Bratman, Searle, and Simplicity. A comment on Bratman, Shared Agency: A Planning Theory of Acting Together

1 Introduction

In the debate between reductionists and non-reductionists about collective action and intention, Michael Bratman’s work offers a middle position. Like Margaret Gilbert and John Searle he accepts a strong form of ‘sociality’, ‘sharedness’, ‘jointness’ or ‘collectivity’ that is substantially different from mere strategic interaction and coordination, while insisting that the step from individual to shared agency can be taken without bringing in any new conceptual or metaphysical resources.

The details about how to take this step have been carefully worked out in previous books and articles by Bratman. Besides providing a clarifying and comprehensive exposition of Bratman’s theory of collective action, his monograph Shared Agency, A Planning Theory of Acting Together (Oxford UP 2014) meets important potential objections to the position, and relates it to alternative views.

The book’s main ambition is to show that Bratman’s planning theory of individual agency gives us all the tools we need to explain what its main rivals attempt to explain about collective actions. Since it does that explanatory work without relying on the introduction of any distinctively new or irreducible entities, Ockham’ Razor will favor Bratman’s analysis of collective actions over two competing types of views – represented by John Searle’s theory of collective actions as explained by we-intentions, and Margaret Gilbert’s theory of collective actions as explicable in terms of joint commitments.

Bratman’s work is established as one of the most important philosophical approaches to group agency so far, and to me the new book confirms that impression. In the present context, it seems proper to express this kind of respect by attempting to challenge what I take to be the book’s most central claim. I will do that in two ways.

Firstly, by questioning whether there must be a fundamental difference in kind between Searle style we-intentions and I-intentions. If not, Searle’s type of theory is not less qualitatively parsimonious than Bratman’s. The issue hangs on how we understand the notions of modes and contents of intentional states, and the relations between modes, contents, and categorizations of such states.

Secondly, by questioning whether Bratman’s theory steers clear of debunking or dismissing collectivity.^[1] Does it capture jointness in the strong sense that we are after? I have argued that Bratman’s theory as presented in previous work to some extent has wavered in this respect. More specifically, I have claimed that the manoeuvres Bratman suggested to avoid circularity in his conceptual analysis (in 1992 and 1997) undermine the strength of his resulting notion of collective action. Bratman responds in detail to this objection in his new book and I return to the issue towards the end of the paper.

I should stress that the two sections of this paper are theoretically disparate. While the first can be read as a possible defence of Searlean ‘we-intentions’, the second presupposes a view close to Bratman’s concerning the nature of group actions.

2 Quantitative and Qualitative Parsimony

Consider the first of the five conditions that together are sufficient for “modest sociality” in Bratman’s analysis.

Intention condition: We each have intentions that we J; and we each intend that we J by way of each of our intentions that we J (so there is interlocking and reflexivity) and by way of relevant mutual responsiveness in sub-plan and action, and so by way of sub-plans that mesh (2014, p. 103).

On the face of it, this condition seems to require the intentions of individuals displaying shared agency to be considerably more complex and cognitively demanding than Searle’s ‘we-intentions’. In Searle’s theory, as I understand it, we-intentions are regarded as intentions that simply have a collective form or mode, without having to fulfil any extraordinary specifications of content.

I assume that Bratman would admit that his theory is less parsimonious than Searle’s in this sense, but he stresses that what matters in this context is qualitative parsimony rather than quantitative.

A doctrine is qualitatively parsimonious if it keeps down the number of fundamentally different kinds of entity. … A doctrine is quantitatively parsimonious if it keeps down the number of instances of the kinds it posits (Lewis 1973, quoted by Bratman 2014, p. 106).

Lewis thinks that qualitative parsimony “is good in philosophical or empirical hypotheses” but he recognizes “no presumption whatever in favor of quantitative parsimony” (quoted by Bratman 2014, p. 106). Let us agree at least that qualitative parsimony have lexical priority over quantitative parsimony in adjudicating between philosophical hypotheses that have the same explanatory scope.

In order to accommodate genuinely collective actions Searle and Gilbert introduce new and irreducible kinds of entities into their philosophies of action. They explicitly deny that the key concepts (‘we-intention’ and ‘joint commitment’ respectively) denoting these phenomena can be analysed further, in terms of non-collectivistic concepts. Bratman’s explanations of collective actions merely utilize the conceptual resources that are already available within his theory of ordinary individual agency. So, it appears that Lewis’ version of Ockham’s razor would favour Bratman’s position.

3 Two notions of ‘Qualitative Difference’

Bratman contrasts mere differences in content of attitudes with fundamental differences in kinds of attitudes.

“The distinction is not between two fundamentally different attitudes, but between two different kinds of contents of the attitude of intending, an attitude described by the planning theory (2014, p. 14)

It is not completely clear to me how Bratman regards the relation between two distinctions concerning attitudes in this and other passages in the book.

The first is the distinction (from Husserl, Frege and others) between the quality, form, or mode of an intentional state, and the content, or “matter”, of that state (Husserl 1900, p. 143–144). Typically, Husserl ties differences in quality to differences in kind – the act of hoping that p is another kind of attitude than the act of doubting that p in virtue of being of a different quality. However, as François Recanati and others have argued, other sorts of qualitative differences – differences in modes, or “perspectives” of believing, for instance – need not involve different kinds of attitudes. In this sort of framework, qualitative differences may or may not imply differences in kind.

The second distinction is the one that Lewis makes use of in the passage about a theory’s parsimony, between quality and quantity. In this sense, ‘qualitative difference’ simply means difference in kind or type.

4 Understanding We-Intentions (Beyond Searle)

So, must we-intentions be of a fundamentally different kind than ordinary individual intentions? Here is a condensed five-step suggestion about how to understand we-intentions in a broadly functional framework.

(1) There is a meaningful distinction between mode and content of intentional states.

Consider your perceiving a flower. As François Recanati points out, what is perceived is the flower, not the fact that the flower causes your experience. However, a condition for your experience to count as a perception (rather than as a memory, for instance) is that there is a flower and that it causes your experience. Perceptions must fulfil a self-referential condition that can be described in functional terms. So, there are essential features of perceiving that do not belong to the content of perceptions. We can say that these features belong to the mode of perceiving (Recanati 2009, p. 131–132).

(2) Modes determine contexts of evaluation.

For your perception of a flower to be veridical, the flower must be present at the time and place of perceiving. The mode of perceiving is essentially such that it it ties the object of the intentional state to its bearer at the time and place of bearing it. In some cases, like memory, the context of evaluation – that is, the context in which the content of the memory should be evaluated – cannot coincide with the time in which the mental state is present. Then, the intentional content “is presented as true with respect to the situation (and the time) of the earlier perceptual experience” (Recanati 2009, p. 141).

(3) Our general conceptual constraints on kinds of intentional states may leave room for variations in the modes of some kinds of states.

Consider my act of asserting that it is raining. In the most common situation, you would assume that I am tacitly referring to the place I am at when I say this. You would regard it as true if it is raining here and now. You could be wrong about that interpretation, though. Maybe I had just been watching the news about a flooded area far away, and I actually tried to inform you about what happens there. So, the full meaning of my utterance is richer than its content and the asserting of that content. It comes with tacit “perspectival” information about time and place for evaluation (Recanati 2007).

In a similar manner, the proper context of evaluating the content of my corresponding belief that it is raining may vary even though the intentional content of that belief is just ‘it is raining’. My mode of believing determines the context of evaluation for that belief. In that sense, the belief’s content can be conceived from different perspectives.

(4) The ‘subject of intention’ of an intentional state should be distinguished from the intentional subject, the individual in whose head the intentional state resides.

For your representation of a flower to count as a perception of that flower, it must be the case that there is a flower related to you in a specific way. Perceptions are self-referential and therefore also subject- or agent-referential. In that sense, perceptions always have a subject of intention, which is a feature of the mode of perceiving.

Other intentional states, like beliefs, need not have any subject of intention, since they are not essentially self-referential. The truth conditions for a belief need not refer to the believer. So, it is conceptually possible to separate the bearer of the attitude, its intentional subject (Mathiesen) or ontological subject (Tuomela) from the subject of intention. ^[2]

(5) We-intentions are intentions held in a collective mode, or from a collective perspective. The subject of intention of a we-intention is the collective.

Action-intentions, like perceptions, are essentially self-referential and therefore have a subject of intention. Unlike my mere desire to see the page turned, my intention to turn the page is not successful unless I perform the action of turning the page. This does not imply that the ‘I’ figures in the intentional content of the intention. The subject of intention is a perspectival feature of the mode of some types of intentional states, like perceptions and action-intentions. That mode determines the context of evaluation. ^[3]

Our general conceptual constraints on success conditions admit that the context of evaluation comes apart in time or space from the actual context in which the intentional state occurs (for most intentional states except perceptions). In that sense, the spatial and temporal perspective of an intentional state – the point in time and space from which its content is conceived – need not be “here and now”.

There are no formal reasons to exclude the possibility that the context of evaluation may come apart from the actual context when it comes to agent perspectives as well. So, there is a plausible way of understanding “subject of intention” in terms of the perspective, or mode, in which the intention is held. Moreover, there is no contradiction in assuming that an intention in the head of an individual can be in the we-mode. ^[4] That is, you and I can have intentions from the group’s perspective.

In this framework, a we-intention has the same kind of functional role as an I-intention; it is successful when the same kind of relations between the agent, i.e., the unit of activity determined by the mode of intending, and the intended event obtains. My intention to turn the page is successful if (I perform the action of turning the page by way of carrying out this intention). Our we-intentions to smash the shop window are successful if (we perform the action of smashing the window by way of carrying out this intention).

It is reasonable to assume that there must at some stage be some form of communication and joint awareness, enabling us to form similar we-intentions, coordinate our actions etc. However, those conditions concern the genealogy and the consequences of we-intentions. They need not enter the functional characterisation of the we-intention “in the head of the individual”.

The ambition of this kind of theory is to characterize the we-mode of intending functionally, in terms of success conditions (admittedly very sketchy in this paper). Some have intuitions about elusive experiential differences between intending individually and intending collectively – a “sense of we-ness”, if you wish. The present account does not rely on such intuitions but is quite compatible with them. “[T]here is absolutely no reason to consider that phenomenology supervenes on content in the narrow sense” (Recanati 2009, p. 151).

So, my claim is that (1) – (5) is a way of understanding we-intentions that does not introduce a fundamentally different kind of entity to make room for collective intentions. It utilizes a conceptual framework that we need to understand intentionality in general, individual and collective. If that claim is plausible, we cannot adjudicate between the two types of theories merely on the basis of qualitative parsimony.

5 A possible Objection on Behalf of Bratman

Searle identifies full descriptions of the content of intentional states with descriptions of their success conditions (Searle 1983, p. 85). It could be argued, therefore, that the distinction between mode (understood in terms of success conditions) and content will collapse within Searle’s general theory. That is correct as far as I can see, but the position outlined above is an elaboration of Searle’s concept of we-intentions, not of his general theory of intentionality in (1983).

Bratman could claim, for other reasons, that everything that can be characterized in terms of functional role or in terms of success conditions is covered by the notion of content. (He wants to remain neutral in the debates about the nature of content though (12)). However, such a broad notion of content would mean that mere differences in content sometimes create different kinds of intentional states – unless we want to deny that perceptions, intentions, beliefs etc. are different kinds of intentional states.

6 A Note on Demandingness

It has been argued that Bratman’s theory of collectivity is too demanding. Four-year olds seem to exhibit modest sociality but can four-year olds have intentions that fulfil Bratman’s intention condition? Bratman responds that given the purportedly primitive status of Searlean we-intentions or Gilbertian joint commitments, it is difficult to know how psychologically demanding these phenomena are (104). That response seems to me correct.

However, if we can go beyond Searle and understand we-intentions in terms of functional roles and success conditions, thereby placing them within the same theoretical framework as other intentions, we might get a better idea of which psychological capacities they require. The place for reflexivity seems crucial to me. While Bratman’s theory requires the parties in a collective action to have the conceptual resources for a certain kind of self-reflexion, the view outlined above does not.

7 Explanatory Scope

In (1992) Bratman attempted to avoid a circularity challenge to his account by explicating ‘we J’ (as it appears in the content of participants’ intentions according to the intention condition) by appealing to act descriptions “characterized in co-operatively neutral ways. The act description merely has to satisfy the “behavioural conditions” for joint activities (1992, p. 97).

Bratman considers the following objection.

Suppose I want the window smashed. When I note your presence on the street, I think that if you act in a certain way, the window can be smashed as a result of both our acts, and I form an intention accordingly. What I intend in that case is merely to get the window smashed, while predicting that your actions will be components in the process leading to that result. This prediction may rest upon my knowledge that your intentions are similar to mine, and that our subplans are likely to mesh in a way that enables me to reach my goal. There is mutuality and interdependence, in line with Bratman’s requirements. Still, I would say, nothing in this picture captures “sharedness” or “collectivity” in any sense distinct from what we can construe in terms of standard individualistic theory of action (Petersson 2007, p. 140–141).

“Petersson thinks that a theory like mine will be led to say that the case he describes is a case of “sharedness”; and Petersson thinks (correctly, in my view) that this would be a mistake” (93).

As Bratman points out, I failed to make it clear in my example “that I intend our joint window smashing in part by way of your intention” (93). To fulfil the intention condition of Bratman’s analysis, one would have to make it explicit that after realizing that your intentions are similar to mine, then I come to intend (and not only predict) that you will act and intend in the ways needed.

Bratman thinks that for this modified version of the example to pose a problem for the basic theory it is important that all the conditions of his analysis are fulfilled: Not only that I intend in the way required by the intention condition, but also that each of us

is appropriately responsive to the other; each is set not to thwart the other and is at least minimally disposed to help the other if needed; there is (feasibility-based) persistence interdependence in relevant intention; there is, as a result of all this, rational pressure in the direction of social rationality; each expects all of this to issue in the joint action each intends; and all this is out in the open (2014, p. 95).

Bratman argues that once all this is in place, the example will capture an acceptably strong version of collectivity. I guess my intuition on this would depend on further specifications of the nature of the “helpfulness” dispositions etc. Furthermore, I am unsure about what such an intuition would tell us about the nature of shared intention. The fulfilment of these conditions would be a consequence of, and presuppose, the agents having formed the right kind of intentions to begin with. To assume about the window-smashers that “each expects all of this to issue in the joint action each intends” I must assume that each intends the joint action. In doing so, I have to apply my intuitive view of what is required for someone to ‘intend a joint action’.

My positive suggestion in 2007 was that Bratman could employ the notion of a group as a unified ‘causal agent’, explicated in terms of dispositions or causal powers. I suggested that ‘we J’ in Bratman’s intention condition could be understood in such terms, thereby avoiding circularity and still making room for a stronger sense of sharedness than the one offered in Bratman’s basic model. Bratman endorses this concept and agrees that we can regard groups of humans as causal units of activity. He shows how his theory can accommodate groups of this kind. “But it does not follow that the contents of their intentions in the basic case must include a further, primitive idea of a group causal agent” (122).

Bratman seems to resist this move mainly because he wants an account that is “conceptually conservative” and thinks that my suggestion would force him to go “somewhat beyond the conceptual resources” of his basic theory (123).

However, while the concept of a collective causal agent leaves room for a form of collective activity that may be ‘primitive’ as opposed to ‘sophisticated’, the concept is not primitive (as opposed to analysable), but explicable in terms of dispositions and causal powers, and defensible on grounds that are independent from questions about collective intentional action. I believe that Bratman could have adopted this understanding of ‘we J’ without employing conceptual resources that are not already available for him.

How substantial is this disagreement? We both believe that there are sets of actions fulfilling the conditions of Bratman’s analysis. We agree, also, that there exist sets of actions in which the participants conceive of their intended activity in terms of group causal agency – they think of their group as the “unit of activity” or “as a body”. Shared Agency, A Planning Theory of Acting Together makes it clear that Bratman’s intention condition is one of five conditions that together are sufficient for shared agency, not necessary (103). So, both ways of understanding ‘We J’ as it appears in the content of the participant’s intention, could be parts of slightly different but fully compatible sets of sufficient conditions for two forms of group agency. The question of how to label these forms would be a matter of linguistic intuitions about terms like “shared”, “joint”, “collective”, etc.

I do not find the difference between these two ways of understanding the intention condition insignificant, though. It could e.g. make a difference when we apply the analyses to some of the moral or legal issues that provide part of the motivation for getting a clearer view of the nature of group actions. Assume, as most of us do, that the content of a person’s intentions is important when we want to know what she is morally responsible for. Then it might matter whether she intended that the group as a unified agent should J, or whether she intended that a set of agents to which she belongs should J, albeit via an intricate pattern of relations between the intentions of the involved. I would argue, and have partly done so in (Petersson 2008), that in the first case we may need to go beyond our ordinary notions of individual responsibility and apply a distinct notion of co-responsibility – a “top down” notion – and hold her to account for the group’s action as a whole due to her fulfilling criteria of membership in the relevant group. In the second case, we can at least in principle disentangle the network of relations and apply an ordinary notion of individual responsibility.