Featured Video Play Icon

Joint Mental States and Quasi-Agential Groups

Luke Roelofs (Australian National University)

[Jump to comments]


Intentional explanation is a strikingly useful form of explanation, and human life would be unrecognisable if we did not routinely apply it to the understanding of individual behaviour. So it is not surprising that we routinely also use the idiom of intentional explanation – of beliefs, desires, and actions aimed at satisfying those desires according to those beliefs – to talk about human collectives. After all, if we could explain collective behaviour by reference to collective beliefs and desires, that would be a major explanatory boon. On the other hand, if we applied this explanatory scheme to phenomena which do not fit it, we would simply be entangling ourselves in unproductive metaphors. The philosophical project of analysing collective intentionality is important because it helps us to see when intentional explanations of collective phenomena are well-founded, and when they are not.

There are two well-discussed sorts of case where intentional explanations of collective phenomena are appropriate; this paper aims to make space for a third in between them. In the simpler case, all the collective claims are read distributively, as simply summarising the mental states and actions of agents. If I say ‘the people in the park believed it was going to rain and wanted to stay dry, so they ran under the gazebo’ (example from Searle 1990), I might simply mean that each individual believed it would rain, wanted to stay dry, and consequently ran under the gazebo. On the other hand, there are cases where a group of people qualifies as an intentional agent in its own right: that is, the very same criteria of agency which apply to individuals are satisfied by the group considered as a whole (see esp. Pettit & List 2011, cf. Huebner 2014). We can then explain group actions by group beliefs and desires, which may be shared by few, or even none, of the individuals making up the group. If we say that ‘the court believed that the defendant had committed the act they were accused of, and wanted to punish such acts harshly, and so it decided to apply a harsh punishment to the defendant’, this can be appropriate even if most individual members of the court do not instantiate all of those intentional states (see List 2006).

In between these two extremes, however, there is joint intentional action: action undertaken by many people together (so it is not just an aggregate of individual actions), but which nevertheless involves no group that meets the demanding criteria for itself being an agent in the full sense. But while there has been considerable philosophical attention paid to joint intentional action (see e.g. Searle 1990, Gilbert 1990, 2000, 2009, Bratman 1997, 2009, 2014, Tuomela 2007), it remains to be shown how it can feature in intentional explanations.[1] This is connected to the preponderance of attention given to joint intention, with much less given to joint belief, and almost none to joint desire. But intentional explanations turn on the interplay of these three types of state. I believe that by focusing on the inter-relation among these three notions, we can develop new ways to characterise joint beliefs and joint desires, which will illuminate the range of cases where intentional explanations are applicable to collective phenomena without being either merely summative or requiring full-fledged group agents. Yet in the course of developing these points, it will emerge that the groups to which they most often apply are likely to be in certain respects akin to agents – they are what I will call ‘quasi-agential’ even without being fully agents.

Section 1: Joint Mentality, Between Groups and Individuals

It will be useful to start by regimenting terminology, because so many terms in this area admit of multiple meanings. First, ‘distributive’ and ‘collective’ are grammatical modes of predication: whenever some predicate is applied to a plural term, it can be read either distributively, in which case the sentence is equivalent to a series of sentences applying that predicate to each individual member of the plurality, or collectively, in which case it is not.

Second, I will use ‘group’ for a singular entity which has many individuals as members, and ‘collection’ or ‘plurality’ as a singular term which is shorthand for plural reference to many individuals. That is, any sentence of the form ‘this collection does X’ is equivalent to a sentence of the form ‘person 1, person 2… and person N do X’ (where ‘do X’ could be read either collectively or distributively).

I will talk about group intentional states, joint intentional states, and individual intentional states: the first belong to a group in its own right, the second to a collection of individuals collectively, and the third to an individual. Individual mental states can also be ascribed by applying mental-state predicates distributively to collections. I will use the term ‘aggregate’ to signal such states: an aggregate belief is a belief that all or most individual members of some collection hold.

Distributive predications of intentional states to collections do not imply collective predications of the same state to the same collection, but they might be implied by the latter. For example, philosophers of joint intentionality disagree about whether some people having a joint intention requires that they each have a corresponding intention with the same content. We can formalise this as follows:

Distribution Principle (DP): When some individuals jointly intend (believe, desire) that P, it follows that all or most of them individually intend (believe, desire) that P.

Stronger or weaker versions of DP would involve higher or lower standards for ‘all or most’: the strongest would be simply ‘all’, while very weak versions might be 51% or more, or perhaps even less (cf. List 2014, pp.1602-1609). For convenience I will tend to speak as though the requirement were ‘all’, but plausibly it may be less than this, especially in very large collections.

Does any version of DP apply to joint intention? Bratman (2009) says yes, while Gilbert (2009, pp.171-173) says no, appealing to cases where two individuals undertake a joint action with a certain end point, but then both privately lose their resolve before reaching that end point. Gilbert thinks it intuitive that until they make this public, they still jointly intend that they complete the task even though neither individually intends that they complete the task (see Kopec and Miller ms, for criticism). I can feel the intuitive pull of Gilbert’s claim, but ultimately go with the pro-DP side. This is because the main sense in which the joint intention seems to remain, when the individual intentions are gone, is that the commitment both made to each other is still binding.[2] In this paper I want to focus on the most basic sorts of joint agency: those that do not require hierarchies of authority, or institutional rules or procedures, or mutual obligations voluntarily entered into, and which are at work in the active, collaborative, setting up of such hierarchies, procedures, or obligations.

What about joint belief – does that obey DP? Gilbert (1994) again argues that it does not, on the basis that people may jointly believe something none of them individually believe, by feeling bound to speak and act as though that thing were true, when speaking and acting in their capacity as group members. For the same reason as above, I disagree: I want to understand the sort of joint mental states which we can attribute to people even before they start doing things like distinguishing their ‘capacity as group member’ from their ‘capacity as individual’.

A related worry is that the ‘joint beliefs’ that Gilbert describes are undertaken voluntarily and with goals in mind, which makes them less like beliefs and more like mere ‘acceptances’ (as argued by, e.g., Wray 2002). Joint beliefs governed by DP would qualify better as beliefs in this respect, since people do not choose (individually, jointly, or as a group) to have something believed by most of them. If joint beliefs (and desires) obey DP, that makes them much more similar to individual beliefs (and desires): unchosen, sometimes inarticulate, states which are formed unconsciously and manifested in a variety of behaviour, rather than actively and deliberately formed one-at-a-time. It is beliefs and desires like this which we primarily appeal to in our intentional explanations of individual actions, while consciously-formed states are more often the things we wish to explain. So in looking for the minimum preconditions for collective intentional explanations, we should focus on states which obey DP.

Section 2: Joint Belief and Desire as the Causes of Joint Intention

The following principle regarding the mental states of belief, desire, and intention seems to be a truism:

Belief-Desire-Intention Principle (BDIP): When S desires that P, and believes that doing A will bring it about that P, S will form an intention to A.

Obviously BDIP is not an exceptionless law – for one thing, S may have countervailing desires, or may not notice the relevance of their belief to their desire. But its being true for the most part is integral to how we think about these three mental states. That is, we go a long way towards explicating our notions of belief, desire, and intention just by saying:

S desires that P iffdef S is disposed to form an intention to A, whenever S believes that doing A will bring it about that P.

S believes that Q iffdef S is disposed to form an intention to A, whenever S desires that P and Q implies that doing A will bring it about that P.

S intends to A iffdef S is in a state that disposes them to A, and which they are disposed to enter whenever they desire that P, and believe that doing A will bring it about that P.

How far these three inter-definitions capture everything that is essential to these three notions is debated – for one thing, they leave out anything either phenomenal or normative. We might still say, however, that they suffice to characterise a thin, ‘functional’, notion of belief, desire, and intention. This is what Pettit and List (2010) do, and they ascribe agency to groups based primarily on their possession of states for which these inter-definitions hold (along with certain others, such as consistency of beliefs and intentions, which I will discuss in section 4).

I propose that if there are useful notions of joint desire, belief, and intention available, they should satisfy BDIP. And we have good accounts – multiple good accounts! – of joint intention. Using one of these accounts, together with the DBIP and the distribution principle, thus offers a cheap and easy way to put constraints on a useful notion of joint desire and joint belief. Namely, joint desires and beliefs are aggregate desires and beliefs together with whatever conditions are necessary to allow for the formation of joint intentions when their contents are right. More formally:

Two or more people jointly desire that P iffdef (i) each of them desires that P[3] and (ii) they are disposed to jointly intend to A whenever they jointly believe that doing A will bring it about that P, and A is not something they can do individually.

Two or more people jointly believe that Q iffdef (i) each of them believes that Q and (ii) they are disposed to jointly intend to A whenever they jointly desire that P, and Q implies that doing A will bring it about that P, and A is not something they can do individually.

Note two shortcomings of this pair of definitions. Firstly, they are clearly dependent on one another, in such a way that it will be hard to speak meaningfully of a joint desire if we do not already have an idea of joint belief, which in turn we cannot get an idea of without some grip on joint desire, and so on. This is unfortunate from the point of view of clarity and ease of application, but it arguably corresponds to a similar holism in individual beliefs and desires. So I do not regard this as an objection to the definitions.

Secondly, as it stands these definitions are not very informative, since we have little idea what condition (ii) of each definition actually amounts to. But although the definitions do not tell us what it takes for some people to be disposed to act in the ways specified, we can look to social science and reflection on familiar social dynamics to give us some idea. Let us label the set of factors which would, in actual human societies, allow for condition (ii) in both definitions to be true, ‘corporality’ (for it allows them to desire and believe ‘as a body’).[4] Thus joint belief and joint desire, for actual humans, are a matter of aggregate belief plus corporality, and aggregate desire plus corporality.

So what is corporality? I propose to flesh out this notion by considering ways that it might be lacking. That is, we can ask ‘under what circumstances could all members of a collection individually desire that P, and believe that doing A will bring it about that P, and know that they can only do A by working together, and yet fail to do A, not through a mere failure to execute a joint intention but through failure to form one?’ More concisely: ‘if everyone wants it, and everyone knows how they can get it, why wouldn’t they at least try?’ By seeing what factors are lacking in such cases we can build up an account of what corporality would have to be – what factors must be present to avoid such failures of collective action. Those factors will be what make condition (ii) in the above definitions true: they are elements of corporality.

In the next section I discuss three ways that collection action might fail even in the presence of aggregate beliefs and desires: by identifying the conditions for avoiding them, I develop a preliminary description of corporality. This in turn can provide an explanatorily useful conception of joint beliefs and desires, as the kinds of things that tend to lead to joint intention and thus to joint action.

Section 3: What is Corporality? Knowledge, Coordination, and Incentivisation

Joint beliefs and desires are meant to provide intentional explanations of joint action, in the sense discussed by Bratman, Gilbert, Searle, and others. This means that they must tend to produce such action when their contents rationalise it. This means that to specify what constitutes a joint belief or desire, we need to rule out cases where aggregate beliefs and desires, whose contents rationalise some joint action, might systematically fail to produce such action. So how might this happen?

I will discuss three conditions where such systematic failure might occur, all of which have been widely discussed by social scientists: ‘mutual ignorance’, ‘coordination problems’, and ‘collective action problems’. Roughly speaking, these are the cases where:

  • Everyone desires that P, and believes that jointly doing A will bring it about the P, but some or all people do not know that the others desire and believe these things.
  • Everyone desires that P, and believes that jointly doing A will bring it about the P, but some or all people cannot settle on mutually consistent ways of doing their parts in A.
  • Everyone desires that P, and believes that jointly doing A will bring it about that P, but some or all are unwilling to do their parts in A, despite wanting the outcome, due to other desires they have.

The first of these possibilities is straightforwardly derived from the inclusion in many accounts of joint intention of a ‘common knowledge’ condition (e.g. Bratman 2009, p.157, Gilbert 2009, pp.180). If we act in ways that in fact accomplish a common goal, or even in ways that in fact depend on each other’s actions, we are still not acting together if we are ignorant of this fact. Admittedly, the exact characterisation of common knowledge is tricky (see Lewis 1969, Friedell 1969, Monderer and Samet 1988). For one thing, belief would seem as good as knowledge if our focus is on explaining what people do.[5] To circumvent such worries, I will adopt a locution Bratman sometimes uses to express the idea behind the common knowledge condition, namely that the relevant facts should be ‘out in the open’.

Joint desire and belief require the same sort of ‘common knowledge’, because where common knowledge fails, joint action is likely to also fail. For an example of aggregate desire not leading to joint action due to a failure of common knowledge, consider five friends, each of whom wishes they would go out to the bar together (we may assume that they all have accurate beliefs about how to go to bars). But each is unaware of the other’s sharing this desire, and so nobody makes the first move and each remains at home. For an example of aggregate belief not leading to joint action due to a failure of common knowledge, consider the ‘fog of war’. Five soldiers share the desire to capture an enemy bunker, and this much is common knowledge. Each has moreover learned, independently of the others, that the bunker has a weak spot at which a few people attacking together could easily get inside, though a single attacker would fail. Because communications between them have broken down, and none knows that the others know of the weak spot, none can expect their attack to be supported and thus no attack is launched.

So common knowledge of both desire and belief are needed for aggregate beliefs and desires to lead to joint action; thus common knowledge is one element of ‘corporality’. What does deserve mention, however, is that there are two necessary components to secure common knowledge, one physical and one sociological. First, there must be means of communication available – this is what the soldiers lack. Second, people must be disposed to share information about their beliefs and desires fairly freely, and not held back by social inhibitions, or by the fear of punishment – this is what the five friends lack. So we may say: people jointly believe (desire) that P only if all or almost all of them believe (desire) that P, and they are both physically able and psychologically willing to share this information freely.

A second component of corporality comes from the requirement of what Bratman calls ‘meshing sub-plans’ (2009, p.158[6]). Joint action requires not only that we both aim at a goal, knowing that we are doing so, but also that we be disposed to co-ordinate our efforts so as to not get in each other’s way. Consider a case where everyone wants P, and knows this, and everyone knows how they can jointly achieve P, and knows this too, but where moreover the joint action required could be done in a number of ways (at a minimum, by assigning different people to its sub-tasks), and people cannot settle on a particular way. Joint action would then not be likely, both because any actual attempt is likely to fail, with different people using incompatible methods, and because people are likely to know this and not even try as a result.

This sort of situation has been studied under the heading of ‘coordination problems’ (e.g Cooper 1998, Chwe 2001). A simple example is two people trying to meet, but not able to communicate and so unable to settle on a particular location. With larger numbers of people the options for failures of coordination are even greater, and so the ability to co-ordinate will be an important factor in moving from aggregate beliefs and desires to joint action: thus ability to co-ordinate is a component of ‘corporality’.

Like common knowledge, the ability to co-ordinate requires both physical infrastructure and certain social dispositions. We need to be able to communicate, and we need to be willing to communicate – and for some joint actions to communicate at tedious length.[7] So we may say: people jointly believe (desire) that P only if all or almost all of them believe (desire) that P, and they are both physically able and psychologically willing to seek and settle on co-ordinated plans.

The third factor pertains not to knowledge or communication but to incentives. It might be that every individual wishes for action A to be performed, but due to their other desires some or all of them are unwilling to do their part to accomplish A. But there are actually four different types of case we should distinguish here, and only some would violate BDIP. That is, there are some ways that people’s desires might prevent them jointly acting on their aggregate desires and beliefs, which are compatible with them having joint desires and beliefs.

The relevant cases differ on three dimensions: the average strength of the aggregate desire, the connectedness or diversity of the countervailing desires, and whether the countervailing desires pertain to the object of the aggregate desire, or to the individual action it requires.

The first dimension is simple: does the collection fail to act on a strong or on a weak aggregate desire? If the aggregate desire is weak, then it is less surprising that it not be acted upon: such a case would be rather like an individual who feels a weak desire for some outcome but does not bother to act on it due to laziness, inattention, or distraction. To put it another way, one of the ways that BDIP should be qualified is to allow that a desire’s tendency to issue in intentional action is stronger, the stronger the desire is. So it is compatible with aggregate beliefs and desires being jointly held that when the desires are very weak, they have little tendency to produce joint action.

The second dimension concerns whether the reasons that various individuals have for not engaging in joint action are the same or different: when many individuals are unwilling to participate, is it just a coincidence, with them each having unrelated personal reasons, or are they all unwilling for similar reasons? To define corporality, we should look at cases where the countervailing desires are at least somewhat cohesive, because the above qualification to BDIP (that action becomes more likely as desire becomes stronger) remains true when the only countervailing desires to consider are diverse.  This is because as the aggregate desire becomes stronger, it becomes statistically more and more unlikely that significant numbers of individuals have independent desires strong enough to override it. Since we are not looking for exceptionless rules but only useful generalisations, it is enough for our purposes that the likelihood of this kind of failure to act declines as the aggregate desire becomes stronger. Thus corporality need not rule out cases where the aggregate desire is very weak, or where the countervailing desires are strong and diverse.

The third dimension of variation concerns what exactly it is that individuals do not want. They may not want the outcome which the aggregate desire itself aims at – that is, they may have conflicting desires about the very same state of affairs, and the desire to avoid that outcome is stronger than the desire to achieve it. Or they may wholeheartedly desire the outcome, but dislike the steps that they individually would have to take to achieve it – i.e. each would ideally like the outcome to come about, but by someone else’s efforts. If it is the former, and if moreover the reasons are cohesive, then it turns out that one aggregate desire is being overridden by another – and this is not a violation of BDIP, which clearly ought to allow for desires to be overridden by other desires. It is only the latter case – when everyone desires the outcome, but is unwilling to bear the individual costs it requires – that corporality needs to rule out.

Cases of this sort are already well-studied, under the name of ‘collective action problems’. Their essential structure is exhibited in the “prisoners’ dilemma” (Tucker & Kuhn 1950), where two people both desire an outcome which they can achieve by both taking a certain action, but where each does better if they let the other take that action alone, and loses more from taking that action alone themselves than if neither takes it. A wide range of social problems, from climate change to consumerism, have been modelled in these terms (Cole 2008, Heath 2000). Thus part of ‘corporality’ will have to be some mechanism for solving collective action problems. Such mechanisms may either be things like legal punishments for free-riding or rewards for contributions to the collective effort, which change the decision-environment that individuals confront, or things that change individual psychology, such as inculcating a strong desire for joint actions to succeed. The various ways of solving collective action problems have been discussed at length, so I will not repeat that discussion here (though see, e.g., Axelrod 1981, Olson 1971,  Ostrom 1990, 1999). So we may say: people jointly believe (desire) that P only if all or almost all of them believe (desire) that P, and they are sufficiently motivated (by whatever motives) to do their part in a joint action to achieve P.

In summary, people have a joint belief or desire when they have an aggregate belief or desire, together with at least the following three things: common knowledge of that belief or desire, sufficient ability to coordinate on the actions which it would rationalise, and sufficient incentives for individuals to willingly contribute to such actions. The first two require both physical mechanisms of communication, and social willingness both to share information and to find and accept coordinated plans. The third could be accomplished either by the inculcation of dispositions to conformity, or by skill in the provision of rewards and sanctions. There might turn out to be other requirements, but these three will suffice for now.

Section 4: Quasi-Agential Groups

Most of the elements of corporality are not specific to particular desires or beliefs. If people are in regular communication, have procedures for coordinating, and are incentivised to bear sacrifices for the sake of joint goals, then all their aggregate beliefs and desires are likely to be joint ones; if they are not, most likely none will.[8] Thus it is useful to look not just at particular states on particular occasions, but at the group in general. A group may have more or less corporality, and so may approximate better or worse to the model of an agent as something which has beliefs and desires and can be expected to perform any action which they rationalise.

It should not be thought that corporality is hard to establish. The physical conditions that enable coordination and commonality of knowledge can be established simply by spatial proximity, and the psychological conditions for these things, and for individual willingness to make costly contributions, can be established by the normal psychology of social relationships (see e.g. MacMillan & Chavis 1986). If a hundred people live in the same village, with regular communication, mutual concern, and a willingness to work together, then they might well show a high degree of corporality, so that whatever beliefs and desires were aggregate would be liable to lead to joint action, and thus jointly held. Obviously, corporality in larger groups will be harder to achieve, and formal institutions (e.g. chains of command) and ideologies (e.g. nationalism) can be seen as partly aiming to render larger groups more corporal than they would otherwise be.

High levels of corporality make a collection of people operate in some respects like an agent, in that some but not all of the defining principles of agency hold of them collectively. What does hold is BDIP: they will tend to act (jointly) whenever it is rationalised by their (joint) beliefs and (joint) desires. What does not hold of them is, firstly, consistency: it is not true that they will generally have only consistent sets of (joint) beliefs or (joint) intentions, or that when they have inconsistent (joint) beliefs or (joint) intentions they will take steps to correct this inconsistency. Another way in which they are not fully agential is the lack of scope of their attitudes: on any topic where many of the individuals disagree, they will neither jointly believe P nor jointly believe not-P, or neither jointly desire that P nor jointly desire that not-P. If they have little agreement, then they may have so few joint beliefs and desires that rationalising explanations only rarely make useful predictions.[9]

These two failings are inversely related: the more we allow ourselves to say that people have a joint belief (say) even when they are not unanimous, but merely show some strong majority opinion, the more easily will inconsistencies arise, while if we try to rule out inconsistencies by insisting on unanimity, the fewer will be the joint states we can posit. This is (a simplified version of) the ‘discursive dilemma’ (List 2006, cf. Arrow 1950), which Pettit and List use to show that even weak versions of the distribution principle must fail when we are dealing with genuine group agents.

I will use the term ‘quasi-agential’ for groups whose members have both high corporality, and high agreement in beliefs and desires (i.e. a high proportion of individual beliefs and desires are ‘aggregate’, shared by many individuals). High corporality ensures that beliefs and desires which are aggregate will be jointly held, while high agreement ensures that there will be enough aggregate beliefs and desires to underwrite a range of interesting predictions. Quasi-agential groups are not group agents, but they are a natural precursor, since their propensity to joint action will both motivate steps to prevent contradictory beliefs and intentions (which would lead to self-defeating actions), and allow them to set up measures to do this which will have enough support and participation to work. Quasi-agential groups are fertile ground for the creation of agential groups, and thus provide a fertile conceptual link between joint agency and group agency.[10]

At the same time, quasi-agential groups allow for explanations at a collective level that are relevantly and usefully different from explanations at an individual level. This is not because an explanation purely in terms of individuals could not be given: I think it could in principle, just as my every move could in principle be predicted by a neuroscientifically-omniscient person with superhuman calculating abilities. But just like the best higher-level explanations, explaining that some people did something jointly because they jointly believed that it would bring about something they jointly desired provides elegance and unity by abstracting from irrelevant details. This is true particularly because so much of the individual-level explanation would have to mention each individual’s beliefs about and attitudes towards the other individuals, who in turn had complex beliefs and attitudes concerning the others, and so on in a web of relations that would be grotesquely complicated to spell out in detail. It is an advantage to cut past those multiple looping feedback relations by saying that the group involved satisfied certain group-level conditions, such as those comprising what I have labelled ‘corporality’.




Arrow, K. (1950). A difficulty in the concept of social welfare. The Journal of Political Economy 58:328–346.

Axelrod, R. (1981). The Emergence of Cooperation Among Egoists. The American Political Science Review 75: 306–318.

Binmore, K., and Brandenburger, A. (1988). Common knowledge and Game theory. ST/ICERD Discussion Paper 88/167, London School of Economics.

Bratman, M. (1997). I Intend that We J, in R. Tuomela and G. Holmstrom-Hintikka, eds., Contemporary Action Theory vol. II, Dordrecht: Kluwer — Synthese Library Series, 1997): 49-63.

Bratman, M. (2009). Modest Sociality and the Distinctiveness of Intention. Philosophical Studies 144: 149-65.

Bratman, M. (2009). Shared Agency. In Chris Mantzavinos, ed., Philosophy of the Social Sciences: Philosophical Theory and Scientific Practice, Cambridge University Press

Bratman, M. (2014). Shared Agency: A Planning Theory of Acting Together. Oxford University Press

Campbell, R., and Sowden, L. (1985). Paradoxes of Rationality and Cooperation (eds.), University of British Columbia Press, Vancouver.

Chwe, M. S.-Y. (2001). Rational ritual: Culture, coordination, and common knowledge. Princeton: Princeton University Press.

Cole, D. (2008). Climate Change and Collective Action. Current Legal Problems 61(1): 229-264

Cooper, R. (1998). Coordination Games, Cambridge: Cambridge University Press.

Eshelman, L. (1980). Reason, Egoism, and the Prisoners’ Dilemma. Canadian Journal of Philosophy 10 (2):169 – 177

Friedell, M. (1969). On the Structure of Shared Awareness. Behavioral Science 14: 28-39.

Gilbert, M. (1990). Walking together: A paradigmatic social phenomenon. Midwest Studies in Philosophy 15, 1-14.

Gilbert, M. (1994). Remarks on collective belief, in F. Schmitt ed. Socializing Epistemology, Maryland: Rowman & Littlefield.

Gilbert, M. (2000). What is it for us to intend? In M. Gilbert, ed., Sociality and responsibility. Lanham, MD: Rowman & Littlefield: 14-36.

Gilbert, M. (2002). Collective guilt and collective guilt feelings. The Journal of Ethics, 6: 115-143.

Gilbert, M. (2009). Shared Intention and Personal Intentions. Philosophical Studies 144(1): 167-187

Heath, J. (2000). Ideology, Irrationality and Collectively Self-Defeating Behaviour. Constellations 7(3): 363-371.

Heath, J. (2001). Problems in the Theory of Ideology, in Pragmatism and Critical Theory, ed. J. Bohman and W. Rehg, Cambridge, MA: MIT Press

Huebner, B. (2011). Genuinely Collective Emotions. European Journal for Philosophy of Science, V1:1, pp.89-118

Hurley, S. (1991). Newcomb’s problem, prisoners’ dilemma, and collective action.  Synthese 86 (2): 173-196

Hutchins, E. (1995). How a cockpit remembers its speeds. Cognitive science, 19, 265-288.

Kagan, S. (2011). Do I Make a Difference? Philosophy and Public Affairs, 39: 364 – 395.

Kopec, M., & Miller, S. (ms). Shared Intention is Not Joint Commitment.

Lewis, D. (1969). Convention: A Philosophical Study, Oxford: Blackburn.

Lewis, D. (1985). Prisoners’ Dilemma is a Newcomb Problem, in R. Campbell and L. Sowden (eds.), Paradoxes of Rationality and Cooperation, University of British Columbia Press, Vancouver.

List, C. (2006). The Discursive Dilemma and Public Reason. Ethics 116: 362-402

List, C. (2014). Three Kinds of Collective Attitudes. Erkenntnis 79: 1601-1622.

McMillan, D., and Chavis, D. (1986). Sense of Community: A Definition and Theory. Journal of Community Psychology 14: 6–23.

Monderer, D. and Samet, D. (1989). Approximating Common Knowledge with Common Beliefs, Games and Economic Behavior 1: 170-190.

Nefsky, J. (2011). Consequentialism and the Problem of Collective Harm: A Reply to Kagan. Philosophy and Public Affairs, 39: 364 – 395.

Olson, M. (1971) [1965]. The Logic of Collective Action: Public Goods and the Theory of Groups (Revised ed.; originally published 1965). Harvard University Press

O’Madagain, C. (2012). Group Agents: Persons, Mobs or Zombies? International Journal of Philosophical Studies, V22:2

Ostrom, E. (1990). Governing the Commons: The Evolution of Institutions for Collective Action. Cambridge, UK: Cambridge University Press.

Ostrom, E. (1999). Coping with tragedies of the commons. Annual Review of Political Science 2: 493-535

Pettit, P. and List, C. (2011). Group Agency: The Possibility, Design, and Status of Corporate Agents. Oxford University Press

Pettit, P. (2007). Responsibility Incorporated. Ethics 117: 171-201.

Searle, J. (1990). Collective Intentions and Actions. in Intentions in Communication, ed. by Philip R. Cohen, Jerry Morgan, and Martha E. Pollack. Cambridge, Massachusetts: MIT Press

Stinchcombe, A. (1980). Is the prisoners’ dilemma all of sociology? Inquiry 23 (2):187-192

Tollefsen, D. (2002). Collective Intentionality and the Social Sciences. Philosophy of the Social Sciences 32 (1): 25-50.

Tucker, A., and Kuhn, H. (1950). Contributions to the theory of games (eds.), Annals of Mathematical Studies

Tuomela, R. (2007). The philosophy of sociality: The shared point of view. New York: Oxford University Press.

Velleman, J. D. (1997). How To Share An Intention. Philosophy and Phenomenological Research, Vol. 57, No. 1, pp. 29-50

Williamson, T. (2000). Knowledge and Its Limits. Oxford University Press, Oxford.

Wray, B. (2001). Collective Belief and Acceptance. Synthese 129: 1-15: 319-333.



[1] List 2014 distinguishes a form of group attitude that is more than just a summary but less than a group agent’s attitude, namely a ‘common attitude’, which is a widely shared attitude whose wide sharing is common knowledge among those who share it. As he argues there, common attitudes are a precondition for many forms of joint action: my notion of a ‘joint attitude’ is stronger yet, in aiming to be a predictor of joint action.

[2] Indeed, Gilbert’s account of joint intention makes joint commitment central, and the account has in consequence been accused of circularity: to intentionally do anything jointly requires first jointly committing, but committing is presumably something that must be done intentionally, so doing it jointly seems to require a prior joint commitment, which in turns requires another, and another, and so on (cf. Tollefsen 2002).

[3] It should be emphasised that P is to be understood as a non-indexical content. If everyone’s desire is that “I get the cake”, they do not desire the same thing – rather, they each desire something different. They would each desire the same thing if there was a particular person who they each desired to get the cake.

[4] I will discuss condition (ii) of both definitions together, since I believe that they will be satisfied by roughly equivalent factors.

[5] For another thing, common knowledge sometimes seems to have an indefinitely iterated character: I know that you want it, and I know that you know that I know that you want it, and I know that you know that I know that you know that I know that you want it… But it’s not clear we should include this as a requirement: if two people can have iterated knowledge of the other’s knowledge up to the third level, that would seem to be enough for practical purposes.

[6] I believe a similar requirement is implicit in Gilbert’s requirement that to jointly intend some action we must strive “as far as possible to emulate, by virtue of the actions of each, a single body that intends” that action (2009, p.180).

[7] Institutional procedures of coordination, like division of labour or a chain of command, might be thought of a ‘stored’ deliberation, in that they allow for a particular coordinated action based not on everyone thinking about and agreeing to that particular plan of action, but on everyone having thought about and accepted the institutional procedures at various earlier times.

[8] This is certainly not an exceptionless rule. For instance, a collection of people might observe taboos about talking publicly about certain topics, with the result that aggregate desires pertaining to those topics were not common knowledge, and hence not joint desires, while aggregate desires pertaining to other topics were. To this extent the discussion here, in which corporality is equal for all aggregate states, represents a simplification. But I believe it is a simplification much less distorting than the opposite simplification, of assuming that corporality for any particular aggregate state is independent of corporality for any other.

[9] Another aspect of agency is the tendency for beliefs to be responsive to evidence: if beliefs have no correlation with the truth of their contents, it is less clear that they are really ‘functioning as beliefs’. By and large aggregate beliefs will tend to be as evidence-responsive as the individual beliefs that constitute them (a joint belief ‘an earthquake is happening’ might be responsive to evidence because it arises only when an earthquake is happening, because individuals will only believe that en masse when an earthquake is happening). But divergences are entirely possible: a group in which people with a track record of being right are more likely to be talked with would be more evidence-responsive than a group where such people are shunned for being ‘know-it-alls’, even if the average evidence-responsiveness of each individual member is the same. The precise way to spell out the requirement of evidence-responsiveness is of course complicated by the difficulty of saying what counts as ‘evidence’ at the collective scale: in this paper I do not attempt to address this interesting question.

[10] Pettit 2007 speaks in a similar way of ‘embryonic group agents’, but this is simply a pragmatic category: they are groups which it would be useful for us to treat as agents, in the hope that they will incorporate themselves and become agents. It is not an intermediate level of intrinsic intentional functioning, as is my notion of a quasi-agential group.

5 thoughts on “Joint Mental States and Quasi-Agential Groups”

  1. In “Joint Mental States and Quasi-Agential Groups”, Luke addresses an important lacuna in the existing literature on joint actions. Although the literature offers careful analyses of joint actions and intentions, there is only relatively little work on how these analyses figure in explanations. This is puzzling given the importance of explanations to theories of intentionality, the social sciences, and human life.

    Luke proceeds in three steps. First, he focuses on intentional explanations. In particular, he sketches the role of different mental states in intentional explanations. Central here is what Luke calls the BDIP principle that connects beliefs, desires, and intentions – the three mental entities that figure in intentional explanations. Second, he defines joint beliefs and joint desires. These are the ingredients of intentional explanations that so far have been under explored in the literature on collective actions. Important here is the observation that, for example, joint beliefs occupy some middle-ground between mere summative beliefs and beliefs of group agents. This leads to the conclusion that intentional explanations of joint actions apply to what Luke calls quasi-agents. Third, Luke elaborates some of the conditions that are necessary for such quasi-agents to function successfully. The relevant term Luke uses to denote the mechanisms by which a collection gets its act together is “corporality”. Specifically, a collection must have common knowledge, coordinate their actions, and make sure that individuals comply to overcome collective action problems. So in short, the three steps in the paper are (1) intentional explanations, (2) joint beliefs and desires, and (3) corporality.

    In my comments I will focus on the notion of intentional explanations and, in a shorter comment, on the definition of joint beliefs. I am quite sympathetic to the project overall and I agree with the main conclusion that intentional explanations are fit to be applied to joint actions. Furthermore, I do not comment on the issue of corporality because, to my mind of the three steps in the paper, it is the issue that is best explored in the literature.

    1. On intentional explanations

    A central topic of the paper is intentional explanation. My worry is that the paper lacks a sufficiently good account of what an intentional explanation is. Luke presents the Belief-Desire-Intention-Principle – BDIP for short. We could take this to give an account of intentional explanation. But I’m not sure that will work. I have worries about BDIP.

    First, it should be noted that the BDIP principle, as the name suggests, relates beliefs, desires, and intentions. But the phenomena that we most often subject to intentional explanation are actions (or behavior). It is not obvious how intentions relate to actions. Not all intentions lead to actions. On one understanding of “intention” there is a gap between intentions and actions. For example, suppose I intend to call my mother this week but, come Sunday, I still did not get myself to do it. Holton (Willing, Wanting, Waiting; 2009) argues that in linking intentions to actions oftentimes an additional special kind of intention is needed, which he calls a resolution. If we understand “intention” as it occurs in BDIP in this way, then BDIP is not an account of intentional explanation because it does not relate to the kind of phenomena that we want to explain: actions. BDIP relates to intentions but there is still this gap between intentions and action. What to do? Well, we could understand “intention” as it occurs in BDIP differently. We could understand “intention” such that it always, or at least typically, translates into action. Then the gap between intention and action would be closed and BDIP could be an account (or part of an account) of intentional explanation. But I think there are still some challenges ahead. I want to suggest a dilemma. Either BDIP does not give us an account of intentional explanation, or if it does, then only if some further apparatus is introduced.

    Before I get into this argument, a clarification. When I say that BDIP is “an account of intentional explanation” I mean the following. I mean that BDIP formulates a true principle between mental entities and the phenomena that we want to explain. Our intentional explanations refer to these mental entities in explaining the phenomena and these explanations are true partially in virtue of this principle being true. I hope this clarifies how I understand the status BDIP and I hope this understanding is consistent with what Luke has in mind in the paper.1 After this clarification, I can go on in my worries about BDIP that it either is not an account of intentional explanations or that it needs some further apparatus.

    Luke writes that there are exceptions to BDIP. For example, BDIP does not hold for countervailing desires. I take it that a pair of desires is countervailing if and only if one desire has the content that P and another that not-P. But even setting aside such countervailing desires, there still seem to be many situations in which an agent will face practical dilemmas. That is, the agent has the intention to A but also the intention to B but cannot do both. Example: Suppose that one morning when leaving your house for your commute you find that you have 1.50$ in your pocket. Lucky you! That is exactly how much the train ticket costs. Of course, you desire not to be caught free-riding, you believe that you need to buy a ticket in order not to be caught, and therefore intend to buy a ticket. But at the ticket machine you meet your handsome neighbor who is out of cash and asks you for 1.50$. Lucky you! That is an opportunity to score some brownie points with your handsome neighbor. You desire to be considered helpful by your handsome neighbor, you believe that giving him 1.50$ will make him think of you as helpful, and therefore you intend to give him 1.50$.

    It seems that you now face some kind of practical dilemma. You intend to buy a ticket and you also intend to help your handsome neighbor by giving him 1.50$ – but you cannot do both. What to say?

    First, one could say that my definition of countervailing desires is wrong. The kind of case I describe is exactly one that is meant as involving countervailing desires and, hence, it is a case in which BDIP does not apply. The good thing about this response is that is salvages BDIP from potentially problematic cases without introducing any additional apparatus. We just need to define countervailing desires as leading into practical dilemmas and stipulate that BDIP does not apply to these cases. They are just out of the principle’s scope.

    But the bad thing about this response is that this seems to severely restrict the scope of situations to which BDIP applies. Situations in which non-contradictory desires lead us into practical dilemma seem very common. Despite them being very common, BDIP would not apply. That’s bad news for BDIP. And it’s getting even worse. Suppose I end up buying a ticket for myself and not giving the money to my handsome neighbor. Without doubt, this behavior of mine can be explained. It is a phenomenon to which intentional explanations apply. But when it is a phenomenon in which an intentional explanation can be given but BDIP does not apply, then BDIP is not an account of intentional explanation.

    Second, we could say that in practical dilemmas only one intention survives. For example, the stronger intention survives in cases where one must choose. Because I bought a ticket and did not give the money to my handsome neighbor, that means that my desire not be caught free-riding is stronger than my desire to be considered helpful by my neighbor. That sounds very plausible as an intentional explanation. In fact, this idea of strength of desire is exactly one way of understanding an agent’s preference (which, it should be noted, in contrast to desires is a binary attitude). There are several good things about this response that only one intention survives – the one that is backed by the strongest desire. Not only sounds this response plausible, we seem to have the apparatus readily available to make this idea precise (think of the idea of preferences, rationalizations, and choice functions from microeconomics). Furthermore, this response also has going for it that intentions can always translate into actions (setting aside blockers to execution like weakness of the will). On this understanding of “intention”, BDIP can be an account of intentional explanation.

    The bad thing about the response is that it needs additional apparatus. If you consider, for example, the definition of “S intends to A” as it is given in the paper, then this definition does not mention the strength of desires. Desires as Luke writes about them in the paper are often not graded. The apparatus that is needed to make BDIP a plausible account of intentional explanation is not sufficiently worked out in the paper. This incompleteness is a problem because the account of graded desires – or preferences – is not straightforward and it is also controversial. How actions are explained by graded desires is a difficult topic in philosophy of economics (with theoretical economics arguably the best account we have of graded desires). There is vast disagreement about what role such graded desires play. It is also controversial whether they are merely theoretical constructs or real entities (this refers to the conflict between mentalism and behaviorism). Of course, none of these are problems that Luke’s paper attempts to solve. But if, as I suspect, BDIP is meant to be the rough outline, at a general level, of an account of intentional explanation, then I think more could be said to motivate BDIP. In particular, I think it would help to clarify how intentions relate to actions and – if Luke wants to take this route – how the strength of desires can resolve common practical dilemmas.

    2. On joint beliefs and desires

    On the definitions of joint desire and joint belief (on page 6 in the paper) I have one quick point. It concerns the sufficient condition part of the definition. I worry that the sufficient condition is not sufficient because, as things stand, the definition seems to overgenerate. If the definition is taken literally, it seems there are intuitively too many joint desires. I think this is not a deep problem. Just as with my previous point on intentional explanation, the point here amounts largely to a request for clarification. That I present my points as objections – totally disregarding the many merits of the paper – is just owed to the conventional format of such a commentary.

    Here is the definition of a joint desire. Two or more people jointly desire P if and only if (1) each of them desires that P, (2) they are disposed to jointly intend to A whenever they jointly believe that doing A will bring it about that P, and (3) A is not something they can do individually.

    Now the problem of overgeneration arises because A can be something that the individuals also cannot do collectively (condition 3 is only that they cannot do it individually).

    Suppose that each individual desires to be rich (condition 1). The individuals are also jointly disposed to intend to find a yet unknown prime number whenever they believe that finding a yet unknown prime number will make one rich (condition 2). Finding a yet unknown prime number is not something one can do individually (condition 3).2 Since these three conditions are jointly sufficient, it follows from the definition that the individuals jointly desire to find a yet unknown prime number. I think there will be many cases of this kind. For a less outlandish case we may consider that of jointly believing that winning the lottery makes one rich. Even if the members were to believe that, and even if they were disposed to form a joint intention to win the lottery, then it seems odd to say that they jointly desire to win the lottery.3 Think for example of my soccer team. Each of us believes that winning the lottery makes one rich. But together we only play soccer. The lottery we play individually.

    It should be noted that the individuals do not actually need to believe that finding a yet unknown prime number will make one rich. Condition 2 requires only that if the individuals were to believe this, then they would be disposed to form an intention. And even if they jointly believed that finding a yet unknown prime number, they would only need to be disposed to form an intention and not actually form it. All this is fairly weak – once a collection has some degree of corporality. So my worry is that perhaps some other condition seems to be missing. Further down on page 6, Luke hints at such a condition. He writes that the individuals “know that they can only do A by working together”. Similarly on page 7 he writes “everyone knows how they can get it”. So perhaps the missing condition – in addition to the condition that A cannot be done individually – is that A is feasible as a joint action. In other words, it can be done collectively. Perhaps this is implicit in Luke’s idea of a disposition to intend. But either way – whether as a new condition or as a clarification on the idea of the disposition – it might help to give a short clarification.




  2. I found textual evidence that Luke seems to use BDIP to be aiming at something akin to rational choice. On page 7 he writes that the mental entities “must tend to produce such action when their contents rationalise it”.
  3. Ok, perhaps this claim is false. But then there might be another counterexample of the same idea: it is something that is intended as an end but the mean cannot be achieved neither by an individual nor by a collection.
  4. This case has, admittedly, the same problem: condition 3 would rule out that it is jointly intended. But suppose that squaring the circle would make one rich. The individual desire being rich, they cannot square the circle individually, but if they believed that squaring the circle would make them rich, they might be disposed to intend to square the circle. (Now perhaps this second condition is not met. The argument would be that an agent is never disposed to take a means that she knows to be infeasible).
  5. 1. Introduction

    In his rich and stimulating paper Joint Mental States and Quasi-Agential Groups, Luke Roelofs argues that there is a need to posit joint beliefs and joint desires in order to provide folk psychological explanations of some (if not all) joint intentional actions—intentional actions performed by several individuals together. So, for example, if a group of five friends jointly desire that they now go for coffee together, and they also jointly believe that by now going to Farley’s—a nearby coffeeshop—then they can imminently have coffee together. By knowing that they have this joint desire and this joint belief, we can predict that it is likely that they will form a joint intention to go to Farley’s and then act on this joint intention. If we observe them going to Farley’s, we can make sense of their decision and the ensuing joint intentional action in terms of these joint states.

    Together, this collection of friends would—given that they really had these joint beliefs and joint desires—have a kind of cohesion that enables us to fruitfully make sense of and predict their behaviours by treating them collectively as one body (at least with respect to some particular context or issue). However, they do not meet the conditions of group agency: the group does not have its own beliefs, desires and intentions that are distinct from the beliefs, desires and intentions of its members (see e.g. List and Pettit 2011). This is how Roelofs characterises the kind of collective phenomena that he is interested in:

    In this paper, I want to focus the most basic sorts of joint agency: those that do not require hierarchies of authority, or institutional rules or procedures, or mutual obligations voluntarily entered into, and which are at work in the active, collaborative, setting up of such hierarchies, procedures, or obligations. (pp. 3-4)

    If the friends we were considering generally have many aggregate beliefs and desires and exhibit a kind of cohesion in their behaviour across a range of different circumstances, then they would make a group that is what Roelofs calls “quasi-agential”. This notion of a quasi-agent is supposed to provide “a fertile conceptual link between joint agency and group agency” (p. 14). The members of a quasi-agential group are likely to frequently engage in intentional joint action and it will often pay off to ascribe joint mental states to them in order to understand what they are doing as a group.

    What are joint mental states? As Roelofs correctly observes, philosophers interested in the phenomenon (or phenomena) of several agents acting together have spent a lot of time and effort developing accounts of joint intention, but they relatively little have been said about the notions of joint belief and joint desire. In his paper, Roelofs first provides constraints for accounts of joint beliefs and joint desires, by asking how joint beliefs and desires are supposed to interact with joint intentions in intentional explanations of joint intentional action.

    According to Roelofs, an adequate account of the kind of joint mental states that we appeal to in explaining basic sorts of joint agency must conform to the following:

    Distribution Principle (DP): When some individuals jointly intend (believe, desire) that P, it follows that all or most of them individually intend (believe, desire) that P. (p. 3)

    An account that adheres to DP is likely to give an account of joint states that correspond to individual beliefs and desires that are, as Roelofs puts it, “unchosen, sometimes inarticulate, states which are formed unconsciously and manifested in a variety of behaviour, rather than actively and deliberately formed one-at-a-time. It is beliefs and desires like this which we primarily appeal to in our intentional explanations of individual actions, while consciously-formed states are more often the things we wish to explain.” (p. 4)

    Roelofs gives the following two interrelated definitions of joint desire and joint belief, where any adequate DP-respecting account of joint intention could be inserted:

    Two or more people jointly desire that P iff (i) each of them desires that P and (ii) they are disposed to jointly intend to A whenever they jointly believe that doing A will bring it about that P, and A is not something they can do individually.

    Two or more people jointly believe that Q iff (i) each of them believes that Q and (ii) they are disposed to jointly intend to A whenever they jointly desire that P, and Q implies that doing A will bring it about that P, and A is not something they can do individually. (p. 6)

    He then goes on to consider what is needed for a group to have the disposition required in each definition by (ii). Before presenting and discussing this part of Roelofs’s paper (in section 3 of this commentary) and his notion of quasi-agential groups (in section 4), I want to consider how his approach to joint intentional action is related to some existing work on the topic.

    2. A contrasting approach to joint intentional action

    While Roelofs never explicitly claims that we always need to posit joint beliefs and joint desires in order to explain or understand the phenomenon of joint intentional action, I think it is easy to read him as if he is suggesting this. Such a claim would follow naturally from the analogy between joint agency and individual agency that seems to underlie Roelofs’s approach. The view of individual agency in the paper seems to take agency as for all relevant purposes an atemporal phenomenon. As a consequence, the analogy suggests that for a collection of agents to perform an intentional joint action, it must at each time collectively be somewhat similar to a rational agent, and as such, it must have joint analogues of beliefs, desires and intentions. A joint analogue of the explication of “S intends to A” on p. 5 would thus suggest that a joint intention by definition is an interpersonal state that a group of people is disposed to enter when they have certain joint desires and joint means-end beliefs, so that there could not be any joint intentions without there being joint beliefs and joint desires.

    On a different approach to joint intentional action, there can arguably be joint intentional action with participants who lack joint desires and joint beliefs. This approach, in its most stark form, sees joint intentional action and individual temporally extended intentional action as two species of one general phenomenon of complex intentional action (for suggestive variations of this stark form, see Rovane 1998, ch. 4; Chant 2007; Butterfill 2015). In a more modest form, the approach at least highlights partial parallels between the way that various component actions are performed by one agent are unified in temporally extended intentional actions and the way component actions performed by several agents are unified in joint intentional actions (see e.g. Laurence 2011; Bratman 2014). On this alternative approach, the question is what unifies various actions that are components of one larger action, rather than on what glues various individual agents together into one larger (quasi-)agent.

    Consider the singular cross-temporal case: An agent who has an intention to, say, make and eat dinner, need not have a cross-temporal analogue of a joint desire (that one no longer be hungry, say). Suppose that half-way through the activity of making dinner, the agent’s hunger might dissipate (perhaps she nibbles on some biscuits while working in the kitchen), but she might nevertheless retain her intention and complete the activity, either out of mere inertia or because some other desire can replace the motivation originally provided by the desire to not be hungry, such as a desire consume certain amounts of nutrients each day or to stay healthy. Nor does this agent need to have a cross-temporal individual analogue of a joint belief (that making and eating dinner is a means to ending hunger, say), at least not one that provides her with a reason for acting for carrying on throughout the activity. After all, the relevant means-end belief that supports the continuation of the intention and action to make and eat dinner will vary across time in the case we just considered: at t1 the relevant belief was that making and eating dinner was a means to ending hunger, but at t2, the relevant belief is that making and eating dinner is a means to stay healthy.

    This alternative approach would thus suggest that, analogously, joint beliefs and joint desires are not necessary for joint intentional action. Some reflection suggests, I think, that this is entirely appropriate: You and I might form a joint intention to make and eat dinner together [A], even if there is no state of affairs P that each non-instrumentally desires nor any one belief in a means-end relation Q that figures in the practical reasoning of each of us. I might desire to stay healthy while you desire to end your hunger. Hence, my practically relevant means-end belief is that making and eating dinner together is a means to staying healthy while your practically relevant means-end belief it is a means to ending your hunger.

    In a sense, this approach makes the explanation of joint intentional action fairly straightforward. An intentional action—unless it is a fairly simple basic action such as raising one’s arm—has an instrumental structure of in-order-to relations organised around a single end. In this structure, the actions of one or several agents can fit. The structure can thus be distributed among the members of a group, as can a prior plan for such a complex action. However, the same cannot be said about belief and desire. If several agents are to jointly believe or desire something, they arguably have to create a sort of agent-like unity to which it is appropriate to attribute that joint belief or joint desire (although the creation of such a unity could involve a joint intentional action whose instrumental structure is distributed among the agents).

    I think that this approach is at least in tension with Roelofs’s view, but I am not sure whether it is inconsistent with it (beyond a terminological difference in the use of the term ‘joint intention’ (or ‘shared intention’). This is because I am not sure whether or not he thinks joint beliefs and joint desires are necessary for joint intentional action. At any rate, I hope that by contrasting Roelofs’s approach with an alternative, it will become clearer that it is an original and distinct approach.

    3. Components of ‘corporality’ and common knowledge

    I will now consider the substantive claims that Roelofs makes regarding what is necessary for several agents to jointly believe that P (desire that Q) beyond that they have an aggregate belief that P (desire that Q)—that is, beyond that each individually having the belief (desire). Roelofs arrives at these claims in a principled way by identifying what is necessary for a group to be disposed to jointly intend to A? Or to put it negatively, “what factors must be present to avoid … failures of collective action”? (p. 7) Roelofs calls these necessary factors “corporality”.

    According to Roelofs, corporality requires at least the following:

    1. common knowledge of the aggregate beliefs/desires in order to avoid systematic failure due to mutual ignorance;
    2. the ability to coordinate plans and actions in carrying out the A-ing in order to avoid systematic failure of coordination/meshing—where such an ability is cashed out in terms of physical ability and psychological willingness—and
    3. each needs to be sufficiently motivated (by whatever motives) to do their part in the A-ing in order to avoid collective action problems (i.e. to rule out free-riding or defection).

    Roelofs presents these as necessary but perhaps not sufficient conditions for corporality. Having the required features will enable an aggregate belief/desire in the group to a joint belief/desire of the group.

    I find the methodological path by which Roelofs arrives at the idea of corporality and these conditions to be congenial. But I have a reservation regarding the specific allegedly necessary condition of common knowledge. I have elsewhere argued against various arguments in favour of a common knowledge condition on joint intentional action and “shared intention” (see Blomberg 2016b). Here, I will suggest that common knowledge also is not necessary for corporality.

    Roelofs mentions that one reason why philosophers have thought common knowledge to be required for joint intention is the following: “If we act in ways that in fact accomplish a common goal, or even in ways that in fact depend on each other’s actions, we are still not acting together if we are ignorant of this fact.” (p. 8) However, absence of such ignorance is compatible with there being a failure of both common knowledge and common belief. Suppose that each believes that each intends that we go for a walk, but I falsely believe that you falsely believe that I merely intend to go for a walk, where this is neutral regarding your participation (such that the satisfaction of the intention is compatible with your participation but doesn’t require it). Here it is not “out in the open” that each intends that we go for a walk, but we may nevertheless each truly believe that we accomplish the common goal that we go for a walk. A similar point holds regarding joint desire and joint belief: Failure of common knowledge among group members that each believes that P (desires that Q) does not imply that the group members are in a situation of mutual ignorance. They may have mutual first-order true beliefs about the fact that each believes that P (desires that Q) even if they have false higher-order beliefs regarding some of these first-order beliefs or desires.

    Now, Roelofs is looking for features that remove systematic obstacles to the formation of joint intentions (see p. 7), so it might be true that common knowledge is necessary if failure of common knowledge undermines the formation of joint intentions in most situations. Roelofs seems to suggest this when he draws on a “fog of war” case where coordination fails and none of five soldiers contributes to an unanimously desired coordinated attack against an enemy bunker. The coordination failure is due to the fact that none of the soldiers expects the others to contribute, and more than one soldier must attack simultaneously for the bunker to be captured. (Assume that the consequences of a lone soldier attacking would be certain death.) Each soldier must here have a true first-order belief about the others’ intentions, and Roelofs might insist that failure of common knowledge that each believes that P (desires that Q) will generally undermine first-order beliefs that P. However, this will only be true in some very specific conditions, such as in situations where the agents involved cannot communicate with each other and where the cost of failure of coordination is high, as exemplified by the “fog of war” case. In most situations, false-higher order beliefs will quickly be revised if agents are disposed to making their beliefs, desires and intentions manifest to each other by initiating communication.

    Perhaps Roelofs would agree with this because when he more explicitly unpacks what is required for joint belief/desire, he writes:

    [P]eople jointly believe (desire) that P only if all or almost all of them believe (desire) that P, and they are both physically able and psychologically willing to share this information freely. (p. 9)

    Given Roelofs approach, this is right I think. But it doesn’t imply that common knowledge is “one element in ‘corporality’.” (p. 8) At least, it is not a necessary such element.

    I have presented (a), (b) and (c) as necessary conditions for corporality, but perhaps this is not the best way of interpreting Roelofs. At the beginning of sect. 4, on p. 12, Roelofs mentions that groups can have “more or less” of the property of corporality, and thus qualify as being quasi-agential to varying degrees. Now, one might think that this is also true of particular joint beliefs or desires, such that particular joint states can have different degrees of corporality. If that is what Roelofs has in mind, then groups can have beliefs and desires that fall on a continuum between being merely aggregative and being maximally joint. In such a scheme, common knowledge that each believes that P (desires that Q) would be a feature of a maximally joint belief that P (desire that Q), but it would not be necessary for there to be a joint state as such.

    4. Quasi-agential groups and group agents

    According to Roelofs, groups that quite generally have the features in (a), (b) and (c), and in addition have many aggregate mental states, qualify as what he calls “quasi-agential groups”. He distinguishes quasi-agential groups from group agents as follows:

    High levels of corporality make a collection of people operate in some respects like an agent, in that some but not all of the defining principles of agency hold of them collectively. […] What does not hold of them is, firstly, consistency: it is not true that they will generally have only consistent sets of (joint) beliefs or (joint) intentions, or that when they have inconsistent (joint) beliefs or (joint) intentions they will take steps to correct this inconsistency. Another way in which they are not fully agential is the lack of scope of their attitudes […]. (p. 13)

    But if we assume that DP [the Distribution Principle] requires that all members of a group believe that P (desire that Q) in order for the group to jointly believe that P (desire that Q), then there consistency will hold regarding the collection’s joint beliefs and joint intentions (insofar as they have joint beliefs and joint intentions). Consistency will hold for the quasi-agential group simply in virtue of the fact that consistency holds regarding the beliefs and intentions of each group member. Suppose that the collection of Olle, Johannes and Luke have the joint belief that P and the joint belief that not-P, then each of Olle, Johannes and Luke will hold the inconsistent set of beliefs that P and that not-P. But if I understand Roelofs correctly, he thinks that it would not be appropriate for a fourth person, Mirja, to criticise the group consisting of Olle, Johannes and Luke for having inconsistent joint beliefs. Mirja would instead have to directed her criticism to each of them individually. If this is right, then there seems to be an interesting asymmetry between the role of joint beliefs and joint desires in making sense of and explaining joint intentional action of corporeal groups on the one hand, and the role of these states in holding such groups responsible and engaging in rational criticisms of their joint actions.

    This point does not hold if DP is weakened so that only a majority of the group members need to believe that P (desire that Q) in order for the group to jointly believe that P (desire that Q) (see e.g. List & Pettit’s 2011; Hess 2013, sect. 1). With such a weakened DP, Olle, Johannes and Luke could have inconsistent joint beliefs and conflicting joint intentions, even if they each have perfectly consistent beliefs and intentions. Roelofs suggests that a weakened version of DP is more plausible for “very large collections” of people. But I’m not sure there is any reason to adopt a weaker version of DP, at least not when it comes to groups where “hierarchies of authority, or institutional rules or procedures” are absent and where individuals don’t distinguish between acting in their “capacity as group member[s]” and in their “capacity as individual[s]” (see pp. 3-4). After all, in virtue of what would an individual count as a member of the group that has a joint belief that P (intention that A) if not (in part) their individual belief that P (intention that A)?

    Roelofs also mentions that the joint attitudes of a quasi-agential group will have much less scope that than the set of attitudes of a paradigmatic agent—that is, there will be many issues regarding which the group doesn’t have any joint attitude. But does this really distinguish quasi-agential groups from groups that are typically taken as paradigm examples of group agents by philosophers such as List and Pettit and Hess? I’m not so sure. The attitudes of many committees, juries, corporations and other groups that are susceptible to the discursive dilemma arguably have a relatively small scope of attitudes too.

    I think there are still important differences between quasi-agential groups and group agents, but I’m not sure Roelofs get these differences right. One important difference seems to be that in the case of group agents, group members will often freely adjust their behaviour and attitudes in various ways according to what the group agent believes, intends and desires. This is a type of top-down influence and control that seems to be absent in Roelofs’s quasi-agential groups.

    5. Conclusion

    On Roelofs’s approach to joint intentional action, the main question is what unifies collections of people into quasi-agents so that they can be fruitfully ascribed joint mental states for purposes of prediction and explanation. This is an original and distinctive approach that I have contrasted with alternative approach that focuses on the question of what unifies several component actions into a larger (joint) intentional action. As far as I can see, there is no direct conflict between these approaches, unless Roelofs takes joint intentional actions to necessarily be performed by quasi-agents with that have joint mental states.

    With respect to the substantive constraints and conditions on what makes it true that a group of people have joint beliefs and joint desires, I have argued that Roelofs need not require there to be common knowledge among the group members that they have certain aggregate beliefs or desires (ability and willingness to share information is sufficient). I also raised an interpretative question: Are Roelofs suggesting that a group’s joint mental states fall along a continuum of jointness from the merely aggregative to the maximally joint, depending on the degree of corporality of that group, or is a more clear-cut distinction possible?

    In the final section, I argued that it is not clear that Roelofs has identified what makes quasi-agential groups different from full-blown group agents, since it is not clear why it isn’t irrational for a quasi-agential group to have inconsistent joint beliefs or joint intentions, nor is it clear that there is much of a difference with respect to the range of issues that a quasi-agential group will have joint attitudes concerning and the range of issues that a group agent will have attitudes concerning.

    Finally, I would like to thank Luke for a very stimulating paper that I thoroughly enjoyed reading and thinking about! I am very much looking forward to the coming discussion.

    Blomberg, O. (2016a). Shared intention and the doxastic single end condition. Philosophical Studies, 173(2), 351–372.
    Blomberg, O. (2016b). Common Knowledge and Reductionism about Shared Agency. Australasian Journal of Philosophy, 94(2), 315–326.
    Bratman, M. (2014) Shared Agency: A Planning Theory of Acting Together. Oxford University Press.
    Butterfill, S. A. (2015). Planning for Collective Agency. In C. Misselhorn (Ed.), Collective Agency and Cooperation in Natural and Artifical Systems (Vol. 122, pp. 149–168). Springer Verlag.
    Chant, S. R. (2006). The Special Composition Question in Action. Pacific Philosophical Quarterly, 87, 422–441.
    Hess, K. M. (2013). Missing the Forest for the Trees: The Theoretical Irrelevance of Shared Intentions. In M. Schmitz (Ed.), The Background of Social Reality: Selected Contributions from the Inaugural Meeting of ENSO (pp. 57–75). Springer.
    Laurence, B. (2011), An Anscombian approach to collective action, in A. Ford, J. Hornsby and F. Stoutland, eds, Essays on Anscombe’s Intention, Harvard University Press, chapter 10, pp. 270–296.
    List, C. and Pettit, P. (2011) Group Agency: The Possibility, Design and Status of Corporate Agents. Oxford University Press.
    Rovane, C. (1998) The Bounds of Agency: An Essay in Revisionary Metaphysics. Princeton University Press.

  6. Thanks to Johannes and Olle for their helpful comments, and to Nick, John, and Cameron for organising this event!

    I don’t have a huge amount to disagree with in the comments, so for brevity I’ll reply in bullet-point form:

    Three Points Raised by Johannes:
    – Johannes points out that I say nothing about the relationship between intention and action. Certainly, intentions don’t always result in action – sometimes I change my mind, sometimes I forget, sometimes I just can’t bring myself to do it. I think all of the same possibilities exist for people acting together (they might jointly change their minds, etc.), but in both the joint and individual cases I think it remains true that the formation of an intention is, other things equal, a good explanation for the performance of the action. A paper could be written on the complexities of that intention-to-action step, and in focusing my attention on the belief-desire-intention step I didn’t mean to suggest otherwise.

    -Johannes also points out that I say little about how to apply BDIP when there are countervailing desires. Again, I’ll plead guilty: BDIP is a deliberately simplified way of thinking about intentional explanation, which ignores the extra complexity of desires that can’t be jointly satisfied. To account for those cases (which I called ‘countervailing desires’, and which are, as Johannes points out, very common), you’d need to complicate the story – roughly, desire+belief sets lead to intention only if not over-riden by others. My aim was just to start slow, seeing how to do joint-level things for artificially simple cases before extending the approach to more complicated cases.

    -Finally, Johannes presents a case that he worries will be a counter-example to my definition of joint desire, where people will count as jointly wanting to discover an unknown prime when they intuitively shouldn’t. I don’t think my definition has that implication in the case described, for several reasons of which I’ll briefly mention 2:

    Johannes talks about the clause ‘and A is not something they can do individually’ as a third condition which can be too easily satisfied. But it’s not a third condition, it’s a qualification of the second condition. That is, the second condition on jointly desiring that P is to be disposed to jointly A whenever certain conditions are met, viz. joint belief that A-ing will secure P, and inability to A individually.

    (This means that if too many cases satisfy this ‘cannot do individually’ condition, that wouldn’t mean an over-generation of joint desires, but an under-generation, because it will expand the range of cases in which the people must be disposed to A if they are to count as having a joint desire.)

    Johannes talks about people who want to gain money, and are disposed to try and discover an unknown prime, or win the lottery, if they believe doing so will get them money: he concludes that in such cases, by my definitions, these people jointly desire to discover an unknown prime, or to win the lottery. But this misreads what is being defined. If these people would do A, were they to think that doing A would get them lots of money, for all sorts of different values of A (winning the lottery, discovering primes, building a new factory, whatever), then by my definition they jointly desire to get rich. And that seems right to me. Nothing follows about a joint desire to do any of those particular As.

    (Note that according to my footnote 3, we have to read ‘each individual wants to get rich’ as meaning ‘each individual wants the group to get rich’ – if each wants to be rich themselves, they don’t all want the same thing, and so there’s not even an aggregate desire.)

    5 Points Raised by Olle:
    -Olle points out that the way I’ve set up inter-linked definitions for belief, desire, and intention, and then inter-linked definitions of joint belief and joint desire, gives the impression that I meant to make joint intention into something that requires joint beliefs and desires. That was not at all my plan: joint intentions may arise from joint beliefs and desires, but may not.

    That is, people sometimes act together because they all, based on separate reasons, happen to have a common goal on that particular occasion. But they also sometimes act together because they all jointly want something, and jointly know how to get it. Both cases admit of individual-level explanations (in which we list the relevant psychological causes in each individual), but the second sort of case also, it seems to me, admits of a useful collective-level sort of explanation. My aim is not to say that this sort of explanation must always be possible, just that it gives us explanatory advantage we wouldn’t otherwise have.

    -More broadly, I’m actually not at all opposed to the Rovanian sort of approach that Olle lays out, treating cross-person and cross-time coordination as basically analogous. Olle seems to see a difference particularly because “If several agents are to jointly believe or desire something, they arguably have to create a sort of agent-like unity to which it is appropriate to attribute that joint belief or joint desire.” I’m not entirely sure what counts as an agent-like unity, but I didn’t intend to commit to any such thing beyond the requirements I laid out: joint belief/desire just require people to share beliefs/desires, talk about them, and be willing to work together in relevant ways.

    -Olle points out that joint action seems to be possible without the indefinitely-iterated structure that has come to be associated with the term ‘common knowledge’ (I know that you know that I know that you know… ad infinitum). I’m happy to accept this: all I insist on is that there is an epistemic condition in place of some sort – people’s shared attitudes must be ‘out in the open’, however we spell this out.

    -Finally, Olle asks whether there “is any reason to adopt such a weaker version of DP? …After all, in virtue of what would an individual count as a member of the group that has a joint belief that P… if not (in part) their individual belief that P?”

    I think the rhetorical question here has force if we proceed by first identifying some joint mental state, and then asking which group it belongs to. But we might equally go in the opposite direction, starting with some group of people (e.g. ‘the 100 people who live in this village’) and then asking what joint attitude they have. In particular, we might do that if we’ve previously identified that it’s this group that has some other joint attitude, or who have engaged in some joint action. So you might find that ‘the villagers’ jointly want more rain, and then find that 99 of them believe that there will be more rain if they do X, and in that case if they’re pretty corporal it seems reasonable to me to say ‘they want rain and they think doing X will get them rain, so they’re going to do X.’

  7. Thanks Luke for your clarifications and responses.

    I saw a tension between the Rovanian approach and your approach, and I suggested that joint beliefs and joint desires, but not joint intentions, requires an “agent-like unity” to which we attribute the joint beliefs and desires. And I also suggested that your view was that such an agent-like unity is required at least for joint beliefs and joint desires (and perhaps also for joint intentions). What I meant was simply that joint beliefs, on your view, require there to be interlinked joint desires and (potential) joint intentions, and that joint desires similarly require there to be interlinked joint beliefs and joint intentions. This is a kind of weak holism (of the jointly mental) that isn’t required, on the Rovanian approach, for joint intention.

    More intuitively, there is an instrumental structure which can be literally distributed and jointly enacted in the case of intention and action. For example, we plan that I cut the vegetables while you boil the pasta, and we then proceed to do precisely this. But in case of desire and belief, several agents must make up an appropriate target for at least a coarse-grained net of interlinked joint mental states, a kind of unity. (To be honest, I might also have attributed the idea to you that agent-like unity is required because I tied your idea of corporality with quasi-agential status too tightly. But clearly, the corporality required for joint belief/desire in a collection of people without this collection being a quasi-agential group.)

    With respect to common knowledge: I don’t see why switching to ‘out in the open’ changes anything. I assume that a false higher-order belief that not-p blocks both common knowledge that p and it being out in the open that p, and that such false higher-order beliefs is compatible with corporality——except in very special circumstances—given that there is ability and willingness to share information in the group. If you show how the ability and willingness to share information will tend to enable coordination and joint action, then this would be sufficient. I don’t see why openness is also necessary. As I mentioned though, perhaps it is best to think of corporality and the jointness of joint mental states as a graded notion, in which case openness could be present in the ideal case.

    Thanks for the answer to my rhetorical question at the end. Your contrast between the two ways in which we pick out the group is illuminating and I think it would be good if you incorporate this into the paper.

    1. Aha, I see what you mean by ‘agent-like unity’ now – it’s that on my approach joint beliefs and desires never exist just by themselves, but always in a sort of holistic relationship (at least potentially) to each other and to joint actions. I’ll cop to that – though I’m happy to allow that there are also some interesting notions in the vicinity that don’t have this feature, like List’s ‘common attitudes’. But the main thing is that I don’t think joint action requires joint beliefs and desires, so I don’t think it requires this holistic network of joint attitudes.

      As for the epistemic stuff: I suppose two people might still be disposed to work together, even if one falsely believes that the other falsely believes that the first doesn’t share the relevant beliefs and desires. So, we might both want that ugly tree pruned, and be disposed to work together to accomplish this when a means becomes known, even if I falsely think that you don’t think I care about the tree. In particular, as you point out, I might be disposed to tell you what I want when it becomes salient, and so a disposition to share information might be enough even if the information isn’t already out in the open. At the same time, there’s also a risk there – I might expect you to be reluctant to trust me, and so consider it not worth telling you that I want the tree pruned (which in fact you already know). So insofar as corporality is gradable, and probabilities will be relevant to those gradations, I think you’re right that ‘out in the open’ is the strongest case but not necessary.

      (In this connection, I believe Anne Schwenkenbecher and Olivier Roy have some work arguing that the relevant notion should be ‘sufficiently pooled information’, which might be another way to develop the epistemic component of corporality.)

Comments are closed.