cause havoc. But that assumption is not an established fact about human psychology; it is just a modelling strategy used by rational-choice theorists. A more empirically grounded form of decision theory might use a different modelling strategy (as Hume's does). So we are not entitled to conclude that an individual who lacks consistent desires and beliefs will fail to make predictable decisions.

How do we human beings recognize agency in one another, if not by looking for behaviour that is consistent with rational-choice theory? One obvious (partial) answer is that we recognize certain characteristic associations between emotional states and actions. For example, we know that when one person A makes an unprovoked attack on another person B, B is likely to feel anger and resentment and an associated desire to seek revenge – even in circumstances in which B has good reason to believe that revenge-seeking will have deeply undesired consequences. Suppose we observe A attacking B, and B showing signs of anger. We may have good grounds for predicting that B will take revenge, even if we believe (and even if B himself believes) revenge-seeking to be irrational. If we then observe B taking revenge, we do not have to conclude that he is not an agent, or that he is not in control of his actions. We can think of him as the author of an irrational act. And since this kind of irrational authorship is characteristically human, it is evidence of human-like agency.

This analysis can apply equally well to groups. Suppose that A and B are nation states. B is the object of an unprovoked attack by A (or perhaps by a terrorist group that is based in A's territory). B then launches an imprudent counter-attack on A. If the criterion for attributing group agency is the capacity of a group to perform like an individual human being, B's response is surely evidence of agency. Whether or not B's action can be explained in terms of consistent group-level beliefs and desires, it can certainly be explained in terms of group-level emotional states – anger and resentment – that are also characteristic of individual human beings who take themselves to be the authors of their actions.

If group agency is analysed solely in terms of the aggregation of judgements, there seems to be no conceptual space for the attribution of emotional states to groups. Because their analysis has this feature, List and Pettit's treatments of some significant aspects of group agency are rather desiccated. A case in point is their discussion of collective guilt. Although they think it may sometimes be ‘sensible’ to hold a national people responsible for the past actions of its governments, they find it hard to square this moral intuition with their official theory of the conditions under which groups can properly be held responsible (pp. 163–169). Their difficulty is that a national people is not a ‘fully fledged group’ that has organized itself to solve judgement aggregation problems. In this context, it might be illuminating to think about group-level emotions as well as group-level judgements. Consider, for example, the action of Willy Brandt when, as Chancellor of West Germany in 1970, he knelt at the monument to the dead of the Warsaw ghetto uprising. This was surely intended as an expression of the collective remorse of the German people – and not of Brandt himself, who had resisted the Nazi regime from its very beginning. As Chancellor, Brandt was authorized to express this emotion, but to understand what he was expressing, one needs group-level emotional concepts.

In a world in which there are group agents, a person can act in her identity as an individual and in her identity as a member of a group; and the demands of those two identities may conflict. In the final pages of the book, List and Pettit ask whether there can be a ‘rational basis for adjudicating issues of identity’. They claim, in opposition to Elizabeth Anderson (Reference Anderson2001), that the answer is ‘Yes’. Anderson's argument – one that I have also defended (see Sugden Reference Sugden1993) – is that the concept of rationality is defined for a given agent and hence for a given identity; there is no neutral identity from which one can make a rational choice between identities. List and Pettit's response to this argument is that

But if, reasoning from an individual identity, it makes sense to choose between retaining that identity and switching to a group identity, why doesn't it make equally good sense to reason in the opposite direction? Just as a person can decide that his individual interests would be best pursued by joining a group agent, so a group can decide that its collective interests would be best served by dismantling the structures that support its group agency. (Think of a country collectively choosing to switch from a command to a market economy, or collectively choosing to demobilize its conscript army after a war.) Here, I think, List and Pettit's strategy of going it alone, rather than building on the existing literature on plural subjects and joint intentions, may have let them down.

To sum up: the judgement aggregation problem is a significant and neglected aspect of group agency, and List and Pettit's discussion of this problem is important and enlightening. But their claim that groups cannot be agents unless they organize themselves to solve that problem seems rather implausible, and I remain unconvinced by the arguments they offer in support of it.