2. Group hunting in wolves and dolphins

While group hunting in early Homo sapiens and late hominins (such as Neanderthals) involved a unique level of sophistication—refined weapons, communication, planning, complex strategy—group hunting behaviors occur widely across the mammalian class. When we explain these behaviors, and make inferences about the cognitive capacities of the individual animals engaged in these behaviors, there is a broad question we can ask: are these behaviors merely the aggregative product of individual decisions, or do they require recognition and coordination between members of the group? To put it in the framework of the current investigation, do nonhuman mammals exhibit a kind of joint agency when hunting in groups? It could be that what looks like group hunting is best explained by opportunistic individuals joining an ongoing activity to maximize their own chances of filling their bellies. On this opportunistic recruitment interpretation, recognition and coordination between members of the group, if they occur at all, would amount to an awareness that other competitors are near and potentially a threat to an individual’s take. Alternatively, individual animals may be coordinating with each other in a way that produces a group behavior that requires an appeal to some sort of joint agency. Evidence that animals recognize one another, fill different strategic roles in the hunt, respond fluidly to the actions of both conspecifics and prey, and communicate in ways that direct the hunt would support a joint agency hypothesis for group hunting in these species. Such evidence should also show that group hunting behavior cannot be easily accounted for simply by aggregating individual, context-independent, and evolutionarily selfish decisions.

We suspect that forms of joint agency occur widely in mammals, particularly in group hunting scenarios. To support this suspicion, we turn to two case studies. Taking a closer look at hunting in wolves and dolphins show that there is convincing evidence that supports genuine cooperation, and thus a form of joint agency, over opportunistic recruitment in some of their more complex foraging behaviors.

The reintroduction of wolves (Canis lupus) into Yellowstone National Park created a new study population, providing a rich source of observational data on wolf social behavior. As a result, biologists now have an extensive and unique data set on wolf hunting behavior. Two key observational studies, one on wolves hunting elk (MacNulty et al. Reference MacNulty, Smith, David Mech, Vucetich and Packer2012) and another on hunting bison (MacNulty et al. Reference MacNulty, Tallian, Stahler and Smith2014), provide some of the best evidence for evaluating whether, and to what extent, wolves need to work together as a pack to bring down the different prey types.

While wolves often hunt both prey types in packs, there are differences between how the wolves pursue the different prey. Elk are easier prey, and a lone wolf can sometimes bring down an elk successfully. Wolves are much more likely to hunt in packs, and their chance of success when hunting elk increases markedly. However, there is a threshold size—four wolves—where increasing the group size of the hunting party no longer increases the chance of success; in fact, there is a statistical trend of declining success probability in bigger groups (MacNulty et al. Reference MacNulty, Smith, David Mech, Vucetich and Packer2012, 79). The explanation given for this trend is observed free riding in larger groups. Individual contribution drops in larger hunting groups, and the drop is more dramatic for wolves lower in the dominance hierarchy or without cubs. The study ranked individual wolves by hunting prowess and so could evaluate whether inept hunting groups or interference from too many hunters might explain the trend. Neither hypothesis fares well given the data. The decline in success probability occurs faster in competent groups compared to inept groups, and there is no evidence of interference increasing across group size (MacNulty et al. Reference MacNulty, Smith, David Mech, Vucetich and Packer2012, 81).

Wolves obviously benefit from working together to hunt elk, and such hunts involve fluid coordination of action as they pursue their prey, a feature we would expect in genuine cooperation. Yet this evidence is not decisive. The wolves are, in an evolutionary sense, pursuing their own interests. Joining a small group increases the chance of successfully capturing elk, but as the group gets larger, individuals start to free ride and avoid the costs of the cooperative hunt. This suggests that even though wolves exhibit some group-level cooperation, individual opportunistic recruitment to the hunt may be the best explanation for the pattern of behavior: cooperate when it makes a substantial impact or if in substantial need; otherwise, ride free.

The situation changes when wolves hunt bison. Bison present a more significant challenge because they can easily injure or even kill attacking wolves. It is extremely rare for lone wolves to approach bison, and MacNulty et al. (Reference MacNulty, Tallian, Stahler and Smith2014) made zero observations of a single individual successfully bringing one down. Perhaps unsurprisingly, more complex cooperation and pack interaction are necessary to hunt the larger, more dangerous prey. MacNulty et al. studied how group size affected hunting success for wolves pursuing bison. In this case, the probability of success scales with group size up to eleven wolves. After that threshold, probability success levels off but doesn’t show the same decreasing trend observed for elk. Also, wolves show restraint when in small groups. In groups of four wolves or fewer, the probability of starting pursuit of bison is much less (approximately half) than the probability of pursuit of elk (MacNulty et al. Reference MacNulty, Tallian, Stahler and Smith2014, figure 3).

The study attributes these trends to two features. First, given the very low probability of success for solo hunters or small groups, hunting bison requires more cooperative investment by the pack. The individual wolves track their group size and adjust their behavior (whether to pursue) accordingly. Second, because bison are more dangerous prey, the wolves adjust their spatial configuration and responsiveness to the pack during the hunt (MacNulty et al. Reference MacNulty, Tallian, Stahler and Smith2014, 5–6). Video supplemental to the study shows an example of a successful bison hunt. The wolves surround a group of bison and take turns harassing individuals over several minutes. In response to the harassment, individual bison charge at the wolves, and the group of bison eventually starts to splinter, with some individuals moving away to avoid the wolves. The hunt ends when a bison calf panics and bolts—the wolves immediately pursue and bring down the calf in a matter of seconds.

Wolf hunting behavior in the bison case demonstrates a higher degree of cooperation and thus suggests at least some elements of joint agency (i.e., cognitive capacities and dispositions that could explain the cooperative behavior in view). The wolves show a tendency to avoid starting bison hunts when groups are small; the success of the hunt scales with group size up to eleven wolves; there is no consistent trend of defection by individuals in larger groups; and their behavior during the hunt is more responsive to packmates. If we want to exclude this as a case of genuine joint agency, we need to articulate the appropriate criteria for drawing a principled line between (say) uniquely human joint agency and cooperative hunting among Yellowstone wolves. There is one criterion that is a good candidate for identifying the kinds of cognitive capacities that support joint agency, that would draw the right line between wolves and humans: communication between individuals before, during, or after the hunt. Although wolves are obviously tracking their packmates and responding accordingly, there is no evidence that they are engaging in explicit signaling to coordinate the hunt. Requiring complex or linguistic communication for joint agency would limit its explanatory power by setting the bar for joint agency rather high—hominin ancestors were engaging in group cooperative activities long before there is evidence of language (Sterelny Reference Sterelny2021; Planer and Sterelny Reference Planer and Sterelny2021). Yet one might argue that explicit communication is essential to planning and executing genuine cooperation or truly joint agential behavior. However, this creates difficulties, because there are cases where we see both highly cooperative hunting and specific signaling between group members in several dolphin species.

Two studies done by Benoit-Bird and Au (Reference Benoit-Bird and Au2009a, Reference Benoit-Bird and Au2009b) document a cooperative foraging strategy in spinner dolphins (Stenella logirostris) off the coast of Hawai‘i. The strategy involves large groups of dolphins (sixteen to twenty-eight) working together to herd fish before feeding. The overall behavioral strategy has four stages and requires careful coordination in four dimensions. Dolphins start with a wide line oriented like a V in the vertical dimension of the water column. When the group encounters a shoal of fish, dolphins adopt a tight line formation to bulldoze the prey. The group then transitions to a circle formation, though each dolphin remains at the same depth, effectively forming a cylinder around the prey. The circle then tightens, and pairs of dolphins, each from an opposing position of the circle formation, take turns darting into the circle to feed. As the biologists summarize:

There is also some evidence that there is significant acoustic coordination, perhaps even communication between the dolphins during the search and feeding. During transitions between stages, there are spikes in echolocating clicks (Benoit-Bird and Au Reference Benoit-Bird and Au2009b). The evidence for communication is not definitive, because disentangling communication from individual echolocation can be tricky and the whistles that tend to be used for communication are absent during the hunt. It is not clear whether the dolphins are signaling or using each other’s echolocation clicks as cues for the position of conspecifics. A further concern with the studies is that position information is inferred from sonar rather than directly observed. Nonetheless, the studies document large group hunting in a marine mammal that minimally involves complex spatial and temporal coordination, and it coheres with observations of hunting and communication in other dolphin species.

Hamilton et al. (Reference Hamilton, Gazda, King, Starkhammer and Connor2022) document a hunting strategy in bottlenose dolphin (Tursiops truncates) off the Florida Keys that they label “driver-barrier feeding.” In this strategy, one dolphin acts as the “driver” and herds small prey fish toward a line of waiting dolphin, the “barrier,” driving them into the air where the dolphins jump to catch them. This hunting strategy exhibits role specialization and coordination among the group. The dolphins sort into a driver and a barrier, and the dolphins in the barrier stop using echolocating clicks—only the driver dolphin uses echolocation clicks during the hunt. In addition, there is clear evidence of communication: the driver dolphin whistles before beginning the hunt in a way that is statistically different from behavior in other contexts (Hamilton et al. Reference Hamilton, Gazda, King, Starkhammer and Connor2022, figure 3). The study suggests that whistles may play a role in signaling motivation, recruiting conspecifics, or coordinating behavior (Hamilton et al. Reference Hamilton, Gazda, King, Starkhammer and Connor2022, 6). All of these involve communication, and the pattern of behavior fits with the role whistling plays in dolphins generally, as Hamilton et al. note. Furthermore, there is also new evidence that dolphins tailor their communication to specific individuals: mother dolphins modulate their whistling when signaling to their own offspring rather than other dolphins in the pod, effectively using “motherese” when whistling to their own calves (Sayigh et al. Reference Sayigh, Haddad, Tyack, Janik, Wells and Jensen2023). In short, dolphins regularly communicate, and they do so during complex group tasks.

Evidence indicates, then, that sophisticated forms of cooperative behavior, highly suggestive of joint agency, have evolved outside of the human lineage. The two cases we discuss here exhibit different forms of coordination deployed to achieve a joint action with the specific form of coordination facilitated by the affordances in each lineage. If we were to expand the scope of inquiry across the tree of life, we would find further gradations and variation in coordinated behavior. This places pressure on any bright-line accounts that aim to distinguish “true” joint agency, and the sophisticated cognition capacities that produce it, from the cooperative hunting behaviors in wolves and dolphins and the cognitive capacities implicated by their behavior. One might respond, of course, that philosophers and evolutionary biologists are simply pursing different explanatory projects.Footnote ² This may be so, but this response leaves important questions unresolved regarding the forms (joint) agency involved in cooperative behavior may take and what types of cognitive capacities might underlie these behaviors. We think these are questions worth answering, as we argue in the next section.

3. Thick versus thin and the diversity of group action

We have argued that wolves and dolphins exhibit forms of joint agency, in the sense that their behavior requires recognition of others, communication, and coordinated, goal-directed adjustment of behavior in light of these. The following important questions arise at this point: What sort(s) of joint agency or cooperation are we studying in animals like wolves or dolphins? How does it compare to the form(s) of joint agency or cooperation we see in other animals, especially humans? How are we to explain the similarities and differences between forms of joint agency found across nature? Evolutionary explanations seek primarily to trace changes in forms of cooperative behavior and the emergence of more sophisticated forms of joint agency. Such explanations have to be sensitive to these questions of categorization and variance (see Moore Reference Moore2017).

The current literature offers no consensus on these questions. The diversity of group behavior is mirrored by a range of approaches to explaining joint agency and cooperation. But various approaches to categorizing forms of joint agency can be mapped along three orthogonal dimensions. The first concerns the level of sophistication of the joint agency or cooperation in view. One way to approach this issue is to focus on the complexity or flexibility of various cooperative behaviors. Much of the philosophical literature focuses less on this and more on the conditions for “bona fide” joint agency (and therefore special, real, or truly collaborative cooperation) of the sort that one sees in paradigmatic human cases (see, e.g., Joyce Reference Joyce2006; Kaufmann Reference Kaufmann2015, Reference Kaufmann2016; Ludwig Reference Ludwig and Fiebich2020). And some approaches seek to characterize more basic forms of joint agency and cooperation. Sometimes these are called “proto” joint agency or “minimal cooperation” (Paternotte Reference Paternotte2014).

A second dimension concerns the cognitive capacities assumed or argued to be critical for joint agency or cooperation. Some approaches make minimal assumptions about the cognitive capacities of the agents involved. This is true of game-theoretic accounts, for example (Gold and Sugden Reference Gold and Sugden2007; see also section 4). Other approaches attempt to make explicit the cognitive capacities required. This is true of more sophisticated as well as more minimal accounts of joint agency. Butterfill and Pacherie (Reference Butterfill, Pacherie and Fiebich2020), for example, engage in an exercise of creature construction and argue persuasively that forms of joint agency and cooperation could emerge in an agent that lacked any capacity to attribute mental states, provided that the creature could track the goals of conspecifics, and provided that certain heuristics (e.g., shared perceptual mechanisms of salience tracking as a proxy for common knowledge) typically hold. They then progressively add cognitive capacities to their creature, and they link these more sophisticated capacities to more sophisticated forms of cooperation.

A third dimension concerns a distinction between internalist and externalist approaches to explaining social phenomena in general. Internalist views appeal to internal items like mental states or representational contents (Bratman Reference Bratman1993; Gilbert Reference Gilbert2009). Externalist views appeal to extramental items like laws and contracts, material resources, or the organizational structure of groups and group roles (Epstein Reference Epstein2015; Ritchie Reference Ritchie and Fiebich2020; Westra and Andrews Reference Westra and Andrews2022).

We highlight these dimensions primarily to isolate explanatory emphases and strategies one finds in recent literature. To some extent, treatments of joint agency and cooperation may draw on all three dimensions. Both externalists and internalists, for example, can and sometimes do draw on accounts of the cognitive capacities involved in cooperation. And interestingly, philosophical accounts of sophisticated joint agency and cooperation tend to be heavily internalist, while approaches that seek to characterize more minimal forms of joint agency and cooperation often appeal, at least in part, to externalist features. Arguably, however, evolutionary explanation of different forms of joint agency and their relationships cannot afford to ignore any of the aforementioned dimensions. Indeed, in our view, understanding these different dimensions of explanations of joint agency affords the ability to notice explanatory resources and opportunities that a reliance on leading philosophical accounts of bona fide joint agency may hide.

If one follows philosophers who seek to explain the most sophisticated forms of joint agency, one might think that these most sophisticated forms are the most relevant target for evolutionary explanation. We argue, instead, that from the perspective of evolutionary explanation, a normative dispute about the conditions for bona fide joint agency is less important than an investigation into the fitness of different forms of group activity in different types of circumstances and how these may lead to the evolution of more sophisticated forms of sociality. Different forms of cooperation can be found across the animal kingdom (Moore Reference Moore2018).

If joint agency and forms of cooperation vary across different dimensions, then the features and capacities constitutive of joint agency may vary across species, rendering the drawing of any bright line between bona fide and proto joint agency less than fruitful for evolutionary explanation. Though philosophical accounts of joint agency can be useful for some purposes, our understanding of the evolution of coordination can benefit from more incrementalist (e.g., Moore Reference Moore2017), and more pluralist, approaches.

4. An evolutionary approach to joint action

Evolutionary game theory provides a natural framework for analyzing joint action in a comparative way. Joint action, especially the sort we see in complex cooperation or divisions of labor, involves strategic interactions between multiple individuals. Game theory provides a way of representing the incentives of the participating individuals and, combined with a dynamics, identifies when cooperative behavior will invade, remain stable, or collapse. Most thinkers on human evolution have embraced game-theoretic characterizations of the cooperative and coordination challenges that faced human ancestors and that an appeal to joint agency purports to solve. One path toward defending the existence and importance of joint agency involves identifying complex coordination and cooperation problems that the capacity enables organisms to solve. Ideally, we would be able to characterize these as evolutionary problems imposed by the challenges of an organism’s social environment, and game theory provides abstract and general tools to do this.

That said, using the formal framework of evolutionary game theory immediately raises a problem for bright-line accounts of joint agency: strategic interaction and cooperation are found across the tree of life from microbes to humans, and sometimes these behaviors are mediated by complex cognition (e.g., wolves, dolphins) and sometimes not (e.g., microbes). So, some lineages can solve these problems without expensive cognition. Are there any features of the strategic setup that require cognition, or is it the case that some lineages have the expensive equipment for other reasons and simply deploy it to handle strategic interactions? In effect, we must confront a version of the problem that Gold and Sugden (Reference Gold and Sugden2007) raised for philosophical accounts of shared or collective intentions that produce group action. They point out that, according to many philosophical accounts, two individuals engaged in Nash equilibrium behavior look like they possess a collective intention to produce the Nash equilibrium. At a Nash equilibrium, players are making simultaneous best responses to the strategies of their interaction partners; thus no one player can change their strategy and improve their specific payoff. What counts as “rational” action and a solution to the game depends on the strategy choices of all the players involved. This interdependence of action seems to meet the criteria for joint intentions or joint agency. Yet many bright-line accounts would classify these cases as involving a sort of pseudo-coordination and not as candidates for true group action, nor do they implicate joint agency despite the player’s best strategy choice depending on the choices of other players.

Consider Gold and Sugden’s (Reference Gold and Sugden2007) key example, a simple competition over resources as represented by the Hawk-Dove game. In this game, each player has a choice between an aggressive hawk strategy and a deferential dove strategy. Hawk beats dove and takes all the resource; doves split the resource evenly; dove loses to hawk but avoids any cost or injury; and finally, hawk–hawk interactions lead to the worst outcome, an escalating conflict that risks injury. Against a dove, you would rather play hawk, but against a hawk, you prefer to play dove and avoid conflict. This game has three Nash equilibria—pairs of strategies such that each player chooses the simultaneous best response to their opponent. There are two pure Nash equilibria where one plays hawk and the other dove and one mixed where each player randomizes between hawk and dove in a way that makes their opponent indifferent between their strategy choices. Gold and Sugden point out that pairs of individuals playing either of the pure Nash equilibria look to have a joint intention to bring about a group action: one player wins the resource, and the other walks away with nothing (111–12). Given their opponent’s play, each prefers their respective strategy of hawk and dove, and neither can do better by unilaterally changing their strategy. Thus the pair seems to collectively intend that they play the Nash equilibrium and act on these intentions to bring about a joint action. Yet this seems strange, especially for the dove player who loses the resource.Footnote ³ Evolutionary dynamics provide further support for Gold and Sugden’s diagnosis, as neither pure Nash equilibrium is evolutionarily stable. The challenge Nash equilibria behavior poses for joint intention based accounts of group action is even more pressing for the classic game of cooperation, the (one-shot) Prisoner’s Dilemma. This game has but one Nash equilibrium: both players defect. Yet accounts of joint action are usually invoked to explain how we can cooperate in these sorts of scenarios, not defect (see, e.g., Tuomela Reference Tuomela1992).

Perhaps another game would better capture the nature of joint agency in cooperation and avoid this unpalatable consequence. Many evolutionary theorists prefer the Stag Hunt as a representation of early cooperation in human ancestral lineages (Skyrms Reference Skyrms2004; Tomasello et al. Reference Tomasello, Melis, Tennie, Wyman and Herrmann2012; Sterelny Reference Sterelny2012). As Rousseau describes the scenario, each individual has a choice between cooperating with their fellows to bring down a stag or defecting to hunt rabbits on their own. The stag is the preferred quarry, but one’s partner must also hunt stag. Hunting rabbit guarantees prey, but of less value. In this game, cooperation is risky but stable, if achieved. Many have argued that the mutual interest of cooperation captured by the Stag Hunt best characterizes the problems human ancestors faced with respect to collective foraging, defense, and reproductive cooperation. This led us to become obligate cooperators and produced the scaffolding necessary to solve more difficult cooperative problems. That the Nash equilibrium for hunting stag counts as joint action seems reasonable, for this game represents a paradigm coordinated activity: group hunting. However, there are more Nash equilibrium solutions to the game. In particular, hunting hare is a Nash equilibrium, and this looks like the paradigm of individual, not joint, action. Furthermore, simply solving a Stag Hunt is not yet evidence that any particularly sophisticated cognitive capacity is at work, as chimps seem to solve the game in a way different from humans (Duguid et al. Reference Duguid, Wyman, Bullinger, Herfurth-Majstorovic and Tomasello2014). Any account of joint action requires the resources to sort Nash equilibria behavior into true and spurious joint actions, if evolutionary game theory is to be of any use.

One obvious solution to this challenge is to make some form of cognition necessary for joint action. While this would cohere with the philosophical approaches to joint agency, it is not an ideal solution for an evolutionary approach for two reasons. First, human cognitive evolution is tightly bound up with human cooperation and sociality. If certain cognitive capacities are constitutive of or necessary for joint action, then such an ability cannot serve the purpose of explaining the origins of complex cooperation and sociality. Joint action would have had to have evolved earlier in response to different selection pressures. Second, and relatedly, making cognition constitutive of joint action compromises the generality and power of evolutionary game theory for explaining the origins and stability of cooperative behavior. A key feature of almost all hypotheses concerning human evolution involves identifying the strategic contours of cooperation to set in motion a feedback loop or ratchet mechanism capable of producing smart, social hominins. The strategic features of behavior should have something to do with the evolutionary pressures on cognition and joint agency, so it would be a costly theoretical move to set evolutionary game theory aside as orthogonal. Having rich cognitive capacities—to form the intention to hunt stag and to communicate this to potential partners—can make it easy to achieve cooperation in the Stag Hunt. Yet we also must confront the fact that such capacities are by no means necessary. Evolutionary dynamics produce (possibly “minimal”) cooperation in the Stag Hunt with regularity when, for example, simple signaling or correlated interaction is introduced (Skyrms Reference Skyrms2004). Bacteria can solve the comparatively more difficult public goods scenarios (Inglis et al. Reference Inglis, Brown and Buckling2012), so solving a Stag Hunt is well within microbial reach. In fact, one of the attractions of evolutionary game theory is that it does not presume the rich rationality and common knowledge constraints common to economic game theory (Skyrms Reference Skyrms2014). Evolutionary dynamics find the stable cooperative equilibrium through either biological (sex and death) or cultural (learning and plasticity) means. Attributing joint agency to these systems is feasible, if we adopt a thin, deflationary account of the notion, but this gives up on the task of connecting joint agency to some suite of cognitive capabilities.Footnote ⁴

One strategy for avoiding such a thin characterization of joint agency yet adhering to a broadly comparative approach involves identifying criteria that determine a spectrum from “aggregative” group action to more “interactive” group action (Conradt and List Reference Conradt and List2009). The movements of bird flocks and fish schools are classic examples of aggregative group action, whereas symbiotic relationships across species or coordinating reproductive roles in cooperative breeding count as interactive. For the wolf and dolphin examples, the opportunistic recruitment interpretation would classify hunting closer to aggregative, whereas the careful coordination between individuals and communication, as actually observed in the wolf–bison or the dolphin prey herding behaviors, provide evidence that group action here is much more interactive. On this approach, true joint agency would be implicated in interactive group action. This is a promising way to connect more extrinsic contours of the strategic interaction with cognitive capacities that mediate behavior.

Such an approach does not (yet) commit to specific cognitive capacities for joint agency but requires identifying the criteria for interactive group action. Some characterization of cooperation would certainly be a part of this account, but additional criteria could help exclude many of the problematic Nash equilibrium cases. Drawing on the examples, careful or specific attention to the number and position of pack or pod mates, fluid and adaptive responses by the group to the situation with respect to prey type or other conditions, and communication between individuals are all good candidate criteria for distinguishing interactive from aggregative group action. Another set of candidate criteria involves causal interaction, spatial integration, and other features of biological individuality. A community of microbes engaged in a public goods game may each be pursuing their individual evolutionary interests and only cooperating or engaging in joint action in an aggregative way. Interactive joint action would involve, say, the tight, specialized coupling of metabolisms between host and symbiont. This combines the approach to biological joint agency with the evolutionary features that involve the emergence of new biological individuals. Okasha (Reference Okasha2018) defends this connection most explicitly, arguing that agents are biological individuals that exhibit sufficient “unity of purpose,” understood in terms of the selected functional integration of their components. This effectively mirrors a move seen in the philosophical literature that links group action to the metaphysics of agency (Rovane Reference Rovane2019).

This contrast between thin and thick accounts of joint agency presents a dilemma: either a (thinner) form of joint agency plays a broad explanatory role in the evolution of animal social lifeways or (thicker) joint agency is the product of a rich social cognition that is a rare, perhaps unique, evolutionary accomplishment. On the basis of the considerations raised thus far, opting for an account that can play a broad explanatory role is the best path forward. Forms of joint agency and cooperation may partake of diverse traits, cognitive capacities, and environmental resources. And if (as we have suggested) joint agency and cooperation exist on a spectrum, determining the cognitive capacities and environmental situations that support more and less sophisticated joint agency and cooperation is a difficult task, and the role of joint agency in evolution becomes a moving target. No single peculiar cognitive capacity would be necessary for joint agency. Instead, a range of capacities implicated in attention, communication, recognition, and sociality will matter to differing degrees in different cases.

5. Joint agency and human cognition

A common inferential pattern found in the study of animal cognition involves identifying—through a combination of designed experiments and careful observations—behavioral patterns, then postulating the minimal cognitive capacities necessary to produce those patterns. To take just one example, researchers designed experiments for the tool-using New Caledonian crows in an effort to assess their cognitive capacities for learning and solving novel problems (Taylor et al. Reference Taylor, Elliffe, Hunt and Gray2010). Based on the crows’ performance, researchers can make inferences about what sorts of learning mechanisms may be implicated in crow cognition. Much like tool use, cooperative behavioral patterns, even complex interactive kinds of cooperation, are not unique to the human lineage. Therefore, unless we opt for the thinnest of accounts, the existence of cooperative behavioral patterns alone is not sufficient for there to be selection pressures in place to support joint agency. So, what else is required for joint agency to emerge or be assembled? What sort of selection pressures are required?

At this point, some skepticism that joint agency is the product of a single, unified cognitive capacity is warranted. Consider what cognitive competences joint agency minimally requires in the context of a group hunt. An individual band member needs to recognize and keep track of other band mates and their positions; to be able to plan for or at least anticipate the actions of their fellow hunters; and to communicate those plans, anticipations, and positions in real time. Once the action starts, individuals may need to track gaze, to respond rapidly to changes in circumstance, to remember the plan, and to react accordingly. This is, by no means, a simple task, and it implicates a range of cognitive capacities that are subject to different sorts of selection pressures and deployed in a wide range of tasks beyond cooperation and social interaction. What’s more, as Butterfill and Pacherie (Reference Butterfill, Pacherie and Fiebich2020) demonstrate, similar forms of cooperation may be stabilized by importantly different cognitive capacities and strategies.

Rather than taking some class of capacities as a canonical package that is evolutionarily novel in humans, we hold that we owe an evolutionary account of the large number of cognitive capacities that make up the purported package. Furthermore, taken individually, components of the joint agency package are not unique to humans. Gaze tracking, for example, is well documented in several mammals and is prevalent in many primates (see, e.g., Tomasello et al. Reference Tomasello, Hare, Lehmann and Call2007 and the overview in Zuberbühler Reference Zuberbühler2008; for corvids, see Bugnyar et al. Reference Bugnyar, Reber and Buckner2016). In Tomasello’s work on this topic, he and his collaborators pick out differences between the gaze following of humans and the great apes, pointing to the need for an evolutionary story that accounts for such differences (Tomasello et al. Reference Tomasello, Hare, Lehmann and Call2007). This would not provide an evolutionary account of the emergence of joint agency but rather an evolutionary account of one part of the package of cognitive capacities that make up joint agency.

There are several alternate accounts of the evolution of the cognitive capacities underlying cooperative behavior in humans. Most are laid out in the context of accounts of the evolution of human uniqueness. The first step of this approach is to pick out a human trait of interest, say, elaborate cooperative behavior, and point to examples illustrating that the version of this behavior in humans is distinct. Furthermore, it is then argued that this distinct behavior, or behavioral suite, is key to our uniqueness. Finally, evolutionary accounts are offered for the relevant trait.

We are not challenging human evolutionary uniqueness here—we acknowledge that we deploy language, culture, and norms in ways that are not replicated anywhere in evolutionary history, as any cursory examination of the tree of life will reveal. However, we do challenge the claim that joint agential cooperation is the hallmark of human uniqueness.Footnote ⁵ Evaluating a uniqueness hypothesis requires a comparative approach. Some point, for example, to chimpanzees and their relative inability to cooperate to the degree of sophistication of human children. For instance, human children coordinate better in riskier strategic scenarios than chimpanzees due to their ability to communicate more effectively (Duguid et al. Reference Duguid, Wyman, Bullinger, Herfurth-Majstorovic and Tomasello2014). Yet, as section 2 demonstrates, evolution has solved group hunting problems time and time again, sometimes using communication to help. This fits a general pattern that we see across the tree of life, where evolution has solved all sorts of cooperation problems, often in elaborate and important ways, including building complex multicellular individuals out of simpler unicellular organisms (Buss Reference Buss1988; Maynard Smith and Szathmáry Reference Maynard Smith and Szathmáry1995; Griesemer and Shavit Reference Griesemer and Shavit2023).

Our perspective differs from some leading accounts of the evolution of joint agency, for which markers of human uniqueness feature. Tomasello (Reference Tomasello2016, Reference Tomasello2019), for example, maintains that rich cognitive capacities are required for cooperative behavior and so accounting for the evolution of such behavior requires an account of the evolution of the relevant rich cognitive capacities. The capacities he has in mind are those emphasized by leading philosophical accounts of bona fide joint agency—capacities for shared or collective intentionality (Schweikard and Schmid Reference Schweikard and Bernhard Schmid2021).

Such capacities can be understood in different ways, of course. Bratman (Reference Bratman1993) accounts for Jill and Sue’s coparenting in terms of their shared intention to coparent. Their shared intention is two (or more) individual states of the form “I intend that we J” and the relations between these states. In contrast, Gilbert (Reference Gilbert1992, Reference Gilbert2009) proposes that ascribing intentions to groups presupposes a “unification of agency.” Joint commitments of the form “I will coparent, if you will coparent” are group commitments on this account rather than individual commitments. When joint commitments hold, intentions and beliefs are properly attributed to the group rather than to individuals in the group.

Tomasello invokes both Bratman and Gilbert in introducing his own account of joint intention. He also understands the evolution of joint intention as a key step in our uniquely human evolutionary trajectory. Developing partnerships to engage in collaborative activities such as antelope hunting led to the formation of joint commitments, and this in turn “created an evolutionarily novel form of moral psychology” (Tomasello Reference Tomasello2016, 5). Thus “human morality is a form of cooperation, specifically, the form that has emerged as humans have adapted to new and species unique forms of social interaction and organization” (2).

For Tomasello, joint intention and other cognitive capacities required for the evolution of human hypersociality are uniquely human traits.Footnote ⁶ Tomasello’s evolutionary conjecture is that the capacity to engage in uniquely human joint action, and form commitments to undertake specific roles to bring about a joint end, emerged against the backdrop of early hominin obligate cooperation and facilitated a transition to more extreme human sociality and specialized cooperation. Some strategic cooperative interactions had already been undertaken in the human lineage, but the peculiarly human forms had yet to emerge (Tomasello et al. Reference Tomasello, Melis, Tennie, Wyman and Herrmann2012; Tomasello Reference Tomasello2016, Reference Tomasello2019, Reference Tomasello2020a). On this view, uniquely human joint agency is supposed to be both a distinct achievement of the human lineage and a capacity (or set of capacities) that facilitates increasingly complex cooperative behavior culminating with morality: “The ability to form shared agencies derives from a uniquely human social psychology of shared intentionality” (Tomasello Reference Tomasello2020b, 3).

This approach divides cooperative behavior into two categories: strategic cooperation of a sort that we see across the biological domain and human cooperation facilitated by uniquely human joint agency. As Tomasello (Reference Tomasello2016, 3–4) puts the thought, “the central claim of the current account is that the skills and motivation to construct with others an interdependent, plural-agent ‘we’ … are what propelled the human species from strategic cooperation to genuine morality.” More recently, Tomasello (Reference Tomasello2019, 7) spells out his evolutionary hypothesis about joint agency in even starker terms: “The claim is that these abilities emerged first in human evolution between collaborative partners operating dyadically in acts of joint intentionality, and then later among individuals as members of a cultural group in acts of collective intentionality.” If joint agency is uniquely human, then it is likely a late-arriving synthesis of cognitive capacities, thus playing relatively little role in the evolution of those very capacities implicated in joint agential action. This contrasts sharply with the view we defend here, which points to the possibility of numerous evolutionary precursors to the cognitive capacities central to Tomasello’s account.

Several alternate approaches to Tomasello’s also prominently feature the evolution of human uniqueness. For example, Hrdy (Reference Hrdy2009) and Wrangham (Reference Wrangham2019) provide alternate accounts. We briefly look at each in turn. Kristen Hawkes (Reference Hawkes2020) identifies a key component to Hrdy’s account of the evolution of human cognitive traits: her adoption of a life history approach. Hrdy’s account of the evolution of the cognitive differences between human infants and apes places the evolution of the relevant traits in ancestral infants: “preferences for participation, engagement and sharing attention and intentions with others that emerge in infancy, evolved as survival responses to the cognitive ecology of ancestral infancy” (as cited in Hawkes Reference Hawkes2020, 2). Just as the tadpole life stage of frogs faced different selection pressures than adult frogs, ancestral human infants faced different selection pressures to ancestral human adults. Hrdy and Tomasello share an explanatory target: the difference between human infant cognitive activity (and human social cognition in general) and that of the great apes. Hrdy agrees with Tomasello that humans have unique social cognition (as do we), but Hrdy provides an alternate answer to the question of how our social cognition evolved. According to Hrdy, one part of the human social cognition package evolved in ancestral infants who “arrive[d] into a socioecology where mothers, depending on allomothers, bear another offspring before the previous one is independent” (as cited in Hawkes Reference Hawkes2020, 6). A key difference between ancient human socioecology and ape socioecology is the presence of grandmothers (see Hawkes et al. Reference Hawkes, James O’Connell, Alvarez and Charnov1998; Hawkes Reference Hawkes2004). Ancient infants were forced to attract the attention of caregivers other than their birth mothers.

Another alternative approach, due to Wrangham (Reference Wrangham2019), treats the cooperative communication behaviors that are hallmarks of joint agency as initially emerging as a by-product rather than as a directly selected effect. Wrangham argues that the abilities to tolerate direct gaze and spend time in close proximity to conspecifics—necessary features of close cooperation and joint action—evolved as part of a process of self-domestication, or selection against reactive aggression. Wrangham offers a comparative argument that selection for domestication (and against reactive aggression) produces a similar domestication syndrome in mammals, and the abilities to follow gaze and tolerate individuals are paedomorphic (juvenile) traits accompanying this syndrome (172–89). He concludes that “in domesticated animals, increased cooperative ability emerges as an incidental consequence of selection against reactive aggression” (189), though he is quick to make clear that this origin is compatible with subsequent ratchet-like selection for increasing cooperation in the human lineage.Footnote ⁷

This brief sample illustrates that not only are alternate evolutionary approaches available to account for a wide range of cooperative behavior but a wide range of alternate evolutionary approaches are available to account for “uniquely human” cooperative behavior. In fact, the continuity of cooperative behaviors across the tree of life undermines the idea that there is a distinct category of uniquely human cooperation—human cooperative behavior simply differs in degree and sophistication from nonhuman cases, underscoring our earlier suggestion that identifying a normative account of bona fide joint agency or cooperation is not the most useful move for evolutionary explanation. Furthermore, human cooperative behavior may well be supported by a cognitive package that some characterize as joint agency, but that package has most likely been assembled from disparate cognitive capacities. The evolution of each of the components of this package may call for different accounts, and these accounts may create trade-offs or conflict among the array of selection pressures.

The various evolutionary accounts from Hrdy, Tomasello, and Wrangham differ in many ways that reflect attention to different facets of human social life and the conditions that made such sociality possible (see also Calcagno and Fuentes Reference Calcagno and Fuentes2012). Unsurprisingly, they reach different conclusions about what features are responsible for human evolution. The formal framework of evolutionary game theory provides a way to unify these approaches as various ways of solving general cooperation problems. This unification can provide some insight, but it does not help gain traction on the cognitive capacities implicated in the various solutions, for the generality of the framework extends across the tree of life. This compromises any explanation for human uniqueness that relies on solving strategic problems associated with complex cooperation. Perhaps more important, there are drawbacks to attempting to give an evolutionary account of the origin of some new cognitive package, such as joint agency, that unlocks human sociality. Such accounts downplay the many components involved in such a package and their likely separate evolutionary histories. These evolutionary histories are almost certain to interact in ways that compromise, synergize, and complicate the emergence of joint agency.