Humans have been characterized as the ultra-cooperative species (Tomasello 2009, 2011). This ultra-cooperativeness has made us one of the most successful species on earth, spreading all over the planet, creating and developing cultural artifacts and practices that are themselves culturally transmitted and accumulate over time, thus giving us a further competitive edge over other species. According to Tomasello, underlying humans’ ultra-cooperativeness are a set of species-unique skills and motivations for shared intentionality, involving “such things as the ability and motivation to form shared goals and intentions with others in collaborative activities, and the ability and motivation to share experience with others via joint attention, cooperative communication, and teaching” (2011, p. 6).
The gist of the social creation myth I am proposing in this section is that the main benefits associated with intentions and with the kind of control over actions they make possible arise in social cooperative contexts where agents have to coordinate their actions to achieve a shared goal. I start with an examination of the special demands for coordination acting jointly with others creates. I then explain how the capacity to form conscious intentions is a crucial component of our ability to meet these demands.
Successful joint action depends on the efficient coordination of participant agents' goals, intentions, plans, and actions. As I argued elsewhere (Pacherie 2012), it is not enough that agents control their own actions, i.e., correctly predict their effects, monitor their execution and make adjustments if needed. They must also coordinate their actions with those of their co-agents so as to achieve their joint goal. For that they must monitor their partner’s intentions and actions, predict their expected consequences and use these predictions to adjust what they are doing to what their partners are doing. The implication of these processes, however, is not unique to joint action nor enough to promote their success. In competitive contexts they also play an important role. For instance, in a fight being able to anticipate your opponent’s moves and to act accordingly is also crucial. What is furthermore required in the case of joint action is that co-agents share a goal and understand the combined impact of their respective intentions and actions on their joint goal and adjust them accordingly. In competitive contexts, an agent should typically aim at predicting his opponents' moves, while at the same time endeavoring to make his own moves unpredictable to his opponents. In contrast, in cooperative contexts mutual predictability must be achieved for efficient coordination towards a shared goal to be possible. Agents should be able to align their representations of what themselves and their partners are doing and of how these actions together contribute to the shared goal.
Various forms of uncertainty can undermine mutual predictability, the alignment of representations and hence coordination. They can be organized into three broad categories. The first category involves motivational uncertainty: we can be unsure how convergent a potential partner’s interests are with our own interests and thus unsure whether there are goals we share and can promote together. The second category involves instrumental uncertainty: even assuming that we share a goal, we can be unsure what we should do to achieve that goal, or, if we have a plan, unsure how roles should be distributed among us, or, yet, unsure when and where we should act. The third category involves common ground uncertainty: we can be unsure how much of what is relevant to our deciding on a joint goal, planning for that goal and executing our plan is common ground or mutually manifest to us.
Philosophical accounts of joint agency, including Bratman’s (2009, 2014) do not ignore these challenges but they are essentially concerned with high-level processes involved in making decisions about whether or not to act together and in advance planning. Their focus is on the coordination of agent’s intentions prior to acting and they pay little heed to the processes enabling people to coordinate during action execution. In contrast, in the last decade, cognitive scientists have investigated joint action by focusing on lower-level online coordination processes in relatively simple joint tasks and on the factors that affect these coordination processes. In what follows, I will argue that there are important limitations to what these advance and online coordination processes can achieve and that high-level online intentional control is crucial to overcoming these limitations. First, however, let us consider the main characteristics of the two sets of coordination processes philosophers and psychologists typically focus on.
Bratman’s account of shared intentions is a good illustration of the way philosophical accounts approach coordination issues in joint action. In addition, its explicitness makes it possible to see clearly what advance coordination involves and how it is achieved.
Bratman (2009) proposes that shared intention involves the following conditions as its main building blocks:
Intentions on the part of each in favor of the joint activity.
Interlocking intentions: each intends that the joint activity go in part by way of the relevant intentions of each of the participants.
Intentions in favor of meshing subplans: each intends that the joint activity proceed by way of subplans of the participants that are co-realizable and can be consistently agglomerated.
Disposition to help if needed: given that the contribution of the other participants to the joint activity is part of what each intends, and given the demands of means-end coherence and of consistency that apply to intentions, each is under rational pressure to help others fulfill their role if needed.
Interdependence in the persistence of each participant’s relevant intention: each believes that the persistence of the other participants' intention in favor of the joint activity depends on the persistence of his own and vice-versa.
Joint-action-tracking mutual responsiveness: each is responsive to each in relevant subsidiary intentions and in relevant actions in a way that tracks the joint action.
Common knowledge among all participants of all these conditions.
Let me offer some comments on these conditions. First, Bratman offers these conditions as a set of sufficient conditions for a shared intention, leaving it open that shared intentions may be realized in other ways, in particular in cases of joint activities involving institutions. Second, conditions (1), (2) and (5) are meant to deal with motivational uncertainty. Bratman points out that the concept of a joint activity that figures in the contents of the intentions in (1) should be understood in a way that is neutral with respect to shared intentionality. So condition (1) only insures that agents share goals in a weak sense of the notion. Rather it is condition (2) that is in charge of insuring that the motivational states of the agents align in the way required for joint cooperative activity: it is the fact that for each participant, the content of their intention refers to the role of the intentions of other participants that, for Bratman, captures the intentional jointness of their actions. Condition (5) in turn specifies how these motivations stay aligned. Third, conditions (3), (4) and (6) relate to means-end uncertainty and are meant to reduce it. According to Bratman, they can be derived from condition (2) taken together with the norms of practical rationality that already govern individual planning and acting. Bratman’s key idea is that the interlocking and interdependent intentions of individual participants, in responding to the norms of practical rationality governing individual planning agency, will also respond to the norms of social agglomeration and consistency, social coherence and social stability shared intentions are subject to. This would involve, in Bratman’s terms, commitments to mutual compatibility of relevant sub-plans, commitments to mutual support, and joint-action tracking mutual responsiveness. Finally, the function of condition (7) is, rather obviously, to reduce common ground uncertainty.
Bratman’s basic idea is thus that this structure of interlocking and interdependent intentions, when it functions properly, frames relevant bargaining and shared deliberation and thus supports and guides coordinated planning and action in pursuit of the intended shared activity. Unsurprisingly, since Bratman’s theory of joint agency is continuous with his planning theory of individual intentions, it is in virtue of the pragmatic functions intentions already serve in the individual action case that the interlocking and interdependent intentions of individual participants can also support coordination in the joint action case.
While Bratman, in his condition (6), stipulates that agents should be mutually responsive not just in their relevant intentions and subsidiary intentions but also in relevant actions in a way that tracks the joint action, his account doesn't tell us by what means mutual responsiveness in action is achieved. To know more about this, we have to turn our attention to recent psychological work on joint agency. In contrast to philosophical approaches, cognitive psychology studies of joint action typically focus on the perceptual, cognitive, and motor processes that enable individuals to coordinate their actions with others online.
Knoblich and colleagues (Knoblich et al. 2011) distinguish between two broad categories of coordination processes, emergent and planned. In emergent coordination, coordinated behavior occurs due to perception-action couplings that make multiple individuals act in similar ways. One source of emergent coordination is entrainment, the process of synchronizing two or more actors’ rhythmic behaviors with respect to phase (e.g., Richardson et al. 2007). A second source of emergent coordination is perception-action matching, whereby observed actions are matched onto the observer’s own action repertoire and can induce the same action tendencies in different agents who observe one another’s actions (Jeannerod 1999; Prinz 1997; Rizzolatti & Sinigaglia 2010; Knoblich & Sebanz 2008). Importantly, emergent forms of coordination are independent of any joint plans or common knowledge, which may be altogether absent. They support basic forms of motor and representational alignment that can facilitate mutual responsiveness in action, but they do not ensure that the agent’s actions track a joint goal. Indeed, the successful performance of some joint actions may require that these automatic coordination processes be inhibited. For instance, the performance of composer Steve Reich’s famous piece, Drumming, based on the technique of phasing, requires the musicians to play the same rhythmic pattern out of sync.
In planned coordination, agents plan their own actions in relation to the joint goal and also to some extent to their partners' actions. As emphasized by Knoblich et al. (2011), shared task representations play an important role in planned coordination. Shared task representations do not only specify in advance what the respective tasks of each of the co-agents are, they also provide control structures that allow agents to monitor and predict what their partners are doing, thus enabling interpersonal coordination in real time. Empirical evidence shows that having shared task representations influences perceptual information processing, action monitoring, control and prediction during the ensuing interaction (Heed et al. 2010; Schuch & Tipper 2007; Sebanz et al. 2006; Tsai et al. 2006). Furthermore, several studies (Sebanz et al. 2005; Sebanz et al. 2006) have shown that actors may form shared representations of tasks quasi-automatically, even when it is more effective to ignore one another.
Several researchers have also suggested that joint attention provides a basic mechanism for sharing representations of objects and events and thus for creating a perceptual common ground in joint action (Tomasello & Carpenter 2007; Tollefsen 2005). To act jointly, it is often necessary not only that the co-agents identify the objects to be acted upon, their location as well as the location of possible obstacles, but also be mutually aware that they do. Joint attention may thus play an important role in ensuring that co-agents track the same objects and features of the situation and be mutually aware that they do. In a recent study, Böckler et al. (2011) showed that attending to objects together from opposite perspectives makes people adopt an allocentric rather than the default egocentric frame of reference. These authors suggest that taking an allocentric reference may support the efficiency of joint actions from different spatial orientations. Independently of mutual manifestness, being able to assess what others are perceiving, or can or cannot perceive at a given moment in time may also facilitate coordination. For instance, a study by Brennan and colleagues (Brennan et al. 2007) demonstrated that co-agents in joint visual search space were able to distribute a common space between them by directing their attention depending on where the other was looking and that their joint search performance was thus much more efficient than their performance in an individual version of the search task.
There are, however, important limitations to what these emergent and planned online coordination processes can achieve. First, to the extent that they exploit perceptual information, they can be of no help unless a certain amount of common perceptual information is indeed available to co-agents. Second, even when common perceptual information is available, there are limits to our processing capacities. An agent may be able to simultaneously track what a small number of other agents are currently doing or attending to, but when the number of agents increases, this capacity soon finds its limits. Our capacity to co-represent the actions, goals, and intentions of other agents we observe acting encounters similar limitations. Understanding of actions through motor resonance and mirroring works only to the extent that the observed actions are part of the action repertoire of the observer. Similarly, when actions are relatively novel, agents may not yet have formed sufficiently detailed shared task representations. Finally, unexpected effects of action execution or failures of coordination may reveal various forms of misalignment between partners' representations or indicate that their representations, though aligned, were inaccurate.
When pre-alignment is insufficient or breakdowns occur due to misalignment in the action execution phase, the deliberate and conscious production of social signals aimed at aligning or realigning relevant representations becomes crucial. Agents cannot count on alignment arising spontaneously. They have to make it happen. Intentional communication, whether verbal or not, is then needed to make it happen.
As emphasized by Herbert Clark (2006), joint activities can typically be partitioned into two sets of actions: a basic joint activity and coordinating joint actions. The basic joint activity comprises all the actions essential to achieving the basic joint goal, while the coordinating joint actions consists in the set of communicative acts about the basic activities that insure relevant representational alignment. To study this partitioning of joint activities, Clark and his co-workers ushered two people in a small room, gave them the parts of a kit for a TV stand and asked them to assemble the stand, videotaping them and recording their verbal exchanges while putting they did it. Here’s a short extract of their exchanges, taken from Clark (2006, p. 128):
Ann Should we put this in, this, this little like kinda cross bar, like the T? like the I bar?
Burton Yeah ((we can do that))
Ann So, you wanna stick the ((screws in)). Or wait is, is, are these these things, or?
Burton That’s these things I bet. Because there’s no screws.
Ann Yeah, you’re right. Yeah, probably. If they’ll stay in.
Burton I don’t know how they’ll stay in ((but))
Ann Right there.
Burton Is this one big enough?
Ann Oh ((xxx)) I guess cause like there’s no other side for it to come out.
Burton M-hm.
[8.15 sec]
Burton ((Now let’s do this one))
Ann Okay
First, it should be noted that, as often happens in daily life, this joint activity was not planned in advance. Instead, Ann and Burton discover that they have to assemble a TV stand and work out together what they should do as they go along. Second, Clark points out that Ann and Burton’s coordinating joint actions are structured in what he calls projective pairs, comprising a proposal and an uptake (i.e., full acceptance, altered acceptance or rejection of proposal). Third, the exchanges can be gestural as well as verbal. For instance, instead of, or concomitantly with, asking verbally whether Burton is ready to fasten the screws, Ann may present him with the screwdriver and his taking it count as acceptance. Fourth, the contents of these exchanges show that they are aimed at reducing instrumental uncertainty. Typically, they are about what should be done and how, who should do what, and when and where it should be done. When the task presents difficulties, they may also serve to reduce motivational uncertainty. For instance, Burton might ask whether they should give it a last try and Ann either acquiesce or reject the proposal. Finally, the structure of the projective pairs shows that at the same time they aim at reducing common ground uncertainty. Proposals are about potential alignments and full acceptance confirms alignment and common ground. Tellingly, with altered acceptance uptakes, projective pairs evolve into projective triads, the third element of the exchange being the proposer’s uptake on the alteration.
Importantly, to negotiate and achieve alignment in this way, we must be aware of our own and others' intentions and beliefs and this at two levels, corresponding to the two sides of the partitioning characterized by Clark. On the one hand, it is essential to the fulfillment of communicative intentions that they be recognized as such by the addressee (Grice 1957; Recanati 1986; Sperber & Wilson 1986). On the other hand, what agents communicate in these contexts is information about their beliefs and intentions regarding the joint action. This suggests that the development of self-consciousness and consciousness of other minds, of intentional communication, and of increasingly complex forms of coordinated joint action go hand in hand.
The success of both individual and joint action depends on representational alignment. In the case of individual action, representation alignment takes two main forms. First, at a given level of action specification, a match should be achieved between representations of desired, predicted and actual states. We can call this first form of alignment intra-level alignment. Second, inter-level alignment is also necessary; that is, despite differences in representational format and resources, action specifications at different levels of the action representation hierarchy should be kept aligned. Conscious online control may be needed to restore alignment when severe intra- or inter-level discrepancies occur. However, it may be argued that in the individual case alignments are taking place within a single cognitive system and that this system is normally sufficiently integrated or unified that serious misalignments are rare and thus that the need for online conscious control is limited.
The main difference between individual and joint actions lies in the coordination demands essential to joint action. Thus, a third form of representational alignment becomes crucial in joint action. In addition to individual intra-and inter-level representational alignment, inter-agent representational alignment is necessary to meet coordination demands. Inter-agent alignment may be achieved in part through advanced planning, as proposed by Bratman. It can also be achieved in part through online emergent and planned coordination processes of the types explored and described in the recent psychological literature. However, there are important limitations to what these coordination processes can achieve. Advance planning, when it takes place, may help define a shared background framework for the joint action, but at this stage it is typically impossible to anticipate all the coordination demands that will arise at the execution stage. Some of these demands may be met by the kinds of online coordination processes reviewed earlier in this section, but, as I pointed out, there are also important limitations to what they can achieve. In many instances, the progress of a joint action is hindered or the action breaks down due to various forms of misalignment between agents' representations. In such instances, individual corrections do not suffice to put the joint action back on track. Rather, to overcome these failures, agents need to align or realign their representations. This process calls for what Clark calls coordinating joint actions, that is, communicative acts about the basic joint activity. These communicative acts in turn are intentional and aim at communicating information about the agents' intentions and beliefs with a view to achieve alignment. But one can only communicate intentionally about one’s beliefs and intentions if one is aware of them. Conversely, one can only understand the communicative acts of other agents if one realizes that these agents have a capacity for intentions. Finally and crucially, as already emphasized by Velleman (2007) in his discussion of Bratman’s account, intentions could not serve their pragmatic functions unless they also had an epistemic role. In other words, if my having the intention to A didn't count as a form of practical self-knowledge and didn't give me grounds to believe that I would act as intended, my communicating (sincerely) about my intention to A would not license other agents to form beliefs about my future actions and thus would not yield the kind of inter-agent representational alignment needed to achieve coordination.
To recap, joint actions create more comprehensive demands for representational alignment than individual actions, since their success depends not just on individual intra- and inter-level representational alignment but also on inter-agent representational alignment. New resources are needed to meet these demands. On the social creation myth proposed here, a capacity for conscious intentions is crucial to inter-agent representational alignment. Having conscious intentions allows us to communicate about them and engage in coordinating joint actions that create common ground and promote the success of basic joint activity. The answer this myth offers to the question what is the purpose of conscious intentions is then that it is to enable more efficient inter-personal coordination in joint action and thus reap the benefits that come with increasingly complex and flexible forms of coordinated actions. The social creation myth doesn't deny intentions an epistemic role. On the contrary, it acknowledges that intentions couldn't serve their inter-personal coordination function if they did not at the same time provide us with a form of self-knowledge. However, it views their epistemic function as subservient to their coordination function. The social creation myth does not deny either that conscious intentions play a role in the online control of individual action. Rather, it proposes that conscious control of individual action may be a by-product of a capacity for conscious control that became established in social contexts because of the role it served in solving inter-agent coordination problems and because of the benefit conferred by the forms of cooperation it made possible.