Minimalism and Maximalism in the Study of Shared Intentional Action

1. Introduction

In recent years, a number of philosophers have challenged the scope and adequacy of established philosophical accounts of shared intentional action ¹ by arguing that these accounts cannot be used to account for the shared activities of young human infants, or other cognitively unsophisticated agents (Butterfill 2012; Michael 2011; Pacherie 2013; Tollefsen 2005). Such “minimalist” philosophers of joint action have often supported their arguments by appeal to developmental research, according to which young human infants do not have the meta-representational capacities that are needed for framing shared intentions (Wellman, Cross, and Watson 2001; Wimmer and Perner 1983). However, human infants do seem to engage in joint activities at an early age (Moll and Tomasello 2007; Warneken et al. 2006). This is how Stephen Butterfill (2012, 23) presents the minimalist challenge to established philosophical accounts of shared intentional action:

. . . Given the premise that joint action plays some role in explaining how humans come to understand minds, what could joint action be? Not what a leading account, Michael Bratman’s, says it is. For on that account engaging in joint action involves sharing intentions and sharing intentions requires much of the understanding of minds whose development is supposed to be explained by appeal to joint action . . .

Butterfill (2012) presents here a special version of the developmental challenge, according to which joint action plays a formative role in the development of the very capacities, which were supposed to explain joint action (Moll and Tomasello 2007; Tomasello and Rakoczy 2003). ² However, this is not a necessary feature of the minimalist challenge against established philosophical accounts of shared intentional action. Deborah Tollefsen (2005) ³ and Elizabeth Pacherie (2013) ⁴ present more modest versions of the developmental challenge, which do not presuppose that shared intentional activities play a formative role in the development of the capacities that are needed for framing shared intentions. Rather, they simply point out that infants seem to engage in rudimentary shared activities prior to the development of the capacities that established philosophical accounts of shared intention take to explain their occurrence. It is this more modest version of the minimalist challenge that I will primarily concern myself within this article.

I will follow Butterfill’s lead by concentrating on the challenge that the developmental research poses to Michael Bratman’s account of shared intentional action, in particular. As Butterfill (2012, 22) points out, Bratman’s account has been the “most influential in psychology,” and “no decisive objection to . . . his account . . . has yet been published,” Bratman’s account is also one of the least demanding established philosophical accounts of shared intentional action in both conceptual and metaphysical terms. By contrast to many other “maximalist” accounts of shared intentional action, Bratman’s account does not assume that the involved individual agents possess the concept of a group agent (Tuomela 2013), or that there are special kinds of we-intentions (Searle 1990), or irreducible joint commitments (Gilbert 2009). We can accordingly take the minimalist challenge against Bratman’s account as a proxy for the kinds of problems that a whole range of maximalist accounts of shared intentional action are likely to encounter.

I take the terminology of minimalism and maximalism from Pacherie (2011). She presents a version of the minimalist approach to joint action, which draws on game-theoretic accounts of team reasoning and social-psychological approaches to group identification. By contrast, Butterfill’s (2012) minimalist account of joint action is based on a distinction between intentions as internal states of agents, and goals as relations between agents and objects or outcomes. Tollefsen (2005) emphasizes the belief component in shared intentional action, and replaces the usual requirement of common knowledge with the less demanding notion of joint attention, in addition to replacing prior intentions with intentions-in-action. Michael (2011) introduces entirely new components to received philosophical accounts of shared intentional action by bringing in minimally construed shared emotions. ⁵ I will focus on the critical points that these minimalist philosophers make in relation to established philosophical accounts of shared intentional action, rather than on their own idiosyncratic features. I will also discuss recent work in cognitive science and psychology that shares a minimalist orientation (Knoblich, Butterfill, and Sebanz 2010; Vesper et al. 2010).

In the next section, I discuss Michael Bratman’s maximalist account of shared intentional action. In the third section, I review the developmental research that has been taken to undermine the adequacy and scope of Bratman’s account in the context of the joint activities of young infants. I also discuss some minimalist accounts of joint action that have been introduced to make up for its shortcomings. In the fourth section, I argue that these accounts should be understood as elucidations of a different kind of joint action from the kind of joint action that sets the paradigm for Bratman’s analysis. In the fifth section, I turn my attention to a minimalist account of joint action in cognitive science. In the sixth section, I accommodate minimalist and maximalist approaches to joint action under a framework of decomposition and localization of shared intentional action into component parts and operations. I also digress briefly on the existential risks that increasing mechanistic understanding of joint action poses for contemporary philosophical accounts of shared intentional action.

2. Bratman’s Account of Shared Intentional Action

Michael Bratman (1999, 2014) analyzes shared intentional action with the aid of a three-pronged strategy. First, drawing an analogy with individual intentional action, Bratman assumes that shared intentional action is a kind of shared activity that is guided and controlled by shared intention. Second, Bratman asks what functional roles shared intention serves in bringing about and structuring shared intentional action. Bratman (1993, 99) says that shared intention “supports coordination of our intentional activities in the pursuit of J, supports coordination of our planning, and (it) structures relevant bargaining,” where J is understood as a joint action type. Third, Bratman asks what complex of states could perform the functions of shared intention. Bratman (1993, 106) comes up with the following account:

1. (a) I intend that we J and (b) you intend that we J

2. I intend that we J in accordance with and because of la, lb, and meshing subplans of la and lb; you intend that we J in accordance with and because of la, lb, and meshing subplans of la and lb

3. 1 and 2 are common knowledge between us.

Bratman argues that the above complex of states can perform the functions of shared intention in the context of what he now calls modest sociality, that is, small-scale instances of shared intentional activity in the absence of asymmetric authority relations. Examples of modest sociality include painting a house together, dancing a tango together, and taking a walk together with a friend. On one hand, modest sociality can be distinguished from less demanding cases of accidentally coordinated activity, such as two strangers walking alongside one another on Fifth Avenue. While they may skillfully coordinate their movements so as to avoid collision in the constant flow of pedestrians, they do not share an intention to walk together. On the other hand, modest sociality can be distinguished from more demanding cases of institutionalized activity in organized groups, such as the Philosophy Department adopting a particular admissions policy. These kinds of cases bring in additional complications, which need to be addressed separately.

The most important features of Bratman’s account of shared intention are the following. First, the requirement of semantically interlocking intentions guarantees that our intentions and plans are responsive to what each intends (Bratman 2014, 48-43). Accordingly, my intending that we go to Chicago together by throwing you in the trunk of my car would not count as a shared intention, because our going to Chicago would not occur in accordance with and because of each of our intentions. Second, the requirement of common knowledge serves as a crucial rationality precondition for each of our intentions that we act together (Bratman 2014, 57-60). If I did not know that you intend that we go to Chicago together, I could not be confident that you will perform your part, or settle on appropriate means-actions that are compatible with what you intend. Such uncertainty would undermine our shared intention to act together. Of course, many of the subplans by means of which shared intention is to be carried out will often remain partial and incomplete at the time when shared intention is formed. For example, we might need to decide whether to take the train or fly to Chicago. Such details can be settled in the course of shared deliberation and practical reasoning, which shared intention supports.

Bratman has subsequently expanded his account by adding a number of further conditions to the original account of shared intention that was cited above. For example, Bratman (2014, 84) now explicitly requires that the persistence of the intentions of each individual must be dependent on the persistence of the intentions of the other agents. Furthermore, Bratman (2014, 56) requires that the individuals have a disposition to help each other in the performance of their parts, if needed. Many of these additional conditions can be regarded as entailments of conditions that were already present in his original account of shared intention. After all, if I intend that we J in accordance with and because of each of our intentions that we J, the satisfaction of my intention requires that I see to it that the satisfaction conditions for your intention obtain. If I did not believe that you are in a position to carry out your intention that we J, my own intention that we act in accordance with each of our intentions that we J would be thwarted. Other things equal, I should accordingly be disposed to help you in the performance of your part, when needed. For our purposes, it is most appropriate to concentrate on Bratman’s original account of shared intention, which has the virtues of clarity and succinctness, and has had the most substantial impact on the subsequent literature. ⁶

3. A Developmental Challenge Against Bratman’s Account

For many minimalist philosophers of joint action, the most troublesome feature of Bratman’s account of shared intentional action is its reliance on a robust capacity for meta-representation. In Bratman’s account, the individuals who intend to act together have beliefs about beliefs and intentions about intentions. Given that these requirements are mutual, this requires a higher order theory of mind. Although the intention condition in Bratman’s account has given rise to many conceptual debates about the conditions under which one can rationally intend someone else’s activities, ⁷ the minimalists have often criticized Bratman’s account on at least partly empirical grounds, rather than on the basis of conceptual considerations alone. This provides them with a more objective vantage point for demonstrating the limitations of Bratman’s account than could be had by stipulative conceptual conditions for defining intentions as psychological states.

A wellspring of empirical evidence that minimalist philosophers of joint action have harnessed in order to illustrate the limitations of Bratman’s account stems from research in developmental psychology concerning infants’ mindreading abilities and the development of a theory of mind. Theory of mind refers in this context to an understanding of the internal psychological states of individuals as the causes of their intentional actions, whereas mindreading refers to the capacity to infer psychological states on the basis of observable behavior. According to standard interpretations of research in developmental psychology, young human infants have demonstrable deficits in both areas of psychological competence. Most importantly, infants do not seem to have a full understanding of or capacity to reason on the basis of ascriptions of propositional attitudes, such as beliefs (Wellman, Cross, and Watson 2001; Wimmer and Perner 1983) and intentions (Astington 1991; Astington and Gopnik 1991) until around the fourth year of their lives. Although these studies have been the focus of renewed debate in recent years (Apperly and Butterfill 2009; Baillargeon, Scott, and He 2010), I follow the minimalists in taking the received interpretation of these results largely as a given.

Several minimalist philosophers of joint action have argued that the developmental data pose significant challenges for the adequacy and scope of Bratman’s account of shared intentional action (Butterfill 2012; Pacherie 2013; Tollefsen 2005). This is because infants seem to engage in rudimentary joint activities as early as 18 to 24 months of age (Warneken, Chen, and Tomasello 2006). The kinds of joint activities that young infants engage in include collaborative problem-solving activities, such as retrieving toys from inside a cylinder with a partner, and social games, such as jointly making a wooden block bounce on a trampoline (Warneken, Chen, and Tomasello 2006). To make matters worse, infants even seem to have a rudimentary understanding of the normative structure of joint activities as early as the second or third year of their lives. This is indicated by their attempts to reengage (Gräfenhain et al. 2009; Warneken, Chen, and Tomasello 2006; Warneken, Gräfenhain, and Tomasello 2012) and to help (Warneken and Tomasello 2009) their partners in collaborative activities. This indicates some understanding of the joint activity as something that the participants are doing together, rather than in parallel, or in the pursuit of their individual goals.

Given the above understanding of the cooperative dispositions of young infants, many minimalist philosophers of joint action have argued that Bratman’s account could not possibly be the whole story about shared intentional action (Butterfill 2012; Pacherie 2013; Tollefsen 2005). After all, Bratman requires that the involved individuals have semantically interlocking beliefs and intentions about the beliefs and intentions of other agents. Moreover, these higher order beliefs and intentions are construed as propositional attitudes, which involve contents that are embedded under that-clauses in conditions 1 and 2 of the account. Given that young infants do not have the cognitive capacities that are needed for forming such higher order propositional attitudes, a less demanding account of joint action seems to be needed in order to account for their rudimentary joint activities.

Beyond these critical points, minimalist philosophers of joint action have diverged from one another in at least two respects. First, they have diverged in their positive assessment of what cognitive capacities infants have at their disposal. Second, they have diverged in their attempts to construe an alternative account of joint action, which is consistent with these capacities. For example, Tollefsen (2005) submits that while infants do not have an understanding of propositional attitudes, they do have the capacities for social referencing and intention reading. Accordingly, she replaces condition 3 of Bratman’s account with the requirement of joint attention, while maintaining conditions 1 and 2 largely intact (Tollefsen 2005, 93). She also suggests that the intentions in her modification of Bratman’s account may need to be construed as “intentions-in-action” (Searle 1983) rather than as prior intentions (Bratman 1987). Butterfill (2012) reviews evidence that infants are able to understand goals as relations between agents and objects or outcomes (Csibra 2008; Gergely et al. 1995; Woodward 1998; Woodward and Sommerville 2000), and to track beliefs (Baillargeon, Scott, and He 2010; Kovács, Téglás, and Endress 2010; Onishi and Baillargeon 2005; Surian, Caldi, and Sperber 2007), even if they do not have a fully developed understanding of propositional attitudes. He comes up with the following conditions for minimal joint action (Butterfill 2012, 40):

a) one goal, two or more agents—there is a single goal, G, to which each agent’s actions are, or will be, individually directed;

b) identification—each agent can identify each of the other agents in a way that doesn’t depend on knowledge of the goal or actions directed to it;

c) expectations about goal-directed actions—on balance each agent expects each of the other agents she can identify to perform an action directed to the goal; and

d) expectations about a common effect—on balance each agent expects this goal to occur as a common effect of all of their actions directed to the goal, her own and the others’. ⁸

Pacherie (2013) makes the ideologically most complete revision of Bratman’s account of shared intentional action. She replaces its central clauses with conditions drawn from game theory (Bacharach 1999; Sugden 1993) and social psychology (Brewer and Gardner 1996) in order to provide an account of what she calls “shared intention lite”, involving team reasoning, framing, and group identification. A frame is a set of descriptions that are used for thinking about a certain situation. For example, we might think of a glass as half full or half empty. When a group frame is sufficiently salient, the individuals are disposed to identify with the group, and to reason from the point of view of what is best for the group, rather than from their own individual points of view. Pacherie’s (2013, 1833) complete account of shared intention is as follows:

“Two persons P1 and P2 share an intention to A, if:

(i) each has a self-conception as a member of the team T, consisting of P1 and P2 (collective self-framing);

(i’) each believes (i) (group identification expectation);

(ii) each reasons that A is the best choice of action for the team (team reasoning from a group viewpoint); and

(iii) each therefore intends to do his part of A (team reasoning from an individual viewpoint).”

Pacherie (2013, 1833-34) considers it to be an important merit of her account that it requires neither prior communication nor common knowledge of team reasoning. She says that her account spells out the implicit sense of collectivity that Searle (1990) argued to be a pretheoretical constituent of shared intentional action. However, her account involves certain controversial assumptions, which make its potential for accounting for the shared activities of young infants difficult to evaluate. Most importantly, researchers still disagree about whether team reasoning is a robust psychological phenomenon (Colman, Pulford, and Rose 2008a, 2008b; Krueger 2008; Sugden 2008; van Lange 2008). Moreover, it is far from obvious how demanding the cognitive capacities that are needed for team reasoning are. By contrast, current evidence in favor of infants’ understanding of goals and their ability to track belief-like states can be regarded as relatively robust due to the large number of experimental manipulations and replications that have been made in recent years. This puts Butterfill’s (2012) minimalist account of joint action on the most secure empirical footing, while Pacherie’s (2013) account holds considerable prospects for future research.

4. Different Kinds of Joint Action

In their introduction to a special journal issue on the nature of joint action, Butterfill and Sebanz (2011, 145) distinguish two broad minimalist strategies for breaking tradition with established philosophical accounts of shared intentional action. One of them is to sever the link between joint action and shared intention. This involves arguing that while infants and other cognitively unsophisticated agents may engage in shared intentional action, their doing so is not explained by shared intention. The other strategy is to formulate a less demanding version of shared intention, which is consistent with the cognitive capacities that infants have at their disposal. Butterfill (2012) presents himself as a candidate of the former strategy, whereas Pacherie (2013) and Tollefsen (2005) are advocates of the latter approach. However, although the distinction between two minimalist strategies for studying joint action is expressive of the terminological choices that many minimalist philosophers of joint action have made, it is misleading in another respect, namely, it diverts attention away from the fact that most minimalist accounts of joint action cannot be regarded as analyses of the same kind of joint action as that which is the target of Bratman’s account, whether they involve appeal to some version of shared intention or not. This conclusion can be supported on both conceptual and empirical grounds.

To start with the conceptual claim, remember that Bratman assumes that there is a tight conceptual connection between shared intention and shared intentional action. By hypothesis, then, joint action that does not involve shared intention cannot be regarded as the same kind of joint action as that which is explained by shared intention. Butterfill (2012, 32) grants as much, and describes his account as elucidating a different species of joint action. What about minimalist accounts that involve a minimally construed version of shared intention? Well, Bratman (1999, 2014) hardly claims that there is only one kind of shared intention. ⁹ Rather, he understands shared intention as a placeholder for any complex of psychological states, which is in a position to perform the functional roles that he associates with shared intention. These roles include the coordination of shared activities in the pursuit of a common goal, the coordination of mutual planning, and the structuring of bargaining about means-actions. However, it is hard to see how the shared intentions that young infants participate in could coordinate mutual planning and bargaining, if they do not have developed meta-representational content. Insofar as they do not perform the same functional roles, then as a a matter of definition, the kinds of joint activities that they give rise to will not count as the kinds of shared intentional activities that are the target of Bratman’s analysis.

There are also empirical grounds for questioning whether the kinds of shared activities that young infants engage in are sufficiently robust to count as shared intentional action in Bratman’s sense. Brownell (2011) presents evidence that the joint actions of young infants are much more dependent on scaffolding by their adult partners and sensitive to environmental disturbances than those of adult-to-adult pairs. Brownell and Brown (1992) point out that joint actions between infant pairs take longer to emerge on an ontogenetic timescale than joint actions, in which at least one participant is an adult. It seems possible that young infants construe shared intentional activities in a manner that does not represent all aspects of their collaborative nature (Bernier, Spelke, and Skerry 2013). If this is the case, then we would once again have reason to say that the infants really are engaged in a different kind of joint action, rather than that they are simply party to a different kind of shared intention.

If this description of the situation is correct, and the minimalists have indeed concerned themselves with a different kind of joint action, then their challenge against the adequacy of Bratman’s account can easily be contained within favorable limits. After all, Bratman (1999, 2014) hardly claims that he has analyzed the only kind of joint action that is possible. We can accordingly say that minimalist philosophers of joint action have pursued the complementarity strategy in relation to established philosophical accounts of shared intentional action. Instead of challenging established philosophical accounts of shared intentional action within their paradigmatic domain of fully socialized and mature human agents, they have identified a hitherto unexplored domain for the philosophical study of joint action. This new domain poses distinctive conceptual and explanatory challenges, which the complementarists have attempted to come to terms with in their theorizing. As I will show in the next section, the complementarity strategy is by no means the only possible approach that minimalistically oriented researchers of joint action can adopt.

5. A Minimal Architecture for Joint Action

In a recent article in Neural Networks, a team of cognitive scientists and philosophers posed the following question:

What kinds of processes and representations make performing joint actions possible? As a step towards answering this question, we propose a minimal architecture that could support joint action. This architecture consists of representations, processes and what we call coordination smoothers; together, these support basic forms of joint action. The minimal architecture does not attempt to define what joint action is and can be made compatible with various definitions. Instead of defining joint action, the minimal architecture aims at specifying building blocks that make performing joint actions possible. (Vesper et al. 2010, 998)

The researchers describe their minimal architecture for joint action as forging a way between dynamical systems approaches to emergent coordination (Marsh et al. 2009; Schmidt and Richardson 2008) and approaches to shared intentional action that draw on propositional attitude ascription (Bratman 1993; Tuomela 2013), language use (Clark 1996), and commitments (Gilbert 2009). Dynamical systems approaches construe coordination as an outcome of more general coordination principles, including perception–actions links, which do not require explicit representations in order to generate synchronized behavior. For example, people who are engaged in a conversation with one another have been shown to unintentionally synchronize their patterns of speech (Shockley, Sabadini, and Fowler 2004), mannerisms (Chartrand and Bargh 1999), and body sways (Shockley, Santana, and Fowler 2003; Shockley et al. 2007) with one another. In contrast to dynamical systems approaches, the minimal architecture for joint action assumes the existence of dedicated mechanisms for joint action. And by contrast to standard philosophical accounts of shared intentional action, it focuses on the mechanisms that make performing joint actions possible, rather than on the stages of decision making and planning that tend to precede shared activities.

The minimal architecture for joint action consists of representations, processes, and coordination smoothers. Joint action can include representations of one’s own task, of the other person’s task, and/or of their common goal. However, joint action need not involve all three kinds of representations, and each party to the joint action need not hold the same kinds of representations as the others. When dancing the tango, the leader might represent both her own body movements and the body movements of her partner, whereas the follower represents only her own movements. Joint action normally involves both predictive processes, which generate expectations of how the joint action will unfold over time, and monitoring processes, which keep track of the constant unfolding of events to adjust to flaws in performance. Coordination smoothers are devices for facilitating coordination by constraining the space of action alternatives, or making certain alternatives more salient than others. For example, musicians who are playing a piano duet in conditions of reduced auditory feedback might emphasize certain finger movements to provide cues about tempo and timing to their partners (Goebl and Palmer 2009).

The contribution that the minimal architecture for joint action makes in relation to established philosophical accounts of shared intentional action can be illustrated as follows. Assume first that the entire complex of interlocking attitudes that Bratman’s account postulates is in place. Still, individuals need to find some way to coordinate their physical activities when their plan comes to fruition, and the time for acting is now. The minimal architecture for joint action fills this gap, which is unattended by established philosophical accounts of shared intentional action. We can accordingly say that it pursues the constitution strategy in relation to established philosophical accounts of shared intentional action. By using this terminology, I do not mean to pick out the heavy-duty metaphysical sense of constitution in which, for example, a lump of clay can be said to constitute a statue of Julius Caesar (cf. Baker 2007). Rather, the relevant sense of constitution is that of exposing the mechanisms, which make a particular phenomenon occur. A mechanism can here be understood in the sense of Bechtel and Abrahamsen (2005, 423) as a “structure performing a function in virtue of its component parts, component operations, and their organization.” The orchestration of the parts and operations can be said to constitute a given phenomenon of interest.

What is the relation between the minimal architecture for joint action and the minimal accounts of joint action that were discussed in the previous section? We can begin by noting that the mechanisms that the minimal architecture for joint action identifies might play a more complete or exhaustive role in minimally construed joint actions, than in the joint actions of fully socialized and mature agents. Additional mechanisms may need to be brought into account for the latter kinds of cases, which fix the paradigm for Bratman’s account. However, the minimal architecture for joint action is in itself entirely neutral between minimalist and maximalist accounts of shared intentional action. Indeed, the researchers make a point of emphasizing that their goal is not to provide a particular definition of joint action, but to individuate building blocks that are compatible with various different functional definitions. This bottom-up approach sets them apart from the complementarists, as we will see in more detail in the next section.

6. A Philosophy of Science Perspective

I have above distinguished two minimalist approaches to the study of joint action, which involve different assumptions about the kinds of constraints that are faced by established philosophical accounts of shared intentional action, and are used to motivate different positive accounts. The “complementarists” argue that established philosophical accounts cannot account for all empirically observed cases of joint action. Thus, an account of a different kind of joint action is needed. The “constitutionalists” argue that established philosophical accounts do not provide sufficient information about the mechanisms underlying the performance of joint actions. Thus, more attention to the cognitive building blocks of joint action is needed. However, the two minimalist approaches are often treated as if they could be used to support the same conclusions and inferences. Thus, Pacherie (2013, 1818) prefaces her minimalist account of joint action by the following considerations about the limitations faced by established philosophical accounts of shared intentional action:

. . . these philosophical accounts suffer from two important limitations. First, they are ill suited to joint actions performed by young children who have yet to develop the psychological structures they deem necessary for joint actions. Second, as their focus is on the stages that precede actual engagement in joint action, they have little to say on what makes it possible for agents to successfully carry out a joint action and thus realize their shared intention . . .

Pacherie seems to believe that both of these considerations favor her minimalist account of joint action over its maximalist alternatives. However, clearly a minimalist account of joint action cannot play both the complementarity and the constitution roles at the same time. The complementarity strategy requires that the joint actions that minimalist accounts elucidate are of a different kind from the joint actions that are targeted by established philosophical accounts of shared intentional action. The constitution strategy requires that the kinds of joint actions that it is used to account for are among the very same phenomena as the kinds of joint actions that are investigated by the maximalists. A potential ambiguity between the complementarity and the constitution strategies accordingly seems like a confusion, which should be avoided. I will help myself to the vantage point of philosophy of science to clear up the terrain, using Bechtel and Richardson’s (2010) account of the heuristics of decomposition and localization to elucidate the research strategies of the minimalists.

Bechtel and Richardson (2010) describe the heuristics of decomposition and localization as a research strategy for studying the behavior of complex systems, which are thought to be nearly decomposable. Roughly speaking, nearly decomposable systems involve subsystems, whose behavior is relatively independent of the other parts of the system. ¹⁰ Examples of the successful use of decomposition and localization in the history of science can be identified in the discovery of cell metabolism and circadian rhythms in the life sciences. However, even when functional decomposition and localization fails due to the fact that near decomposability is not satisfied, or is not satisfied in the manner expected by the researchers, this failure can be instrumental in guiding us to understanding how the system actually functions (Bechtel and Richardson 2010, 24). To borrow Bechtel’s (2009, 545) example, the gyri and sulci of the cerebral cortex turned out not to be operational parts of the brain, but incidental byproducts of how the cortex folds in the human skull. The failure of simple localization led researchers to investigate the brain on the basis of cytoarchitectural criteria and new experimental techniques, such as stimulation and inhibitory studies, which fared better in developing an understanding how the brain actually functions.

Why should we believe that joint action makes an appropriate target for investigation using the heuristics of decomposition and localization as a research strategy? Well, joint action is arguably a complex phenomenon in the sense that it involves the interaction of numerous different motor control and perceptive processes in the human mind. Joint action is also complex in the sense of multiple realizability, which implies that the same behavioral patterns could be produced by many different psychological mechanisms. Third, the obvious fact that joint action involves several physically distinct individuals guarantees that at least some aspects of the phenomenon are likely to be nearly decomposable. Fourth, additional support for near decomposability can be gleaned from evidence that there are important modular components to certain human mindreading processes (Apperly and Butterfill 2009). Thus, we can treat the near decomposability of shared intentional actions as a useful working hypothesis. Indeed, it seems that much existing research on joint action can be re-described in terms of the research strategy of decomposition and localization, even if reliance on this kind of methodology has been tacit, rather than fully spelled out.

Let us consider Bechtel and Richardson’s (2010, 18) distinction between two different methodological approaches to decomposition and localization. First, we can adopt the top-down procedure, which Bechtel and Richardson call the “synthetic” approach. If we adopt this procedure, we start by providing a description of the functions that a given system must perform. We then move down from these functions to a hypothetical system that could perform them by imagining possible mechanisms that might underlie performance. Alternatively, we can adopt the bottom-up procedure, which Bechtel and Richardson call the “analytic” approach. If we adopt this procedure, we attempt to directly identify parts in the system and the operations that they perform. We then build up from these parts and operations to the behavior of the complex system as a whole by combining and organizing them in various ways until something resembling the composite behavior of the system comes out. In the synthetic procedure, we fix the functional description first, and treat the parts and operations as variables to be manipulated in the course of theory construction. In the analytic procedure, we fix the parts and operations first, and treat the functional description as something to be determined at a later stage by the orchestration of these parts and operations. Bechtel and Richardson (2010, 18) identify a paradigm exemplar of the top-down procedure for decomposition and localization in the functionalist tradition in philosophy of mind, while the bottom-up procedure is exemplified by the developmental trajectory of modern chemistry.

Given the above description of the two procedures, it requires no stretch of imagination to identify the top-down procedure in Bratman’s maximalist of shared intentional action. After all, Bratman (1999, 144) explicitly characterizes his account of shared intention as a functional account, and he even seems to treat the underlying notions of belief and intention as functional notions (Bratman 1987). Equally clearly, the constitutionalists have exhibited the bottom-up procedure. They have attempted to identify parts and operations that make performing joint actions possible, while making few commitments about the functional patterns of behavior that might emerge from the orchestration of these parts and operations. More surprisingly, the complementarists seem to have also adopted the top-down procedure. This brings them methodologically closer to Bratman’s functional account of joint action than to the research strategy of the constitutionalists. To be sure, the accounts of the complementarists have been informed by empirical research concerning the cooperative dispositions and mindreading abilities of human infants. However, this research is primarily behavioral in nature, and involves little detailed information about the underlying psychological mechanisms (Bernier, Spelke, and Skerry 2013). Given the shortage of mechanistic knowledge, the range of complementarist accounts reflects the space of possible alternatives for explaining rudimentary forms of joint action, which have not yet been closed out or settled by empirical research.

As more information accumulates about the mechanisms that make performing joint actions possible, the bottom-up and the top-down perspectives might be expected to eventually merge with one another. However, this should not be taken to entail that the integration of the two perspectives is bound to occur without any strain on either side. Imagine that the mechanism for a particular pattern of behavior cannot be localized, despite repeated attempts to do so in favorable epistemic circumstances. Then there will be no basis for postulating associated functional roles, no matter how characteristic or familiar they may seem to common sense. This gives rise to a kind of existential risk for many contemporary philosophical accounts of shared intentional action. In principle, neither maximalist nor minimalist accounts are immune to this kind of epistemic vulnerability. However, the minimalists have so far been much more receptive to empirical research, and willing to modify their accounts in response to new empirical inputs. By contrast, the maximalists have been more inclined to hold their accounts hostage to conceptual intuitions, and to grant empirical inputs only a secondary role in theory construction. Given the rapid pace of progress in the cognitive science of joint action during recent years, this strategy is unlikely to remain successful in the long run.

How should established philosophers of shared intentional action be prepared to respond to this challenge? Well, we can begin by noting that the top-down approach to decomposition and localization is often most useful when we are dealing with phenomena that we do not yet understand very well. As more information accumulates about the mechanisms that make performing joint actions possible, the bottom-up approach can be expected to eventually take precedence over the top-down procedure. This is a natural development, which philosophers should not resist. However, the top-down and the bottom-up approaches should not be regarded as substitutes for one another. As Bechtel and Richardson (2010, xl) emphasize, a mechanism is always a mechanism for something. An account of the parts and operations that bring about joint actions would accordingly be incomplete without the right functional description of its target behavior. Articulating such functional descriptions may in principle involve both philosophical analysis and empirical research. However, the relevant functional descriptions will often need to be updated, as more information comes in about the constraints and affordances posed by the parts and operations that underlie performance. To avoid becoming obsolete, established philosophers of shared intentional action need to learn to grant empirical inputs a much more important role in theory construction than they have previously done. They have many valuable lessons to learn from the minimalists in this respect, complementarists and constitutionalists alike.