The Anterior Cingulate Gyrus and Social Cognition: Tracking the Motivation of Others

1Accounts of the anterior cingulate cortex (ACC) have highlighted its role in fundamental cognitive processes, including motivation, decision making, learning, cost-benefit calculation, as well as conflict and error monitoring. While such theories have dominated research on ACC, there is also a long, if less influential, tradition associating ACC with social behavior. For example, many years ago, links were drawn between ACC and the basic affiliative and communicative behaviors in which all mammals engage. That link has persisted even as the neuroscientific approaches and behavioral paradigms used to investigate social interaction have increased dramatically in their sophistication. It is apparent when neuroimaging is used to examine human brain activity when people engage in socially oriented modes of cognition, such as empathy, that might once have been thought recalcitrant to neuroscientific examination. In addition, disorders of social cognition have long been linked to the structure and function of the ACC. Understanding the mechanistic contribution of this region to social behavior is therefore vital for understanding social cognition in health and disease.

2It is sometimes assumed that any role ACC has in processing social information simply reflects an aspect of more generalized cognitive control processes. Here we review recent experiments, primarily in primates such as macaques and humans, to examine whether this is indeed the case. Drawing on anatomical, lesion, single-unit recording, and neuroimaging studies, we suggest that the notion that social information processing in the ACC simply reflects general cognitive control processes can be refuted. Anatomical evidence points to considerable heterogeneity in ACC cytoarchitecture and connectivity. Such heterogeneity is indicative of functions being localized to discrete sub-regions. We highlight how evidence points to social information being processed in a specific ACC sub-region in the gyrus (ACCg). Both single-unit recording and neuroimaging studies suggest that this region computes “other-oriented” information (i.e., information about other agents that might be animals or people, rather than ourselves;, ). We also review this region’s connectional and functional properties in health and in disorders of social cognition before finally describing a model of how this region might compute information that guides our understanding of how motivated other people are. We suggest that this computational framework has potential for providing mechanistic understanding of variability in social behavior in healthy people and also disorders of social cognition.

3Classical as well as recent accounts of the ACC have often posited that this region plays a vital role in processing rewards and also in decision making. Recent work has highlighted that this region may also be particularly important for evaluating cost-benefit information that influences how motivated we are and foraging decisions such as whether to maintain or switch behavior. The ACC encodes variables determining the reward rate of a behavior, including its probability of reward and its effortfulness, which are key determinants of how vigorously behaviors should be pursued. On first glance it may therefore seem surprising to suggest that this region is also particularly important for processing information about others. However, from an ecological perspective, it isn’t surprising at all. Despite major differences in avian and mammalian neuroanatomy, in birds, species in which foraging is sophisticated, information about the presence of conspecifics is processed in adjacent regions to those concerned with both the costs and benefits of the bird’s own behavior. Moreover, in humans, social information is processed in the same or adjacent brain structures as those that are engaged during foraging decisions, which may in part be because information about competitors also crucially determines foraging decisions. Such a confluence of information appears to be a general principle of neural circuits for decision making.

4Electrophysiological recordings in rodents suggest that neurons in a homologous region to human ACC are concerned with cost-benefit decisions made in social contexts and with learning through observation of others. recorded from the ACC as rats evaluated the costs and benefits of competing with a conspecific for a reward. They found that a large proportion of ACC neurons coded the net value (benefit minus cost) of competing for a reward but did not respond similarly to equal levels of reward when no competition was required. Lesions of the same ACC region (Cg1/Cg2) in rats disrupt both cost-benefit decision making and diminish the normal interest they take in other rats. In mice, the typical “freezing” response when receiving painful stimulation can also be learned by observation. showed, however, that when the ACC of mice is inactivated, they are no longer able to learn the freezing response through observation. Inactivation of the lateral amygdala did not abolish the ability to learn from observation in the same way. This would suggest that the ACC may have a particularly important role in learning about the behavior of conspecifics in rodents as well cost-benefit information about an animal’s own behavior.

5Over the last decade many neuroimaging studies in humans have shown that ACC is active when people interact with others during economic games in which, typically, the experimental participants make decisions that affect not only their own subsequent payoff but also the payoff that other players will receive. A recent study suggested that such responses in the ACC may be driven by neurons that are predictive of the behaviors of another. recorded from the ACC as monkeys performed a similar type of game to those used in human neuroimaging studies—the prisoner’s dilemma game. In this game, players have to choose between cooperating with another or defecting—failing to cooperate. Cooperating resulted in a medium-sized reward for the monkey but only if the other cooperated too; the monkey only received a small reward if it cooperated and the other defected. Defecting led to a large reward payoff although, again, the payoff depended on what the other did too. Not surprisingly ACC neurons responded to the monkey’s own decisions of whether to cooperate or defect but intriguingly they found a second class of neurons predicting the decisions of the other monkey. Such findings suggest some specificity for monitoring and predicting the behavior of others within ACC. However, such experiments give rise to important questions: (1) is such activity a corollary of the social interaction that such tasks entail, or (2) is it simply the case that any difficult decision making, whether it involves other people or not, is associated with ACC activity? After all there are many reports linking ACC to decision making even in the absence of social context, so at the heart of this Review is the question: is there anything special about responses in the ACC when tracking others’ behavior?

6If primate ACC activity tracks the behavior of others, then we might expect ACC lesions to disrupt social behavior. It is indeed the case that lesions in humans that impact on large areas of medial prefrontal cortex, including portions of the ACC, result in changes in social behavior, social decision making, and the ability to learn from observation. However, the identity of the key areas that, if damaged, cause social impairments has been difficult to determine because the lesions are large and directly affect several anatomical areas and undercut the connections of even more. In macaques, circumscribed lesions to the ACC, more so than some other frontal areas, disrupt the processing of social information and lead to changes in social behaviors. There is therefore evidence of a causal link between social behavior and the ACC.

7Taken together, research points to the ACC playing a key role not just in reward-guided learning and decision making but also in social cognition. However, to truly understand the “reference frame” in which the ACC responds during social interactions and to examine the specificity of any region for processing social information, it is important to use tasks where such features can be disentangled. In the following sections, we consider some simpler tasks that allow us to examine how ACC activity tracks specific features of others’ behavior and the motivations that inform those behaviors. First, however, we consider the anatomy of the ACC in greater detail and examine whether some of its component zones may have distinct functional properties.

8The cingulate cortex in both humans and monkeys contains multiple distinct cytoarchitectonic zones that imply localization of different functions to each zone (Figure 1A). Although many discussions of ACC focus on the anatomical distinctions that arise along the rostro-caudal axis, an equally crucial cytoarchitectonic distinction also exists in the dorsal-ventral dimension between gyral and sulcal regions. Furthermore, differences in connectional anatomy, related to the direction and strength of inter-regional communications, suggest that the ACCg may be functionally distinct from the more frequently studied regions in the adjacent ACC sulcus (ACCs), often referred to as the dorsal ACC (dACC). Notably, the distinction in anatomical properties between the sulcus and gyrus extends across a large extent of the ACC. In many atlases the ACCg contains areas 24a/b and 32 in both humans and macaques, whereas areas 24c and 32′ lie predominantly in the ACCs (Figure 1A).

9Throughout this Review, we use the terms ACCg and ACCs to refer broadly to locations anterior to the posterior cingulate cortex (PCC) in the gyrus and sulcus, respectively. We do this because the location of the border between areas 32 and 24a/b has varied across anatomical studies and it was not possible to carry out detailed histological analysis of brain tissue in the humans and monkeys in the experiments we discuss. Moreover, although not acknowledged by every authority, there is evidence for cytoarchitectural subdivision within area 24 (24a, a’, b, and b’; ). However, the anatomical connections of these subdivisions, which constrain and determine function, appear broadly similar throughout 24a/24b. In fact many of the connections of both area 32 and 24 are similar (Figure 1B;,, ). Patterns of activity coupling between these areas and the rest of the brain, which only partly reflect monosynaptic connections, suggest important similarities between human and macaque 24a/b, macaque area 32 and human area 32pl, and the manner in which they interact with the rest of the brain in both species (; Figure 1B).

10Importantly, it has also been argued, on the basis of cytoarchitectural analyses, that there is homology between human and macaque ACC and that the ACCs and ACCg differ from one another in similar ways in the two species. This evidence suggests that it may be possible to integrate findings from lesion and neurophysiological studies in monkeys with neuroimaging evidence in humans.

11Humans live in large, complex social groups. It has been argued that living in such large groups has led to evolutionary pressure for larger brains in the primate order. As a result, it is often suggested that there are networks in the brain that have evolved to play important roles in social cognition and behavior. Historically, neurobiological models of social cognition have placed importance on three distributed neural networks, each of which may perform different functions that contribute to social behavior: (1) a “mentalizing” network involved in inferring others’ mental states and abstract information about social situations, comprising a region in the temporo-parietal junction (TPJ) and the dorsal portions of the medial prefrontal cortex (dmPFC); (2) an action observation network (AON) including the ventral premotor cortex (PMv) and adjacent BA44 and the anterior, inferior parietal cortex; and (3) a network involved in affective and value-based processing, including the amygdala, ventral portions of the medial prefrontal cortex (vmPFC), and anterior insula (AI).

12Although there are exceptions, the ACCg has classically not been considered as a part of these networks. Nevertheless, anatomical tracer studies in monkeys and neuroimaging studies in humans examining human brain connections (Figures 1B and 1C) suggest that this region may in fact have strong connections to each of these three social networks. This region connects to the TPJ and the dmPFC, which are engaged in mentalizing processes; to several amygdala nuclei, the vmPFC and the AI regions, which are engaged during affective and social processing; and also to the PMv and the inferior parietal cortex regions engaged during action observation. This connectivity profile places the ACCg as a potentially important region that integrates a variety of distinct forms of social information from the different social networks.

13Is the connectivity profile of ACCg distinct from other cingulate sub-regions? Notably, both tracer and resting-state studies in macaques and humans highlight that its connectional fingerprint is distinct from that of the adjacent cingulate regions. Specifically, none of the connections from the ACCs overlap with those of the ACCg to the mentalizing network or the AON and the connections of the ACCg and the ACCs to the vmPFC and the amygdala also do not fully overlap. This would suggest that overall the ACCg has a connectivity profile that plays an important role in processing social information, whereas the ACCs may not have access to all of the same information.

14Obviously not all ACCg’s connections are distinct from those of the ACCs. The ACCg also interconnects with many regions to which the ACCs does also. This includes strong interconnections between the ACCg and ACCs, between both regions and several other portions of the prefrontal cortex, as well as parts of the striatum, the parietal lobe, and both the ACCg and ACCs receive monosynaptic input from dopamine neurons in the ventral tegmental area. A diverse array of functions has been ascribed to these regions. However, it is clear that they play important roles in processing the costs and benefits of acting. This would therefore suggest that while the ACCg may be functionally distinct from the ACCs in terms of its role in processing social information, it may also be sensitive to some similar information. This may therefore highlight an important role for the ACCg in processing the costs and benefits of acting in social contexts, much how the ACCs does when evaluating the value of acting for ourselves.

15Lesion studies in macaques suggest that the region of the ACC that is most important for the processing of social information lies in the gyrus. showed that lesions to the ACCg decreased the value that macaques placed on social stimuli and also led to reductions in the execution of social behaviors. In contrast, lesions to the ACCs, and other parts of the frontal cortex, do not lead to similar changes to the valuation of social stimuli or social behavior, suggesting a special role for ACCg in social cognition. Neuroimaging studies in primates provide further support for the notion that the ACCg processes social information. Resting-state connectivity in the ACCg has been shown to correlate with the size of social group in which macaques were living.

16Single-unit recording studies in macaques also suggest that the ACCg processes other-regarding information that is particularly important for social behavior, in a manner that is distinct from other regions of the cingulate cortex. used a modified dictator game in monkeys, in which the actor monkeys made a decision to share or withhold reward from a conspecific or a decision to reward the conspecific or no one (Figure 2). Using this design they were able to disentangle whether neurons responded to rewarding outcomes being delivered to self, other, no one, or combinations of conditions. As well as ACCs, they recorded from neurons in a relatively anterior portion of ACCg, approximately at the border of areas 24a/b and 32 (Figures 2A and 2B). They found that a significantly higher proportion of neurons in the ACCg responded to rewarding outcomes delivered to another, with many responding exclusively when another monkey was receiving a reward. Specifically, they found that approximately 60% of neurons that responded to any choice, cue, or reward outcome (approximately 25% of total neurons recorded) in the ACCg (Figure 2C) signaled when another monkey was receiving a reward (whether exclusively for another monkey [other-referenced] or in a mirrored fashion when either self or other obtained a reward [both-referenced]). Nearly half of these reward-sensitive neurons in ACCg only responded to the other monkey’s reward. While 25% may seem a small proportion of neurons, it is comparable with the proportion of neurons in the premotor cortex exhibiting “mirror neuron”-like profiles and yet such PMv neurons putatively influence behavior and impact on electromyographic activity at several synapses distance. In contrast, approximately 70% of reward-modulated neurons in the ACCs and 80% in the OFC were self-referenced and did not respond to any information about other monkeys. Thus, the proportion of neurons in the ACCg sensitive to other’s reward outcome, including those that exclusively use an other-oriented frame of reference, is striking and distinctive. This finding would therefore suggest that, compared to the ACCs and other cortical regions, the ACCg may play a more significant role in processing “other-oriented” information.

17Is there a similar specialization for other-regarding information in the ACCg in the human brain? Recently, a number of studies have asked whether ACCg processes self or other-oriented information. With a striking consistency, they have localized an ACCg region near area 24a/b that only responds to other-oriented information (Figure 3A). Crucially, all of these studies have highlighted that ACCg processes cues that are informative as to what events will impact another person even if it does not always process cues informative of events that will impact upon ourselves.

18The ACCg is activated when processing cues that are predictive of others’ reward (,; Figures 2A and 2B) or provide information about whether rewarding outcomes are being delivered during social interactions (,,,; Figures 3C and 3D) or when subjects process stimuli that are predictive of painful stimulation being delivered to another person (,; Figure 2A). However, the ACCg is not activated by similar cues indicating that the subject themselves will receive painful stimulation or a reward. In addition, a study by examined activity in the ACCg when processing the outcomes of either another’s decisions or the responses of a computer. Crucially, they found that the ACCg—albeit in a slightly more posterior region from the other neuroimaging and neurophysiological recording studies—responded to the outcomes of the other person’s decisions but not to the outcomes of a computer’s responses. Taken together, neurophysiological and neuroimaging evidence reveals that the ACCg responds to other-oriented information but not to information about ourselves or about non-biological agents.

19As highlighted in the previous section, there is accumulating evidence that the ACCg processes other-oriented information. Although teasing apart signals that are related to decisions, predictions, or outcomes can be challenging, in the this section we review evidence that suggests that the ACCg may play an important role in processing cues that allow for predictions about cost-benefit evaluations to be made about the value of a behavior for another.

20In the study by, not only did the ACCg activity reflect reward delivered to others, but it also, at earlier time points in each trial, reflected the monkeys’ expectations about the reward about to be received by the other. Specifically, the normalized response profile of ACCg neurons reflected reward that the other monkey would receive prior to their delivery, but it did not change when rewards were to be delivered to no one or just to the self (Figure 2B). Therefore, ACCg activity reflects predictions about the motivated state of a conspecific prior to the actual reward delivery.

21Neuroimaging studies have highlighted that human ACCg activity is found exclusively for cues that are predictive of reward for others. examined activity at the time of cues that was predictive of a high probability or a low probability of a reward for the subject or for another person. Crucially, they found that activity in the ACCg was sensitive only to the probability of another receiving a reward.

22Some neural circuits are not just sensitive to primary reward but they are also sensitive to other factors that determine the net value of a course of action, such as costs entailed by the action made to obtain reward. This certainly is the case for cingulate regions concerned with reward-guided decision making even in the absence of social context, but is it also true of the ACCg? used fMRI to examine activity at the time of cues that indicated whether the subject or another person would have to exert effort to obtain a reward (Figure 3A). They found ACCg activity covaried with net value (reward – effort) at the time of the cues, but only on trials when the other person would have to exert the effort to obtain the reward. Such an activity pattern is analogous to that seen in the ACCs in non-social reward-guided decision making. Other studies have also reported activity in the ACCg at times when predictions can be made about the value of a behavior for another. Importantly, this includes studies in which subjects are evaluating reward that will be received by others that are delayed in terms of their receipt–another key factor that devalues rewards.

23How does ACCg activity manifest when there is a possibility of a net negative state for another individual? If the ACCg is truly signaling value, then its activity should decline even further in such situation, but if instead the ACCg is signaling the net motivational impact of another person’s state, then its activity will increase as the net negative impact increases (cf. ). ACCg activity conforms with the latter pattern; it encodes motivational salience; cues indicative of affective states closely linked to motivation, such as fearful faces or pictures of others in pain, also increase activity in the ACCg in the same manner as rewarding stimuli. But, as highlighted above, activity in this region shows an opposing profile in terms of being sensitive to the costs of others’ actions.

24This profile suggests that at the time of cues that are predictive of behavior by another, activity increases in the ACCg when the benefit associated with the behavior is greater and actions will be more motivated (invigorated). The prospect of negative states as well as positive ones can lead to similar motivation or invigoration and it is with this motivational factor that ACCg activity covaries. That is, activity increases when another will avoid pain or obtain a reward. However, activity in this region also decreases with costs that decrease motivation, such as the effort costs or temporal delay before a beneficial outcome is received. This suggests that the ACCg is signaling the motivational value of acting for another.

25While many studies suggest that the ACCg responds to stimuli that are predictive of events that impact another, the ACCg also appears to be sensitive to the outcomes of decisions that impact upon another. A large number of EEG and fMRI studies have shown during social interaction tasks that the ACCg signals when the outcomes of others’ actions are unexpected. As highlighted above, identified that a group of reward-sensitive neurons in the macaque ACCg signal only when reward will be delivered to another and not to no one or ourselves. In humans, asked participants to make decisions on the basis both of the history of reward that they had experienced with different choices, and also on the basis of advice that they were given. When the participants were given feedback at the end of the each trial, they were able to update both their estimates of action values and their estimates of how good the advisor was. The weight of influence, or impact, of the feedback information on these two updates was associated with activity change in the ACCs and ACCg, respectively; the activity increases were, respectively, proportional to the action learning rate parameter size and the advisor learning rate parameter size in the model used by Behrens and colleagues. This would point, therefore, to the ACCg playing an important role in other-oriented learning—learning about the behavior of others, and perhaps also, to learning about the value other’s place on behavior.

26Other studies suggest that not only does the ACCg signal information at the time that outcomes are delivered to other people, but it signals when other people’s expectations about the consequences of their actions are revealed as erroneous. Such signals are typically referred to as prediction errors (PEs), a signal for tracking the statistical properties of the environment. In the ACCg, the statistical properties of the contingencies between stimuli and rewarding outcomes for others are tracked by PEs. As reported above, found that the ACCg was active when the outcomes of another’s actions are revealed to a subject. Specifically, they showed that ACCg activity signaled when the outcome of another’s action was discrepant from the predicted outcome. However, it signaled this PE only for the unexpected outcomes of another’s action and not for the outcomes of a computer. ACCg activity has also been found to covary with PEs when monitoring the outcomes of another’s strategy as well as when updating beliefs about other people’s strategies. Crucially, all of these studies point to the ACCg being engaged when monitoring others’ outcomes and suggest that this region may signal “outcome”-related PEs (OPEs).

27Recent evidence suggests that the ACCg may also signal prediction errors (PEs) while we monitor the actions of another person, not only when viewing the outcomes of their actions. asked subjects to act like a teacher and monitor the actions of another person (student) engaged in learning stimulus-response associations by trial and error (Figure 3B). As the subjects had learned all of these associations themselves, when they saw the action of the other person they could infer a PE for how different another’s prediction was from the actual outcome the subject knew they would receive. They found activity in the ACCg signaled this PE. In other studies, the ACCg was found to respond more strongly when observing others’ actions that have unfamiliar or unusual kinematics. It could be postulated that such a response in the ACCg might reflect the signaling of a discrepancy between the expected and actual kinematics of another’s action. That is, the ACCg may signal a PE when monitoring another’s movement, coding for the difference between an expectation about the value of a movement for another and the actual value of the movement executed.

28In contrast to the ACCg, the adjacent ACCs (or dACC as it is often called) does not seem to have such a specialization for other-oriented information. Based on single-unit recordings, lesion studies, and neuroimaging approaches (see or for reviews), the ACCs, including its anterior perigenual portion, signals the value of our own behavior, and updates those estimates when feedback suggests that they may have been incorrect. As a result of these findings, many attempts have been made to develop domain-general models of the contributions of this region to cognitive processing.

29In studies directly investigating social information processing, there is mixed evidence as to how important a role the ACCs plays in processing social information. In human neuroimaging studies, the ACCs is activated when painful stimulation is delivered to either ourselves or other people. However, it also responds to the unexpected outcomes of both other people and computers. This suggests that ACCs processes information regardless of whether it pertains to another biological agent or not. In, a study discussed in the preceding sections, the vast majority of ACCs neurons differentiated between reward delivered to self and other but the same neurons also similarly differentiated between conditions in which the self or no one obtained a reward. In fact, within the ACCs very few neurons distinguished between reward delivered to the other or to no one (Figure 2B). This would suggest that neurons in this region predominantly respond to “foregone” reward. That is, these neurons code for when a reward is not going to be delivered to ourselves. This would place these neurons within a “self” reference frame and not an “other” frame of reference.

30In contrast, recent studies recording from the ACCs during more complex social interaction tasks could be interpreted as suggesting that neurons in this region signal other-oriented information. found neurons in the ACCs that responded when another monkey made an error during a reward incentivized action selection task. While this could be interpreted as reflecting “other-oriented” information, it is important to note that the subject’s own outcome and subsequent behavior was critically dependent on the action selection of the other monkey. A failure of the other monkey to act appropriately would therefore lead to an absence of reward for the monkey themselves. Thus, this finding is remarkably consistent with that of, who showed that the majority of responses in the ACCs reflect a “foregone” reward for ourselves in a self-referenced manner.

31In another study, recorded from neurons in ACCs as monkeys played the prisoner’s dilemma game with another monkey. They found neurons that were predictive of the other monkeys’ choices of whether to cooperate in the iterative game. While these signals may be interpreted as reflecting the other-oriented process, it is also equally plausible that these signals may be self-referenced. With differential probabilities of the monkey themselves receiving a reward depending on the history of choices of the two monkeys, these neurons could be reflecting a prediction of a foregone reward as in. These neurons may therefore reflect the updating of the first monkey’s decision-making strategy on the basis of the other’s previous actions, which is critical for social interactions.

32While this notion of domain generality might suggest that the ACCs has limited specialization for social information processing, it is important to consider the connections highlighted in previous sections between the ACCg and other areas of the brain. As noted, the ACCg has strong connections to the OFC, ventral striatum, and ACCs. Without doubt, these regions are important for guiding behavior, including social interactions. However, neuroimaging and neurophysiological evidence suggests that they may not process information in the same reference frame as ACCg. Moreover, the ACCg is also connected to portions of the dmPFC and TPJ that have been shown to signal “other-oriented” information in a variety of different tasks (see for a review). This therefore places the ACCg within a distributed network that guides social behavior, with the ACCg processing information about the level of motivation of others agents during social interactions.

33Social abilities vary considerably even between healthy individuals. Notably, the structural and functional properties of the ACCg have been linked to variability in both social behavior and factors that influence socio-cognitive abilities.

34As highlighted above, the connectivity of the ACCg has been shown to reflect social group size (Figure 4A). There is also evidence that activity in the ACCg in response to other-oriented information—and particularly to stimuli that carry information about the motivational value of behavior for another—varies with trait levels of social abilities. Levels of our ability to empathize with others have been shown to be correlated with activity in the ACCg when processing information about pain being delivered to others or cues that are instructive of the probability of another receiving a reward (Figure 4B).

35There is also evidence that variability in the ACCg response is linked to social behavior during tasks in which people interact. Specifically, the extent to which activity in the ACCg responded to feedback about the advice of another is related to variability in the degree to which people are influenced by that advice (; Figure 3C). In addition, the extent to which ACCg signals information at the time of the outcomes of others’ actions, and particularly the degree to which this region signals PEs, correlates with how influenced by others’ behavior people are during competitive interactions (; Figure 3D).

36These findings therefore suggest that while the response of the ACCg in the typical population is predominantly “other-oriented,” the degree to which the ACCg processes information in this reference frame varies between individuals. As a result, there is considerable variability in the response of the ACCg to predictions about the value of a behavior for another and also to the extent to which PEs update our predictions about other people. These studies point to individual differences in the response of the ACCg or the degree to which the response is in an other-oriented frame of reference being closely linked to variability in the ability to understand the value of a behavior for another individual.

37In clinically diagnosed psychopathy and ASD, there is also evidence of disruptions to the structural and functional properties of the ACCg. Post-mortem evidence has highlighted changes in the histological properties of the ACCg and ACCs as one of the most replicable differences between the brains of individuals with ASD and healthy control individuals. Resting connectivity of an area including the ACCg and also ACCs has also been shown to be different in a large sample of individuals with ASD compared to a matched control group. It has also been shown that symptom severity in ASD is correlated with activity in the ACC during an economic exchange with another (; Figure 4C). In addition, psychopathic individuals and children who are rated as highly callous show differences in gray matter volume in the ACCg and connectivity between this region and other regions involved in processing social information such as the AI and the AON.

38Existing evidence suggests that the functional properties disrupted in the ACCg in these disorders may be linked to the processing of information about the pain another is receiving. Activity in the ACCg to the pain of others has been shown to be attenuated in psychopathic individuals and those who exhibit callous traits (,,; Figure 4D). Similarly, the response of the ACCg to stimuli that indicate another is in pain is reduced in ASD compared to healthy controls. Prominent theories of ASD have also argued that deficits in social behavior may arise from the inability to effectively process motivation related information during social interactions.

39There is therefore evidence to suggest that the ACCg may be a key region that is disrupted across different disorders of social cognition. Moreover, the deficit in the ACCg appears to be specifically tied to the processing of “other-oriented,” motivationally relevant information. Thus, while there have been limited specific investigations of the ACCg in disorders of social cognition, these results suggest that this region may be a fruitful target to examine in disorders of social cognition.

40In the previous sections, we have highlighted how the ACCg may be the key region of the ACC that processes information in an “other-oriented” reference frame. Disruption to this region changes social behavior, a group of specialized neurons in this region respond to others’ reward, the region signals when information learned specifically about others is unexpected, and variability in this region is related to variability in social behavior. However, to date, there has been an absence of a theoretical framework that can inform future basic or clinical research. Moreover, while studies have hinted at the computational mechanisms that may operate within this region, this has never been formalized in a generalized manner that could characterize variability in behavioral or neural signals across a variety of tasks. This is in stark contrast to generalized models of dACC/ACCs, of which theoretical accounts are numerous and classical accounts suggest that the ACCs may be important for conflict monitoring, error detection, learning, and decision making. Here, we suggest that the ACCg shares some of the computational properties of the ACCs but processes this information in an other-oriented frame of reference. In this next section, based on the extant literature and relevant research on the computations that underlie value-based motivation, we put forward a model that may be useful in explaining the previously identified contribution of the ACCg to processing information about others.

41In order to successfully interact with others, an agent must monitor the behavior of other individuals closely in order to predict the behaviors that they are likely to produce next and respond in the most adaptive manner. Crucially, to do so, an agent needs also to estimate the motivation of others. Accurately processing the level of motivation another has to obtain a desirable outcome (i.e., vicariously processing the value of a behavior for another individual) is a fundamental ability that is distinct from many other more frequently discussed components of social cognition (e.g., theory of mind, perspective taking, or empathizing). For example, when I see someone running fast, this tells me that they are highly motivated at that moment in time. The vigor (speed) of their behavior therefore informs how motivated they are. However, it does not necessarily provide information on what their mental state is, nor is it necessarily informative as to their current mood. That is, we are able to infer whether someone is highly motivated or not, but their level of motivation can be orthogonal to the inferences we can make about their cognitive or affective state. Tracking another’s motivation may therefore be a key, but often neglected, component of social cognition.

42The level of motivation or vigor in another’s behavior can therefore be thought of as a key piece of information that must be tracked during social interactions. Importantly, it can also be quantified. Anything that increases vigor can be thought of as motivating behavior toward a beneficial outcome that is being sought (e.g., obtaining a reward or avoiding pain), and anything that decreases vigor when there is a highly beneficial outcome offered can be thought of as a cost. This can be quantified as a “value” signal, with the benefits of moving fast traded off against the costs. As argued in previous sections, we suggest that ACCg is sensitive to factors determining the motivation of others (their costs and benefits). There is little evidence that many other brain regions signal as much information about the reward or pain that is, or may be, received by another apart from perhaps the AI, a region to which the ACCg is strongly connected. This includes frontal lobe regions linked to reward-guided decision making, such as vmPFC and ACCs, and other brain regions such as TPJ and dmPFC linked to other aspects of social cognition and meta-representation.

43We break vicarious motivation (Figure 5) processing down into the three distinct phases. (1) We must infer and represent the expectations another has about the value of their own behavior in terms of the beneficial outcome they desire for themselves, and the costs they will incur to obtain the outcome. This vicarious value (Vv) can either be derived from exogenous cues from the environment that have been learned to be predictive of costs and benefits for others, from communication directly, or from cues such as facial expressions. (2) The Vv ascribed to others’ behaviors are updated dynamically, online during the monitoring of their actions; and (3) the value of cues associated with such behaviors are updated when a subsequent outcome reveals that the initial expectation at the time of a cue was erroneous (i.e., when we learn that the value of a particular cue for another is different from what we had expected). In this section, we outline how a model derived from computational accounts of value processing and motor control may be useful for quantifying the mechanisms underlying the vicarious processing of motivation. From henceforth, we refer to this computational framework as the Motivation and Vicarious Error model (MoVE).

44As noted in previous sections, the ACCg appears to process information that increases the value of acting (e.g., avoiding pain or approaching reward) and also that which decreases value (e.g., effort or delays). Computationally, we therefore characterize a prediction of how motivated another is as processing the vicarious value (Vv) of an action they will perform. This Vv is dependent upon the magnitude of the benefit (R) for another individual. This benefit can be thought of as being always positively valenced or unsigned (i.e., regardless of the positive or negative valence of the outcome, the Vv increases with the magnitude of the outcome such as greater pain or reward). So avoiding a large loss, obtaining a large reward, or avoiding a significant amount of pain would all reflect a large benefit for another person and thus increase “value.” The benefit magnitude is devalued by three distinct costs: (1) the temporal delay (D) before another receives the benefit; a benefit received in a week is valued as worth less than the same benefit received immediately; (2) the probability (P) of the benefit being received (if a particular behavior has a low probability of resulting in the desired benefit, then motivation to execute that behavior will likely be low and so reaction times and movement velocities will be slower;,, ); and (3) the effort that must exerted in order to obtain the beneficial outcome (E). This “effort” is determined by the amount of control that must be exerted during the performance of the behavior. The greater the effort to be exerted, the more the behavior is devalued.

45This model therefore predicts that activity in the ACCg will be highest for a highly beneficial behavior for another, which is immediate, has a high probability of being received, and for which minimal effort needs to be exerted. However, each of these is weighed by a parameter that dictates how sensitive we believe another is to these sources of information (k,θ,Φ). That is, how impulsive someone is (k), how sensitive they are to probability (θ), and how averse they are to effort (Φ) all dictate how motivated they will be. The response of the ACCg therefore depends on how these sources of information are weighed, as well as the real delays, probabilities, and effort that another must exert. Importantly, we do not suggest that these specific characteristics—that define other people’s traits—are necessarily “stored” in the ACCg, merely that when monitoring another’s behavior, they can influence the valuation another places on a behavior and how invigorated their actions will be. These parameters reflect how different individuals may therefore value benefits differently. Through its connections with other regions that are important for learning about others’ traits and states, the ACCg may therefore form part of a network that allows the Vv to be constantly updated based on how much another weighs benefits against costs.

46As outlined in the previous paragraphs, a prediction of how motivated another is can therefore be characterized in the following equation:

47This equation therefore represents the key properties of the MoVE, evaluating the costs and benefits of a behavior for another. As noted in previous sections, activity in the ACCg has previously been shown to be modulated by all of these features and to signal the net value of others’ behaviors.

48As highlighted in previous sections, the ACCg signals PEs both when the outcomes of others’ actions are processed, but also when monitoring other’s actions. Similarly, two distinct updating signals are crucial for dynamically controlling one’s own behavior and learning from exogenous feedback (Figure 5B). At the time of an outcome (feedback) of a behavior, a comparison can be made between the actual value of an action and the expected value of the action. When these are discrepant, PE signals drive learning based on the magnitude of the error. These PEs are scaled by learning rate parameters that dictate how much is learned from a PE. The outcomes of others’ actions reveal whether expectations about the value of a behavior for that individual were accurate. At the time of the outcome of another’s behavior, we can compare our expectation about the value of an outcome of a particular behavior for another to the value of the actual outcome they receive. When there is a discrepancy, vicarious outcome PEs (vOPEs) update our expectations of the level of motivation of another in the future following similar cues in the environment. That is, if we observe another obtaining a reward after they have performed an action, having never previously been rewarded for that behavior, we will come to expect they will be more motivated to perform the action in the future.

49Control theories also postulate another form of PE signal that occurs during movements. During our own actions, we monitor their consequences and correct errors in our movements by identifying discrepancies between expected and actual movement kinematics. During actions we can also dynamically update our evaluations of others’ motivations on the basis of the properties of the others’ movements that are witnessed, such as the speed (vigor) of their movement. When monitoring another’s behavior, we can therefore compare our expectation about the value of a behavior—and an expectation about the vigor of their movement—with the actual behavior we observe. These can therefore be thought of as “dynamic” PEs as they code for errors during movements. If the movement is faster than predicted, then the model increases the estimate of the value the other agent is placing on behavior (i.e., their motivation is higher than was expected), whereas the opposite is true when the movement is slower than predicted. Such discrepancies between the actual and expected speed of another’s movements are updated by a vicarious dynamic PE (vDPE) that improves estimates of how motivated another currently is while executing a behavior.

50Crucially, vDPEs and vOPEs can be used to update estimates of the effort required, as well as the probability of receiving a reward. In the MoVE model, these two forms of PE update estimates of the value of another’s behavior.

51The responses in the ACCg to the value of behaviors for others and to PEs that were discussed in earlier sections approximate the signals predicted by the MoVE model. This model therefore provides a new framework for understanding previously identified signals in the ACCg during social interaction tasks. Notably this model also has the potential to quantify individual differences in people’s processing of the motivation of other individuals. This model may therefore be able to quantify and provide a mechanistic understanding of failures to accurately represent the motivation of others (Figure 5C).

52Notably, the MoVE model shares commonalities with other models that have been proposed that can account for strategic learning and reasoning during social interactions including models of simulation, strategic inference or trust. There are two key differences between these models and the MoVE. First, these models are predominantly aimed at computing abstract information about the strategy of others, and how we ourselves should adapt our strategy to reach goals during social interactions. In the MoVE model, the aim is to account for much lower-level information about how motivated other people are. Rather than attempting to explain strategic or theory of mind type inferences, this model makes predictions simply about the value of a behavior for another individual that allows predictions about the invigoration of another’s movement to be optimized. Second, in no previous model has the effort cost to be exerted by another been included within a model. This is pertinent given the significant influence that effort costs have on motivation. The MoVE model therefore makes predictions about a component of social behavior that is not directly considered or parameterized in any of these models and as highlighted in other sections have been shown to be tracked in the ACCg.

53We note, however, that did report activity in the ACCg at the time of the outcomes of decisions that scaled with the volatility (or the current parametric value of a dynamically changing learning rate) of another’s advice. While the MoVE model does not directly predict the response in the ACCg witnessed in that study, it may have approximated some of what was observed, as in many circumstances the two models make similar predictions. Future studies should therefore use tasks where the properties of the two models lead them to make distinct predictions that can be tested against the response of the ACCg to determine whether volatility needs to be incorporated into the outcome response present in the ACCg.

54By proposing a new conceptual framework of the contribution of the ACCg to social cognition, there are a number of potential avenues of research. From an anatomical perspective, perhaps the most interesting question is what are the precise locations of ACCg zones processing other-oriented information. To date, recordings from the gyrus of the cingulate cortex in macaques have not extended posterior to the genu of the corpus callosum. In contrast, neuroimaging studies show activity across the ACCg to social information in both more posterior and anterior locations. Future research should therefore examine whether the other-oriented zone of the ACCg lies just in a small circumscribed region or whether a large portion is concerned with social information. Moreover, it has recently been shown that in the ACCs outcomes delivered to different parts of the body evoke activity in different zones. It would be intriguing to examine whether a similarly topographic representation of others’ outcomes is also evident in ACCg. For example, is feedback for another’s arm movement, as opposed to feedback for facial movements, represented in a different portion of the cingulate gyrus.

55To fully support the framework we have proposed, it would also be important to demonstrate that single neurons in the ACCg respond in a manner predicted by the MoVE model. For example, it will be important to examine whether neurons in the ACCg signal PEs relating to the unexpectedness of the speed of others’ movements and also to the unexpected outcomes of their actions.

56An interesting question, which was beyond the scope of this Review, is whether different components of computing the value for others’ behaviors can be tied to distinct neuromodulators. Past studies suggest dissociable roles for oxytocin, dopamine, and serotonin in influencing social behavior. However, future research should examine the specific influences of each of these on the processing of costs and benefits in the ACCg when they are vicariously processed.

57Finally, we believe this model and framework may be more generally useful for understanding variability in social abilities such as empathy, cooperation, competition, as well as disorders of social cognition. Do all disorders of social cognition have deficits in processing others’ motivation and, if they do, can they be parameterized by different components of the MoVE model? Future work using tasks tailored to understanding how motivation is tracked during social interactions may be able to tease apart distinct deficits in different disorders in which there is ACCg dysfunction.

58In summary, research across different species in terms of anatomical connectivity, neuroimaging, and neurophysiology support the notion that a region in ACCg—adjacent and dorsal to the genu of the corpus callosum in humans—plays a crucial role in evaluating the behaviors of others and in estimating others’ level of motivation. We review the anatomical and functional role of the ACCg in social processing and put forward a mechanistic, computational account of the contributions of the ACCg to social cognition. Our account predicts many of the responses observed in the ACCg including those recorded from individual neurons and in neuroimaging studies. In addition, the MoVE model is easily generalizable to other domains of complex social behavior in which it is also necessary to understand others’ motivation. Looking forward, this framework may help to characterize and provide a mechanistic understanding of disorders of social cognition, by parameterizing potentially dysfunctional components of social behavior.

59M.A.J.A., M.F.S.R., and S.W.C.C. wrote the paper.