Mechanistic Explanation, Interdisciplinary Integration and Interpersonal Social Coordination

ABSTRACT Prominent research programs dealing with the nature and mechanisms of interpersonal social coordination have emerged in cognitive science, developmental psychology and evolutionary anthropology. I argue that the mechanistic approach to explanation in contemporary philosophy of science can facilitate interdisciplinary integration and division of labor between these different disciplinary research programs. By distinguishing phenomenal models from mechanistic models and structural decomposition from functional decomposition in the process of mechanism discovery, I argue that behavioral and cognitive scientists can make interlocking contributions to families of mechanism models representing a) what central forms of interpersonal social coordination consist in, b) what types of agents engage in such forms of coordination, c) what types of cognitive mechanisms give rise to such forms of coordination, d) how the operation of such mechanisms is modulated by the natural and social environments that the agents populate. Thus mechanistic philosophy of science can play a heuristic role in guiding the research strategies of scientists in different scientific disciplines in a manner that is complementary to the research strategies of scientists in other disciplines and thereby efficient from the point of view of the division of epistemic labor.


Interpersonal Social Coordination as a Topic of Interdisciplinary Investigation
The human capacity for interpersonal social coordination manifests itself across a wide variety of different circumstances, from subconscious bodily synchronization (Dumas and Fairhurst 2021;Gallotti, Fairhurst and Frith 2017;Schmidt and Richardson 2008) to joint attention between a toddler and her caretaker (Rakoczy 2017; Siposova and Carpenter 2019;Tomasello 2019) and turntaking during linguistic conversation (Levinson 2016;Stivers et al. 2009).The nature of this capacity (or group of capacities giving rise to similar phenomena), as well as the cognitive and social mechanisms underlying its manifestations in various environments, have become prominent topics of research in numerous different scientific disciplines, including cognitive science, developmental psychology and evolutionary anthropology (which are the disciplines that I will focus on in this paper). 1However, genuinely interdisciplinary investigations of interpersonal social coordination have remained rare, and there have been few attempts to integrate many different disciplinary perspectives into a more comprehensive, multidisciplinary theory of social coordination (however, see Thonhauser and Weichold 2021;Tomasello 2019).
My goal in this paper is to contribute to our understanding of the division of labor between different scientific disciplines studying interpersonal social coordination.The mechanistic approach The idea that different approaches to social behavior can be integrated by analyzing them in terms of hierarchical or interlocking levels of mechanisms has become more common during recent years (Bechtel and Richardson 2010;Craver and Darden 2013;Glennan 2017).A recent review about mechanisms for collective action by Georg Thonhauser and Martin Weichold testifies to this trend (Thonhauser and Weichold 2021). 2 They describe philosophical analyses of collective intentionality, game-theoretic models of team reasoning, social identity theory, ecological social psychology, enactivist approaches to participatory sense making and sociological practice theory as providing grounds for identifying mutually complementary mechanisms for collective action that ought to be integrated with one another.Unfortunately, their paper suffers from some ambiguities, because they do not substantiate this promissory note on interdisciplinary integration by a systematic analysis of mechanisms and mechanistic explanation that would make it possible to articulate precisely what is meant by a mechanism, how models of distinct or overlapping mechanisms can be integrated with one another or how to proceed if there are some incongruities between different models of mechanisms (as is surely likely to sometimes be the case).Given that I am skeptical of the mechanistic credentials of some of their cases as well as their scope for integration, I will initially focus on a more restricted literature than Thonhauser and Weichold (2021)-my focus will be primarily on empirical studies of small-scale forms of interpersonal 3 social coordination in the cognitive and behavioral sciences.This will allow me to demonstrate clearly how the framework of mechanistic integration that I will present works in practice, and I will return to some of the research programs discussed by Thonhauser and Weichold's (2021) at the end of this paper.
The next section of this paper will introduce the concept of mechanism, different stages of mechanism discovery and the ways in which the search for mechanisms can guide interdisciplinary investigation through mutual constraints.In Section 3, I will move on to empirical research on interpersonal social coordination in the fields of developmental psychology and evolutionary anthropology.My focus there will be on the stages through which the capacity for interpersonal social coordination develops during human ontogeny and the ways in which human capacities for interpersonal social coordination contrast with those of other primate species, especially the chimpanzees.In Section 4, I will turn to research on the mechanisms of emergent and planned coordination in cognitive science and consider whether this body of research supports the so-called Vygotskian intelligence hypothesis about the evolutionary distinctiveness of human capacities for social coordination.In Section 5, I will apply the distinction between mechanistic decomposition and phenomenal decomposition (introduced in Section 2) to carve out a heuristic division of labor between behavioral and cognitive scientists studying interpersonal social coordination.In Section 6, I will consider the boundaries of mechanistic integration and some examples of research programs on social coordination that do not appear to be plausible candidates for interdisciplinary integration under a mechanistic framework.At the end of the paper, I provide brief concluding remarks and reflect on the role of mechanistic integration in scientific progress.

Mechanisms, Mechanistic Explanation and Mechanism Discovery
The mechanistic approach to explanation in the philosophy of science was originally proposed as an alternative to the deductive-nomological account of explanation, which was based on the idea that scientific explanation consists in logical derivation of an observation statement from statements describing a general law and auxiliary conditions describing the conditions under which the law manifests itself in particular circumstances (Hempel 1966;Nagel 1961).For example, a deductivenomological explanation of why an oar submerged in water appears bent might appeal to the laws of refraction and the particular angle in which the oar is viewed.However, the deductivenomological account of explanation never seemed like a good fit in accounting for explanatory relations within other sciences than physics, and it encountered several crucial objections during the latter half the 20 th century.First, the deductive-nomological account seems unable to account for explanatory asymmetries, such as the idea that light hitting a flagpole at a certain angle is causally responsible for the shape of a shadow but the shadow of the flagpole is not causally responsible for the shape of the flagpole (Salmon 1989).Second, the deductive-nomological account presupposed that explanation is an essentially linguistic affair and neglected the role that visual diagrams and scale models (e.g.representing the molecular building blocks of a cell) can play in explanation (Bechtel and Abrahamsen 2005).Third, the deductive-nomological account misrepresented the role that general regularities or laws play in explanation, given that regularities in many sciences (e.g.relating to the functional constraints of short-term memory) seem to pick out phenomena to be explained, rather than explanatory facts (Cummins 2000).Philosophers in the mechanistic tradition of philosophy of science responded to these challenges by arguing that explanation in many circumstances requires identifying constituent entities and activities that underlie, produce or maintain the general regularities that were (erroneously, they believed) taken by the deductivenomological account to be explanatory of particular facts (see e.g.Bechtel and Abrahamsen 2005;Glennan 2017;Machamer, Darden and Craver 2000).
The notion of a mechanism can be understood in a minimal sense as referring to 'entities (or parts) whose activities and interactions are organized so as to be responsible for the phenomenon' (Glennan 2017, 17;cf. Bechtel and Abrahamsen 2005;Machamer, Darden and Craver 2000).Three features of Glennan's minimal definition are worth highlighting here.First, the entities and activities that make up a mechanism are individuated relative to the phenomenon to be explained.For example, the behavior of the human circulatory system can be mechanistically explained by referring to the operations that are carried out by (among other things) the heart, the lungs, the arteries and the veins (Bechtel and Abrahamsen 2005).Other features of human physiology, such as the skin or the joints, are not relevant for the mechanistic explanation of this particular phenomenon.Second, mechanisms are hierarchical in the sense that the entities and activities that make up a mechanism can themselves typically be mechanistically decomposed into further entities and activities (with the possible exception of fundamental physics).For example, the heart, which forms one central part of the human circulatory system, can be further decomposed into its four chambers, the vena cava, the pulmonary artery and the aorta (as well as further constituents).Going one step further, these entities and their activities can be decomposed into molecular parts and activities, which can be decomposed into atomic entities and activities, and so on.Third, the deliberately permissive notion of responsibility can be understood in a causal or constitutive sense, making it possible for the definition to accommodate both synchronic mechanisms that take place at a time and diachronic mechanisms that take place across time.
The mechanistic approach to explanation is based on the idea that many generalizations in the special sciences can be explained by describing underlying constituent entities and activities that underlie, produce or maintain a phenomenon (Craver and Darden 2013).Mechanistic explanations are conveyed by linguistically or visually represented mechanism models, which may involve a greater or lesser degree of detail, and incorporate some abstractions or idealizations relative to the phenomenon to be explained as well as the underlying entities and activities.The process of mechanism discovery can be understood as the piecemeal and gradual filling in of a mechanism model, beginning with an incipient and incomplete mechanism sketch, which is then used as the basis for constructing a more detailed and satisfactory mechanism schema.A mechanism sketch is an 'incomplete representation of a mechanism . . . it characterizes some parts, activities, and features of the mechanism's organization, but it has gaps' (Craver and Darden 2013, 31).These gaps may be conveyed by black boxes or question marks in a visual diagram or by filler terms in a verbal description, such as 'activate, cause, encode, inhibit, produce, process and represent [which] are often used to indicate a kind of activity in a mechanism without providing any detail about how that activity is carried out' (Craver and Darden 2013, 31).A mechanism schema replaces such black boxes with grey boxes or glass boxes, which 'are complete enough for the purposes at hand [and] include all of the entities, properties, activities, and organizational features that are relevant to the pragmatic ends for which the description is being used' (Craver and Darden 2013, 31).
The formulation of a mechanistic explanation can often be understood as a multi-part endeavor (Bechtel and Richardson 2010;Glennan 2017, 66-68).The first part is phenomenal decomposition, or describing the causal roles that the mechanism performs, and distinguishing it from other, related phenomena.These causal roles are sometimes described as functions of the mechanism (e.g.Bechtel and Abrahamsen 2005), but they should not be conflated with etiological functions that brought about the mechanism through evolution and/or design or 'proper functions' of the larger system that the mechanism is part of (Craver and Darden 2013, 53).Consider the HIV virus, whose behavior can be mechanistically explained, although attacking the immune system of the body is highly dysfunctional from the perspective of its host organism, rather than a proper function of that organism.Or consider features that have been brought about as indirect side-effects or 'spandrels' (Gould 1997) of features that were selected for by evolution, such as the floppy ears of many domesticated animals (Trut et al. 2009).In addition to ontic constraints, the epistemic interests of researchers may also set boundaries on phenomenal decomposition.For example, it may sometimes be appropriate to treat learning as a singular phenomenon, while for other theoretical purposes it may be necessary to distinguish between different types of learning, such as association learning, reward learning and social learning (Frith and Frith 2012).
The second part in formulating a mechanistic explanation is mechanistic decomposition, or describing the constituent entities and activities that give rise to the phenomenon.This process can be further decomposed into structural decomposition and functional decomposition.Structural decomposition involves analyzing a complex system into constituent parts, without yet necessarily identifying what activities and interactions those parts engage in.Functional decomposition involves analyzing a complex system into constituent activities, without yet necessarily identifying which physical parts carry out these activities.Thus, a complete mechanistic explanation can be described as consisting of two parts: a phenomenal model, which describes the causal roles that the mechanism performs (e.g. the heart pumps blood to the circulatory system), and a mechanism model, which describes the entities and activities that give rise to the phenomenon (e.g. the four chambers contract and relax, the valves open and close to let blood flow through the chambers and to block reverse movement, etc.).
The processes of mechanistic decomposition and phenomenal decomposition constrain one another during the process of mechanism discovery.While phenomenal decomposition guides the search for mechanisms, mechanistic decomposition may force us to revise our conception of the phenomena to be explained (Craver and Darden 2013, 61-62).In general terms, a constraint can be described as a representation or a finding 'that either shapes the boundaries of the space of possible mechanisms or changes the probability distribution over that space' (Craver 2007, 247).Thereby constraints 'constitute the relevant evidence for evaluating how-possibly descriptions of mechanisms' (Craver 2007).Paradigmatic bottom-up constraints on mechanisms come in the form of spatiotemporal claims about the ontic nature of mechanistic building blocks -for example, that a certain molecule occupies a certain space, has a certain shape or is connected to another molecule (Craver 2007, 249).Top-down constraints, in turn, often relate to the ways in which phenomena are described -for example, the idea that memory involves encoding, storing and retrieving information constrains mechanistic explanation in the sense that any cellular-level process that does not serve these roles does not count as a form of remembering (Craver 2003).Some further constraints are based on representational norms that have to do with the logic of explanation and need not (even if they may) be regarded as laws of nature -for example, the idea that causes precede their effects and that mere temporal sequences are not explanatory (Danks 2014;Miłkowski 2016). 4 To avert one important misunderstanding, the distinction between mechanistic decomposition and phenomenal decomposition should not be confused with the distinction between empirical and theoretical investigation.Rather, even phenomenal decomposition almost always requires (or must be related to) empirical study in order to provide informative constraints on mechanistic decomposition.For example, the phenomenal distinction between short-term and long-term memory was made largely on the basis of experimental studies, which demonstrated contrasting functional constraints that are associated with these different memory systems (without yet demonstrating what brain mechanisms are responsible for these functional differences) (Atkinson and Shiffrin 1968).If phenomenal decomposition did not require empirical study, it would be too facile to present, say, contrasting functionalist analyses of intentional attitudes in analytic philosophy as contributing to mechanistic explanation simply because they involve contrasting definitions of what certain central categories of mental phenomena consist in (e.g.contrasting definitions of 'intending' (Bratman 1987;Harman 1986)).Yet most functionalist analyses in philosophy are much too permissive to provide many informative constraints on mechanistic decomposition, as exemplified by the challenges that are present in mapping out even basic elements of these analyses, such as the subject-mode-content distinction, to any types of cognitive primitives that play a role in cognitive science (I will return to this point at the end of this paper).

Interpersonal Social Coordination from a Phylogenetic and Ontogenetic Perspective
The extraordinary ecological spread and adaptability of the human species, which together with its domestic animals makes up over 90% of the Earth's mammalian biomass (Bar-On, Phillips and Milo 2018), has been attributed to numerous different behavioral traits of human individuals and social groups -including (but not restricted to) tool use (see Seed and Byrne 2010), natural language (Scott-Phillips 2015), dietary changes due to the use of fire in cooking (Wrangham 2009) and social conventions, such as cooperative child rearing (Hrdy 2009).However, as cultural innovations, which are passed on from one generation to the next through imitation or emulation, explicit instruction and various types of collaborative activities (Frith and Frith 2012;Tomasello 1999), these contrasting accounts of the phylogenetically distinctive characteristics of the human species all share one thing in common: they are all manifestations of human social intelligence, which depend on the capacity of individuals to learn from one another and to coordinate their attitudes and actions cooperatively with those of other individuals.This makes it possible for human social groups and individuals to build on the cultural achievements of prior generations in the form of cumulative cultural evolution, which seems to be unique to the human species (Henrich and Muthukrishna 2021).
The centrality of interpersonal social coordination in human ontogeny and phylogeny has been recognized by proponents of the social intelligence hypothesis in evolutionary anthropology.They have argued that central aspects of human intelligence and problem solving capacity evolved in response to adaptive demands posed by social complexity (see Byrne and Whiten 1988;deWaal 2007 ; Dunbar and Shultz 2017; Moll and Tomasello 2007; Tomasello et al. 2005; Whiten and Byrne  1997).Social complexity is often operationalized in terms of group size or the average number of enduring pairwise social relationships that an average species member engages in.In this respect, both absolute and relative brain size (and especially neocortex size) have been shown to covary with social complexity in most anthropoid primates (Dunbar 1992;Dunbar and Shultz 2017).Moreover, relative brain size has been shown to covary with monogamous pair-bonding in many species of birds, pointing at the possibility of convergent evolution (Dunbar and Shultz 2017).Originally, the social intelligence hypothesis was presented as an alternative to the ecological complexity hypothesis, which highlighted the adaptive demands posed on memory and problem-solving capacity by nutritionally resource-constrained environments (Humphrey 1976).However, the two hypotheses should not be regarded as competing, because large brain size imposes significant energetic demands on nutrition intake (requiring new technical-instrumental skills to emerge), and technological innovations can in turn be transmitted in hominid populations by means of cultural learning processes, such as imitation and education (Kline and Boyd 2010;Muthukrishna et al. 2013).
The two most prominent versions of the social intelligence hypothesis are the Machiavellian intelligence hypothesis of Richard Byrne and Andrew Whiten (1988;Whiten and Byrne 1977) and the Vygotskian intelligence hypothesis of Michael Tomasello and his followers (Moll and Tomasello 2007;Tomasello 2019;Tomasello et al. 2005).As a rough approximation, the Machiavellian intelligence hypothesis emphasizes the trade-off between the reproductive fitnesses of particular individuals within a primate social group, as well as the uses of social intelligence for tactical deception, transitive reasoning about dominance hierarchies and the formation of potentially hostile in-group coalitions (Byrne and Whiten 1988;de Waal 2007;Whiten and Byrne 1977).The Vygotskian intelligence hypothesis in turn pays attention to differences between the reproductive fitnesses of competing social groups to the effect that groups with high degrees of internal coordination and cooperation may have outcompeted groups with a more competitive in-group bias (Moll and Tomasello 2007;Tomasello 2019).Typically, proponents of the Machiavellian intelligence hypothesis have also emphasized the continuity between human social intelligence and the social intelligence of other primates, while proponents of the Vygotskian intelligence hypothesis have highlighted the (by hypothesis) uniquely cooperative nature of human social intelligence. 5This paper will mostly follow proponents of the Vygotskian intelligence hypothesis in emphasizing the more benevolent aspects of human social coordination in the service of mutually beneficial cooperation and collaboration, but I will also consider challenges to this hypothesis and ways in which it could be made more robust.
The most compelling evidence for the Vygotskian intelligence hypothesis comes from comparative studies juxtaposing human infants of various ages with non-human primates in acting together or solving various types of collaborative tasks.For example, using a battery of 25 cognitive tests, Herrmann et al. (2007) found that chimpanzees and orangutans are as adept as 2.5-year-old human infants in tasks requiring spatial, numerical and causal reasoning, but that they perform less well in tasks requiring communication, theory of mind and social learning (see also Wobber et al. 2014).Warneken et al. (2006) found that chimpanzees were able to solve an instrumental task together with a human partner to access food, but they were unable to engage chimpanzees in simple social games that human infants seem to find intrinsically rewarding, such as bouncing a ball jointly on a large hand-held trampoline.Hare and Tomasello (2004) found that chimpanzees did not understand the cooperative motives behind helping behavior (e.g. when an experimenter points to a location where food is hidden), although they do make pragmatic inferences about what others know or believe in competitive settings.By contrast, it has been well established that human infants make spontaneous inferences about others' cooperative motives, and they help others in reaching their personal goals even in the absence of an extrinsic reward (Melis and Warneken 2016, 299-301;cf. Tennie, Jensen and Call 2016).In addition to cognitive traits, humans even seem to have distinctive physiological adaptations to facilitate social coordination: according to the cooperative eye hypothesis, the unusually large white sclera in human eyes evolved to advertise gaze direction, while most non-human primates seek to hide their gaze from others (Kobayashi and Kohshima 1997).
The Vygotskian intelligence hypothesis has been criticized by some evolutionary anthropologists on the grounds that it misrepresents the nature of primate social intelligence and that it is explanatorily extraneous to postulate distinctive cognitive capacities for a single taxon.For example, Dunbar and Shultz (2017, 5) note that 'contrary to what they [proponents of the Vygotskian intelligence hypothesis] assume, all (anthropoid) primate societies are in fact based on cooperation: primate groups are cooperative solutions to the central problems of survival, in particular predation risk'.They go on to argue that the Vygotskian intelligence hypothesis 'fails to explain why group size varies across primates, or why it correlates so robustly with brain size . . .since it appears to be a categorical phenomenon (you have Vygotskian intelligence or you do not)'.As far as I can see, these critics are correct in emphasizing that 'special pleading [for a single taxon] should always be an explanation of last resort', and therefore, the weight of the evidence should be conclusively in favor of the Vygotskian intelligence hypothesis in order to postulate a categorical distinction between human social intelligence and the social intelligence of other primates.Yet the behavioral evidence that proponents of the Vygotskian intelligence hypothesis have provided can at best be regarded as relatively weak, because the sample sizes in most laboratory studies of primate social cognition are very small, and it is difficult to employ suitable control conditions in field studies (see Tomasello and Call 2011).Moreover, the study of primate social cognition has historically been a moving target, where increasingly more sophisticated cognitive capacities have been postulated in order to make room for our evolving understanding of the complexities of primate social life (see Tomasello 2019;  cf.Tomasello 1999). 6This suggests that one should exercise caution before wholeheartedly endorsing the Vygotskian intelligence hypothesis.
The next section of this paper will consider whether research on the mechanisms of social coordination in cognitive science could provide additional support to the Vygotskian intelligence hypothesis and make it more robust.In particular, it seems that if research in cognitive science were to show that distinctively human forms of social coordination are underpinned by dedicated cognitive mechanisms that are unique to the human species, this would make the Vygotskian intelligence hypothesis much more compelling as a general account of human social intelligence.If, on the other hand, we were to find out that human social coordination is largely a function of general-purpose cognitive capacities that are shared with other primate species, then the burden of proof would remain on proponents of the Vygotskian intelligence hypothesis to explain why these capacities were recruited in the service of mutually beneficial social coordination and cooperation in the case of our species, while in the case of the other primates, they were (by hypothesis) recruited primarily in the service of Machiavellian manipulation and deceit.This would make it hard to sustain the hypothesis that human culture is qualitatively distinctive in the manner suggested by many proponents of the Vygotskian intelligence hypothesis (esp.Tomasello 2019;Tomasello et al. 2005). 7

Interpersonal Social Coordination from a Cognitive Perspective
The cognitive scientists Günther Knoblich and Natalie Sebanz together with the philosopher Stephen Butterfill have proposed a distinction between two different forms of interpersonal social coordination: emergent coordination and planned coordination.According to Knoblich et al. (2011, 59), in emergent coordination 'coordinated behavior occurs due to perception -action couplings that make multiple individuals act in similar ways' (Knoblich, Butterfill and Sebanz 2011, 59), whereas in planned coordination, 'agents' behavior is driven by representations that specify the desired outcomes of joint action and the agent's own part in achieving these outcomes'.Synergy of these two forms of social coordination is needed for many joint activities 'because there are complementary limits on what each can achieve . . .Emergent coordination is likely the key to dealing with the real-time aspects of joint action . . .[but it] alone . . .does not allow people to distribute different parts of a task among themselves, nor to adjust their actions to others so as to flexibly achieve joint outcomes'.Apart from their different functional roles, Knoblich et al. (2011) argue that emergent and planned coordination are also based on different and non-overlapping cognitive mechanisms.While planned coordination seems to be subserved by representational processes that are largely available for conscious awareness, emergent coordination is underpinned by the mostly subpersonal mechanisms of entrainment, action simulation, perception-action matching and (perceptions of) common object affordances (Knoblich, Butterfill and Sebanz 2011).In this section, I will briefly discuss the mechanisms of emergent and planned coordination, before considering how they might give rise to some of the behavioral manifestations that were discussed in the previous section.
To begin with the mechanisms of emergent coordination, the well-established phenomenon of entrainment refers to the motor behaviors of two or more individuals becoming synchronized when they perform the same action or receive the same auditory, visual or haptic information (Knoblich, Butterfill and Sebanz 2011, 66-69).For example, two individuals in adjacent rocking chairs have been shown to involuntarily adjust their rhythms to the rhythms of the other individuals (either in-phase or out-phase), even when explicitly instructed to maintain their own rhythm (Richardson et al. 2007).Action simulation refers to individuals using their own motor control systems to predict what another individual will do next (Goldman 2008), for example, when two individuals are dancing (Cross, Hamilton and Grafton 2006) or playing music (Keller, Novembre and Hove 2014) together.Perception-action matching is exhibited when individuals unconsciously mimic one another's mannerisms or facial gestures (Chartrand and Bargh 1999) or are facilitated in their motor responses by observation of similar behaviors in other individuals (e.g. in the case of grasping (Rizzolatti and Sinigaglia 2010) or tapping (Brass, Bekkering and Prinz 2001) movements).And common object affordances refer to objects in the environment that can be manipulated by several individuals in ways that they cannot be manipulated in by one individual acting alone, such as a two-handled saw (Knoblich, Butterfill and Sebanz 2011, 63), or a long plank that is too unwieldy to be lifted by one individual alone (Richardson et al. 2007).
The phenomenon of planned coordination is based on explicit mental representations about shared goals and mutual expectations that may be more or less detailed (Knoblich, Butterfill and Sebanz 2011, 77-84).Minimally, planned coordination includes explicit mental representations of 'the joint action outcome, one's own part in a joint action, and some awareness that the outcome can only be brought about with the support of another agent or force' (Knoblich, Butterfill and Sebanz 2011, 65; see also Vesper et al. 2010).Typically, planned coordination also involves representations about the identities of the other participants and their parts in carrying out the joint activity.However, not all aspects of the division of tasks need to be represented in advance, since they may also be filled in by means-ends reasoning (Bratman 2014) or emergent coordination processes as the joint action unfolds in time.Knoblich et al. (2011, 65-66) argue that planned coordination can also involve joint perceptions and shared task representations, in which the activities of other agents are represented under a common executive format with the agent's own activities (Sebanz, Bekkering and Knoblich 2006;cf. Dolk et al. 2014).In addition, planned coordination may involve the deliberate use of coordination smoothers (Vesper et al. 2010), such as positioning one's body in a suggestive manner or exaggerating certain movements when performing a manual task.
To illustrate how the cognitive mechanisms discussed by Knoblich et al. (2011) might give rise to some of the forms of social coordination studied by developmental psychologists and evolutionary anthropologists, consider a few of the comparative paradigms used by Warneken et al. (2006) to study coordination among chimpanzees and 18-24-month old human infants.One of the experimental paradigms they used was a simple social game, where infants were bouncing a ball up and down on a large hand-held trampoline together with an experimenter (pictured below).Here, the participants were performing similar activities synchronically with parallel roles, suggesting that they were drawing on the mechanisms of entrainment to adjust to the same rhythm, perception-action matching to learn the right types of bouncing behaviors and action simulation to adjust to subtle changes in tempo or to prepare for the ending of the activity.Another experiment involved an instrumental task with a more complex division of labor in the form of a 'trap door' paradigm, in which one individual had to raise a panel on top of a cage at the same time as the other individual reached inside the cage for an item of food (the instrumental reward).Solving this task seems to require explicit mental representations of the other individual's mental states as well as a joint plan with a detailed division of labor (one raises the panel, the other reaches for the food).Remarkably, Warneken et al. (2006) found that chimpanzees were able to solve the types of incentivized tasks represented by the trap door experiment with human partners, but they were unable to engage chimpanzees in the types of non-instrumental social games represented by the trampoline task at all. 8   The success of chimpanzees in coordinating in instrumental joint activities with a complex internal division of labor, such as the trap door experiment, is a challenge for proponents of the Vygotskian intelligence hypothesis, given that these tasks seem to require more developed capacities for interpersonal social coordination than non-instrumental joint activities with parallel roles, such as the trampoline task.After all, the largely sub-personal mechanisms of entrainment, perceptionaction matching and action simulation are likely to be ones that we share with our primate relatives due to their older evolutionary ancestry (see e.g.Rizzolatti and Sinigaglia 2010).However, if chimpanzees are successful in coordinating instrumental coordination tasks with a relatively complex internal division of labor, such as the trap door experiment, there should be nothing in the much simpler non-instrumental social games that they cannot do, even if they are uninterested in engaging in them.This suggests that the Vygotskian intelligence hypothesis according to which humans possess species-distinctive skills and motivations for social coordination (e.g.Tomasello 2019; Tomasello et al. 2005) should be replaced by the somewhat weaker social motivation hypothesis, according to which humans possess distinctive prosocial motivations, which make their cognitive capacities for social coordination be manifested in different ways (see Godman 2013; Godman, Nagatsu and Salmela 2014; Nagatsu and Salmela 2022).
The social motivation hypothesis can also be supported on independent grounds.For example, Coralie Chevallier and her colleagues have argued that deficits in social motivation drive subsequent deficits in social cognition (especially theory of mind capacities) in the case of autism spectrum disorders (ASDs), rather than vice versa (Chevallier et al. 2012) -reversing how traditional theories of ASD have characterized the direction of causality (e.g.Baron-Cohen, Leslie and Frith 1985).They associate deficits in social motivation with the orbitofrontal-striatalamygdala circuitry in the human brain and with the functioning of oxytocin, dopamine and opioid receptors in particular.We may also point to the well-known overjustification effect, according to which the introduction of extrinsic rewards (e.g. money or sweets) may undermine spontaneous social behaviors (see Lepper, Greene and Nisbett 1973;Tang and Hall 1995;Warneken and Tomasello 2008), as well as the prevalence of overimitation among young infants, where infants often choose to imitate an exact sequence of behaviors even when the instrumental goal of the activity could be more straightforwardly attained (see Lyons, Young and Keil 2007;Nielsen and Tomaselli 2010).In addition, we can point to infants' spontaneous tendencies to altruistic helping (Melis and Warneken 2016, 299-301), as well as the intense interest that they show towards the non-instrumental norms that govern various types of joint activities (Gräfenhain et al. 2009;Hamann, Warneken and Tomasello 2012;Kachel and Tomasello 2019).
These studies suggest that human beings in general do not view joint activities merely as a means to achieve their own private goals but as social interactions that are valuable in their own right.In accordance with this view, Marion Godman (2013) argues that social coordination in general serves two separate functions: 1) achieving the intended target outcome of an action and 2) attaining the benefits related to being part of a social bond (see also Godman, Nagatsu and Salmela 2014;Nagatsu and Salmela 2022).Her view strikes me as plausible and is also compatible with recent attempts to avoid over-intellectualizing social coordination by representing social intelligence as a form of mindshaping (Zawidzki 2013), where agents build adaptive cognitive niches for themselves and one another (cf.Laland and O'Brien 2011;Sterelny 2007), instead of attempting to read one another's minds from a purely detached third-person perspective (cf.Gopnik and Meltzoff 1997).These views suggest that both the Machiavellian intelligence hypothesis and the Vygotskian intelligence hypothesis may be guilty of over-intellectualizing the mind, and an evolutionarily more plausible view of social intelligence can be achieved by focusing on lower-level motivational processes that are sufficient for bringing about the behavioral and phenotypic effects that these hypotheses were originally conceived to explain.

Discovering the Mechanisms of Interpersonal Social Coordination Through Interdisciplinary Investigation
The preceding sections have discussed research on interpersonal social coordination in the fields of cognitive science, developmental psychology and evolutionary anthropology.While my review has been far from exhaustive, the point has not been to put forth a new theory of interpersonal social coordination, but to use it as background in order to better understand how different disciplinary perspectives on the same phenomenon fit together, what types of challenges may arise in integrating their respective contributions and how to go about resolving such challenges.With regard to my case study, I have argued that the triangulation of evidence from cognitive science, developmental psychology and evolutionary anthropology fails to provide robust support to the Vygotskian intelligence hypothesis, according to which human beings possess distinctive skills and motivations for social coordination that other primates do not possess (see esp.Tomasello 2019;Tomasello et al. 2005).Rather, it favors the weaker social motivation hypothesis that refers to species-distinctive social motivations, which make these capacities for social coordination manifest in different ways (Godman 2013;Godman, Nagatsu and Salmela 2014).This being said, my primary goal has not been to evaluate the feasibility of the Vygotskian intelligence hypothesis as an overarching account of human social intelligence, but to use my case study to ground a schematic division of labor between behavioral and cognitive scientists studying interpersonal social coordination and to provide a blueprint for how their theoretical outputs can be integrated with one another (when they are compatible).Drawing on my discussion, we can now distinguish between the following questions that they have addressed: representational processes that are available to conscious awareness, or sub-personal processes of perception-action matching, action simulation, and the like)?
(d) How is the operation of such mechanisms modulated by the natural and social environments that the agents populate (e.g. in the laboratory or in the wild, with conspecifics or with human partners)?
From the perspective of the mechanistic approach to explanation, the first two questions can be roughly understood as belonging to the domain of phenomenal decomposition, or 'describing the causal roles that the mechanism performs, and distinguishing it from other, related phenomena' (see Section 2 above).The latter two questions, in turn, address mechanistic decomposition in the sense of 'describing the constituent entities and activities that give rise to the phenomenon' (question 4 presupposes that social coordination may be influenced by environmental as well-brain-bound mechanisms).My discussion of the Vygotskian intelligence hypothesis in the previous sections illustrates how our answers to these questions set constraints on one another: for example, the identification of distinctive forms of social coordination (question A) that only human individuals seem to engage in (question B) may be suggestive of dedicated cognitive mechanisms (question C) or social-environmental scaffolding (question D) that explain such species-distinctive forms of social coordination, while failure to identify dedicated social-environmental (question D) or cognitive (question C) mechanisms for these forms of social coordination (question A) may in turn undermine hypotheses about the true distinctiveness (question B) of the forms of social coordination in question.This illustrates how the search for mechanisms is a constant back-and-forth endeavour, where our increasing mechanistic understanding of the world forces us to revise our classifications of the world and the phenomena that need explaining (Reijula 2021).
To go into some more detail, consider the analytically (but not always chronologically) first stage in formulating a mechanistic explanation or phenomenal decomposition.This task consists in individuating the phenomenon to be explained in a manner that is fine-grained enough for the purposes at hand, distinguishing it from related phenomena by the causal profile of behaviors that they give rise to and identifying the range of systems that manifest this phenomenon.In the cases of interpersonal social coordination that I have discussed in this article, this task has been addressed primarily -but not exclusively -by behavioral scientists.For example, developmental psychologists have distinguished between the stage of primary intersubjectivity (Trevarthen and Aitken 2001) and the nine-month revolution (Tomasello 2019) by the types of interpersonal social coordination that infants of a certain age manifest, such as seeking and maintaining eye contact, imitation and joint attention.For their part, evolutionary anthropologists have distinguished between instrumental and non-instrumental coordination by the presence of an extrinsic reward (Warneken et al. 2006).They have also studied under what conditions infants and non-human primates engage in these forms of social coordination.For example, there is some evidence that chimpanzees may be more skilled collaborators with human agents in laboratory conditions than with conspecifics or in the wildperhaps due to a lack of trust in sharing rewards (see Tennie, Jensen and Call 2016).However, while these studies shed light on what forms of social coordination various types of agents engage in, they do not yet identify the mechanisms by means of which these agents are (or are not) able to engage in the forms of social coordination in question.For this, a mechanistic decomposition is needed.
The analytically second stage in formulating a mechanistic explanation consists in mechanistic decomposition, or describing (in satisfactory detail) the entities and activities that are responsible for the phenomenon under investigation.The process of mechanistic decomposition can be further distinguished into a structural decomposition (identifying constituent entities) and a functional decomposition (identifying constituent activities).These two stages of mechanistic decomposition need not take place at the same time nor be carried out sequentially.For example, the identification of mirror neurons in the ventral premotor cortex of macaque monkeys (Rizzolatti and Sinigaglia 2010) can be understood as an instance of structural decomposition, although the range of activities that mirror neurons play in human social coordination remains a matter of considerable debate (Hickok 2014).Conversely, describing perception-action matching as one of the constituent activities underlying emergent coordination (Knoblich, Butterfill and Sebanz 2011, 69-71) can be understood as an instance of functional decomposition, although it is not certain which other areas of the brain engage in perception-action matching over and above mirror neurons narrowly conceived (Hari and Kujala 2009, 461-465).Interestingly, despite their background in cognitive science, Knoblich et al. (2011) suggest that the proximate mechanisms of interpersonal social coordination may extend beyond the brains and bodies of the individuals involved in the form of common object affordances, i.e. physical entities in the environment that can be manipulated by two or more individuals in a manner in which they cannot be manipulated by one individual acting alone (e.g. a see-saw or a two-handled saw) (see also Vesper et al. 2010). 9 The distinction between phenomenal and mechanistic decomposition raises the question of how complete mechanistic explanations should be, given that the entities and activities that serve to mechanistically decompose a phenomenon may often themselves be mechanistically decomposed into further underlying entities and activities (e.g.mirror neurons can be decomposed into sub-cellular building blocks).The proper degree of mechanistic detail seems to depend on the explanatory context and the epistemic interests of researchers (Ylikoski and Kuorikoski 2010), and even in cognitive explanations of behavioral or social phenomena, it is not always necessary to go down to the neurobiological level of the human brain (Sarkia, Kaidesoja and Hyyryläinen 2020).Nevertheless, an explanatorily adequate mechanism model should arguably always localize the activities that are identified at a particular level of mechanistic decomposition into entities that perform those activities at that level and vice versa (Boone and Piccinini 2016).Problematically, Knoblich et al. (2011) seem to depart from this intra-level constraint on mechanistic decomposition by using the term mechanism loosely to refer to either entities (e.g. common object affordances) or activities (e.g.perception-action matching) that play a role in interpersonal social coordination, without drawing a clear separation between the two.This may lead to some confusion, as it is not clear, for example, how common object affordances depend on the cognitive processes of individuals and to what extent they play an autonomous role in interpersonal social coordination.
The preceding discussion has sought to illustrate how the mechanistic framework of explanation can provide guidance to both behavioral and cognitive scientists about what types of questions their accounts have been successful in answering, and where inputs from other scientific disciplines may be needed.This being said, the distinction between mechanistic decomposition and phenomenal decomposition that I have made use of should not be understood as a straitjacket that scientists must conform to, because behavioral scientists can also sometimes contribute to mechanistic decomposition, and cognitive scientists can sometimes contribute to phenomenal decomposition.For example, behavioral scientists have put forth hypotheses about the types of capacities for intention reading (Tomasello 1999) and sharing intentionality (Tomasello 2019;Tomasello et al. 2005) that are required for various forms of interpersonal social coordination, although they have typically not associated these capacities with detailed mechanistic hypotheses about specific brain regions or environmental mechanisms that are responsible for the exercise of the capacities in question.For their part, cognitive scientists have distinguished between the phenomena of emergent coordination and planned coordination (Knoblich, Butterfill and Sebanz 2011) partly on the basis of underlying cognitive mechanisms but also by behavioral criteria.Although the distinction between phenomenal decomposition and mechanistic decomposition is not intended to carve out a categorical division of labor between the behavioral and cognitive sciences, it would still clearly be a mistake to believe that one is engaged in mechanistic decomposition when one is in fact engaged in phenomenal decomposition, or vice versa.Thus, distinguishing between these two types of inquiry can shed light on the nature of interdisciplinary investigation and contribute to a more efficient division of labor between behavioral and cognitive scientists studying interpersonal social coordination.

Exploring the Boundaries of Mechanistic Integration
As noted in the introductory section of this paper, talk of mechanisms and mechanical explanation has become more prevalent across many different scientific disciplines during recent years (Bechtel and Richardson 2010;Craver and Darden 2013;Glennan 2017).The study of social coordination is no exception to this broader interdisciplinary trend, with researchers from cognitive science to social psychology and even philosophy ever more likely to describe their approaches as identifying mechanisms for various aspects of social coordination (e.g.Thonhauser and Weichold 2021).Unfortunately, this widespread convergence towards a new meta-theoretical terminology has not always (outside the philosophy of science, that is) been accompanied by equally explicit methodological reflection on what mechanisms are, how to integrate mechanisms at different levels of analysis or how mechanistic explanation relates to certain alternative ways to describe the activities that take place within and between disciplines, such as theoretical modeling, conceptual analysis or hermeneutic interpretation.For this reason, it seems to me that a superficially shared terminology may hide deeper underlying disagreements about the most appropriate ways to explain social coordination, and some approaches that are loosely described as mechanistic may turn out (upon closer analysis) not to satisfy the types of adequacy criteria that are associated with mechanistic explanation after all.This section will seek to substantiate these claims, taking the integrative review by Thonhauser and Weichold (2021) introduced at the beginning of this paper as my primary foil.
Remember that the goal of Thonhauser and Weichold's (2021, 1) paper is to integrate six different traditions of thought that have each 'carved out fruitful mechanisms for explaining collective action, but is also faced with limitations': collective intentionality, team reasoning, social identity theory, ecological social psychology, participatory sense making and practice theory.However, I will argue below that the first two traditions of thought are not necessarily in the business of mechanistic explanation to begin with -given a suitably demanding notion of mechanism -while the latter approaches (as they are described in the paper) go only some way towards mechanistic integration, insofar as they pose few constraints on one another (mere mutual compatibility in the sense of an absence of contradictions is the weakest possible form of coherence analyzed by Danks (2014) and Miłkowski (2016)).To counterbalance these critical remarks, I should emphasize that my goal is not to otherwise undermine the intellectual merits of the approaches that Thonhauser and Weichold (2021) discuss -my point is simply that not all attempts at understanding social coordination are best described in terms of mechanisms and mechanistic explanation.Accordingly, while the previous sections of this paper have defended the prospects of mechanistic integration across different scientific disciplines, the present section will seek to identify the boundaries of mechanistic integration and circumstances where it may seem less than fully promising.
Let us begin with the collective intentionality approach.Thonhauser and Weichold (2021, 3) correctly recognize that much of the philosophical literature on this topic began with an attempt to distinguish genuinely intentional collective actions (e.g. two people taking a walk together) from merely accidentally coordinated actions (e.g. two people walking alongside one another).This attempt might, although with some strain 10 (because it does not involve any empirical investigation), be described as a first pass at a phenomenal decomposition of socially coordinated behavior into a variety of possible types.Thonhauser and Weichold (2021, 3) follow Schweikard and Schmid (2013) in classifying alternative philosophical approaches to collective intentionality based on whether they focus on the subject (e.g.Gilbert 1990), mode (e.g.Searle [1990Searle [ ] 2002) ) or content (e.g.Bratman 1993) of intentional states.Their assumption seems to be that such contrasting analyses of collective intentionality pick out alternative mechanisms for collective action.However, what they fail to pay attention to is that the subject-mode-content distinction primarily reflects the subject-mode-content structure of linguistic attributions of collective intentionality (e.g.'We' (subject)/'intend' (mode)/'to walk together' (content)), rather than any set of well-established explanatory primitives that are used in cognitive science. 11Accordingly, it seems that we would at least deserve a closer analysis of the relation between conceptual analysis of folk psychological categories and cognitive scientific explanation before accepting that philosophical analyses of collective intentionality pick out mechanisms for collective action (for one take on this issue, see Sarkia (2023)).
The second approach that Thonhauser and Weichold (2021, 5) discuss is the team-reasoning approach in economics.Here, Thonhauser and Weichold (2021) join the ranks of numerous philosophers, who have distilled the complex technical details behind game-theoretic models of team reasoning to the rather simple idea that team reasoners proceed by first deciding what their team should do and then working out their share in the optimal team decision.What they fail to pay attention to is that the tractability assumptions underlying many prominent models of team reasoning are at odds with the assumptions underlying their other preferred candidate for theoretical integration, collective intentionality theory.For example, Bacharach's (1999) model of team reasoning is based on a Bayesian approach to game theory, where degrees of beliefs are assigned to agents on the basis of hypothetical lotteries that they would participate in.By contrast, philosophers writing on collective intentionality have concerned themselves almost exclusively with binary (all-ornothing) attitudes, such as collective acceptance (Searle 2010;Tuomela 2013). 12Of course, it is arguably an open question whether agents operate primarily with arbitrarily fine-grained degrees of belief or with binary attitudes.However, it seems that we would deserve an explanation of how this apparent contradiction is to be resolved if we really think of the two approaches as identifying complementary mechanisms for collective action.Again, I do not wish to question the intellectual merits of either the team reasoning approach or philosophical analyses of collective intentionality, but I think that the Thonhauser and Weichold (2021) are too optimistic in their assessment of their mutual compatibility.My own view is that accounts of collective intentionality and team reasoning are both best understood as abstract and idealized theoretical models (Giere 1988;Hausman 1992)albeit ones that involve contrasting abstractions and idealizations -but that they are not necessarily best construed as mechanistic models in particular (see Sarkia 2022).
The next three research programs that Thonhauser and Weichold (2021) consider are social identity theory, ecological social psychology and enactive approaches to participatory sensemaking.Here, the complementarity of the approaches seems to be genuine.According to Thonhauser and Weichold (2021, 5-6), social identity theory 'strives at a psychological understanding of social groups based on the self-identification of individuals' but 'only accounts for the intellectual aspects of group participation' (cf.Turner et al. 1987).Thonhauser and Weichold (2021, 6) suggest correcting this cognitivist bias through ecological social psychology, which 'characterizes the environment in terms of possibilities for action that exist in relation to an organism', who is in a 'constant loop of perceiving (multiple) affordances, reacting to them, perceiving new affordances, and so on' (cf.Gibson 2015;Vesper et al. 2010).However, given that ecological social psychology goes to the other extreme from social identity theory by under-intellectualizing the mind, enactivist approaches to participatory sense-making allow us to understand how the 'perspective of a collective is dynamically constituted through processes of collective sense-making' (cf.De Jaegher and Di Paolo 2007).The types of dynamical processes of entrainment and perception-action matching that were discussed under the banner of emergent coordination in Section 4 of this article arguably play an important role in participatory sense making, while the collective affordances of ecological social psychology were also discussed there.Thus, Thonhauser and Weichold (2021) seem to be on the right track here: the three traditions really do seem to fill in each other's lacunae and are not mutually incompatible.Thus, they are plausible candidates for mechanistic integration, even if many details remain to be filled out.
The final research program that Thonhauser and Weichold (2021, 9) consider forms an interesting case with respect to the limits of mechanistic integration.At first sight, practice theory, which has its roots in a mix of continental and anglophone social theories including the works of Wittgenstein, Heidegger, Bourdieu, Giddens and Butler, may seem like an unlikely candidate for a mechanistic interpretation.For example, Thonhauser and Weichold (2021, 9) quote Schatzki (1996, 89) as saying that practice theory describes individuals as engaged in a 'temporally unfolding and spatially dispersed nexus of doings and saying'.Thonhauser and Weichold (2021, 9) go on to emphasize that a practice is to be understood as 'more fundamental than individuals and "the society"', since 'individual human beings only become the subjects they are in and through social practices, and . . ."the society" is nothing but a configuration of social practices'.This level of abstraction may seem to be in fundamental tension with the idea of mechanistic decomposition.However, an important group of cognitive sociologists have argued during recent years that many central insights of practice theory can in fact be accommodated within a cognitive scientific framework (for an explicitly mechanistic interpretation of these results, see Kaidesoja, Hyyryläinen and Puustinen 2022).For example, Vaisey (2009) interprets Bourdieu's notion of habitus and Giddens's conception of practical consciousness under the framework of dual-process theories of cognition and relates it to the motivation-justification split in the explanation of action (with implicitly learned social practices serving to explain behavior and cognition serving a role in its post hoc justification). 13 Lizardo (2019), in turn, connects Bourdieu's habitus to patterns of embodied cognition and associated theories of meaning and metaphor (Barsalou 1999;Lakoff and Johnson 1999).And Theiner and Fogle (2018) respond to Turner's (1994) challenge regarding how habitus is transmitted between individuals by building on mechanisms of extended cognition, including cultural niche construction theory, scaffolding and entrenchment (Laland and O'Brien 2011;Wimsatt and Griesemer 2007).Accordingly, when suitably interpreted, it seems that practice theory might well be integrated as part of a broader mechanistic theory of social coordination, although some more hermeneutically oriented forms of practice theory are certain to resist a strictly mechanistic interpretation.
To draw together the results of the preceding discussion, my goal in this section has been to temper too much optimism about the prospects of interdisciplinary integration under a mechanistic framework by indicating how loose talk of mechanisms can obscure differences between mechanistic explanation and other forms of scientific practice (e.g.theoretical modeling and conceptual analysis) and how it can hide from view genuine sources of disagreement and incompatible tractability assumptions that occur between rival theoretical approaches.Moreover, I have attempted to show that the high degree of abstraction that is involved in some social scientific theories is not as such an obstacle to mechanistic integration, since even some comparatively highlevel theories, such as practice theory, have been interpreted mechanistically (even if there are surely also forms of practice theory that do not lend themselves equally easily to a mechanistic interpretation).Finally, I have attempted to make clear that the task of integrating different approaches to interpersonal social coordination does not end with my paper, since there are many further approaches, such as social identity theory, ecological psychology and participatory sense-making, which my paper has not discussed in detail.Indeed, the primary goal of my paper has not been theoretical but methodological -to show how the mechanistic framework of explanation can play a role in interdisciplinary integration and what possible epistemic risks may be associated with it (including the risk of interpreting mechanisms too liberally).

Conclusions and Summary
Many of the most celebrated examples of mechanistic explanation in the history of science have involved interactions between different scientific disciplines studying closely related phenomena.For example, the evolutionary synthesis connects Darwinian mechanisms of natural selection across populations of individuals to hereditary processes at the level of the gene.Accordingly, increasing mechanistic understanding of the world can be regarded as one important aspect of scientific progress.However, interdisciplinary integration may also sometimes fail, or advances in one field may force substantive revision of ideas that have been taken as established knowledge in another scientific field.By studying such cases, we can learn how different disciplinary bodies of knowledge and new empirical discoveries interact and constrain one another.We may also learn where the limits of mechanistic integration lie and what types of theoretical approaches may be best described in terms of some alternative methodological or meta-theoretical vocabulary altogether.Accordingly, while mechanistic explanations are ubiquitous and useful for intervening in the world, there is no reason to assume that all intellectually satisfactory approaches to social phenomena must involve mechanistic decomposition.Describing them in terms of mechanisms and mechanical explanations may serve to distort the phenomena under investigation and impose success criteria that are inappropriate for their study.Thus, it is equally important to recognize the limits of mechanistic integration, as well as to appreciate its unifying potential.
The case of interpersonal social coordination, which I have discussed in this article, provides one example of an increasingly prominent field of research where interdisciplinary integration under a mechanistic framework is far from straightforward.With regard to my case study, I have argued that certain ideas about the phylogenetically distinctive character of human capacities for social coordination that have been popular among developmental psychologists and evolutionary anthropologists subscribing to the so-called Vygotskian intelligence hypothesis are not robustly supported by cognitive scientific research on the mechanisms of interpersonal social coordination.On the other hand, I have also argued that this body of cognitive scientific research is still compatible with the weaker social motivation hypothesis, which is entailed by most versions of the Vygotskian intelligence hypothesis but is not equivalent to it.Accordingly, it seems to me that integrating these different disciplinary bodies of knowledge still provides us with a more complete and comprehensive picture of interpersonal social coordination than we would have on the basis of the type of understanding coming from any single scientific discipline alone.Moreover, understanding different research programs on social coordination as contributing to the mutually complementary enterprises of mechanistic decomposition and phenomenal decomposition allows us to appreciate the division of labor between behavioral and cognitive scientists studying social coordination.This may lend them confidence that their approaches will eventually fit together, despite the fact that they often pursue their proximate explanatory tasks in only partly overlapping institutional environments.Thus, it seems fair to say that even if much work remains to be done in integrating different disciplinary approaches to social coordination, the mechanistic approach to explanation provides one promising avenue by which such interdisciplinary integration can be pursued.

Notes
1.In addition to these disciplines, there are lively research programs concerning the mechanisms of social coordination in experimental economics and evolutionary game theory (see e.g.Bicchieri 2006;Gintis 2009;O'connor 2019;Ross 2012;Roth 2016).I will not discuss these research programs in the present paper.2. To be precise, Thonhauser and Weichold's (2021) focus is on mechanisms of collective action, rather than interpersonal social coordination (which is the focus of this paper).However, the capacity for social coordination is of course a prerequisite for many forms of collective action, so there is significant overlap between the two.3.By speaking of interpersonal social coordination, I mean to convey that I am interested in coordination that takes place between physically interacting individuals, not the type of coordination that takes place, for example, in the context of market institution.To avoid cumbersome expressions, I will occasionally drop the qualifiers 'interpersonal' or 'social' and speak simply of coordination.4. Miłkowski (2016) suggests that Thagard's (2000) computational account of coherence can be used to provide a general account of the roles that positive and negative constraints play in mechanistic explanation.Miłkowski views constraints as primarily a representational affair that relates to models of mechanisms, while Craver (2007) seems to view constraints as stemming largely from the ontic nature of the mechanisms themselves. 5.For example, Tomasello et al. (2005, 675) state that 'human beings, and only human beings, are biologically adapted for participating in collaborative activities involving shared goals and socially co-ordinated action plans (joint intentions)' (see also Moll and Tomasello 2007).6.Indeed, one of the most prominent proponents of the Vygotskian intelligence hypothesis, Michael Tomasello (1999;Tomasello and Call 1997) argued only a few decades ago that non-human primates do not understand intentions or false beliefs -claims that have since then been (to his credit) conclusively refuted by many of his own studies (cf.Call and Tomasello 2008;Krupenye et al. 2016).7. Tomasello et al. (2005, 680) suggest that species-distinctive capacities for shared intentionality and social coordination are responsible for numerous highly abstract features of human societies, which are as farreaching as 'money, marriage, and government, which only exist due to the shared practices and beliefs of a group'.8. Images reproduced from Warneken et al. (2006).
(a) What do central forms of interpersonal social coordination consist in (e.g.instrumental and non-instrumental coordination; emergent and planned coordination)?(b) What types of agents engage in such forms of coordination (e.g.neonates, toddlers and/or non-human primates, such as bonobos and chimpanzees)?(c) What types of cognitive mechanisms give rise to such forms of social coordination (e.g.