Social groups are studied in a variety of fields in the behavioural and brain sciences, mirroring their central role in human and nonhuman societies. As three researchers studying groups and their individuals from different perspectives (e.g., Faber, Häusser, & Kerr, Reference Faber, Häusser and Kerr2017; Lockwood et al., Reference Lockwood, Wittmann, Apps, Klein-Flügge, Crockett, Humphreys and Rushworth2018; Rütgen et al., Reference Rütgen, Seidel, Silani, Riečanský, Hummer, Windischberger and Lamm2015), we very much welcome an attempt for a unifying framework on groups. A shared conceptualization of groups would indeed allow researchers to bridge gaps between disciplines and could be extended to core group topics beyond conflict, such as cooperation and coordination problems. However, to really be unifying, we argue a broader scope than the one presented in the target article would be needed. Such a broader scope should, in particular, consider (1) a neuroscientific angle that (2) incorporates motivations and emotions.

The target article dissects the cognitive mechanisms allowing individuals to navigate social groups. It argues that group computations arise from considering costs that one agent imposes on another. Representing costs in group settings presses basic relational primitives into action. Relational primitives are the computational scaffold that enables us to think of the relationships between individuals in a group. However, we know that the computational architecture proposed, should it exist, must arise from neural processes. In neuroscience, we routinely consider learning from rewards and losses using computational approaches (Sutton & Barto, Reference Sutton and Barto2018). Pietraszewski's framework suggests that this quantitative framework of learning and decision-making might, in fact, be an adequate route to understanding the relational primitives at the heart of social group membership computations.

Neural networks in prefrontal cortex and interconnected subcortical regions have been identified that determine how people weigh costs and benefits in non-social settings (Basten, Biele, Heekeren, & Fiebach, Reference Basten, Biele, Heekeren and Fiebach2010; Klein-Flügge, Kennerley, Friston, & Bestmann, Reference Klein-Flügge, Kennerley, Friston and Bestmann2016). One intriguing implication from Pietraszewski's framework is that the same machinery that has evolutionary arisen to guide cost–benefit decisions in non-social domains may be used to instantiate relational primitives in group contexts. Should this be the case, then this raises a fundamental question about the social nature of group representations. This question is: To what degree are the component processes underlying our ability to navigate social groups specifically social? Neuroscientifically, it is likely that many component processes underlying social cognition are shared between social and non-social domains (Wittmann, Lockwood, & Rushworth, Reference Wittmann, Lockwood and Rushworth2018). For instance, the amygdala and the lateral orbitofrontal cortex are important for learning from rewards and associating them with specific stimuli (Murray & Rudebeck, Reference Murray and Rudebeck2018). These computations might underlie some of their contributions to social cognition (Munuera, Rigotti, & Salzman, Reference Munuera, Rigotti and Salzman2018; Sliwa & Freiwald, Reference Sliwa and Freiwald2017). However, other brain regions have been more specifically linked to our ability to think about other people and infer their beliefs such as the temporoparietal junction and dorsomedial prefrontal cortex (Lamm, Bukowski, & Silani, Reference Lamm, Bukowski and Silani2016; Saxe, Reference Saxe2006) and it is possible that at least some of the computations performed in these regions are specifically needed for navigating social environments.

Applying this perspective to group computations, we might speculate that group cognition relies similarly on a mixture of social and non-social mechanisms. Following the rationale of the target article, it might draw particularly on the ability to compute rewards and costs that ensue from the actions of ourselves and others. In addition, an ability that seems particularly pertinent for group cognition may be the ability to infer relationships between multiple agents. Recent studies have explored the specific computations via which the brain computes relationships between objects and even abstract concepts (Behrens et al., Reference Behrens, Muller, Whittington, Mark, Baram, Stachenfeld and Kurth-Nelson2018). This might be central for instantiating the relational primitives underpinning our ability to think about social groups. Nevertheless, despite the potential existence of such a domain general mechanism for forming relationships, it is possible that, in particular, dorsomedial prefrontal cortex, one of the most prominent regions in social cognition research, is specifically important for forming relationships between social agents (Izuma & Adolphs, Reference Izuma and Adolphs2013). Dorsomedial prefrontal cortex represents self and others in an interdependent frame of reference and appears to be causally important to separate out self and other related information (Wittmann et al., Reference Wittmann, Kolling, Faber, Scholl, Nelissen and Rushworth2016, Reference Wittmann, Trudel, Trier, Klein-Flügge, Sel, Verhagen and Rushworth2021).

Therefore, by employing a neurocomputational perspective, we may gain more precise information on which aspects of group representations may be genuinely social. However, by no means do we suggest being “reductionist” in a computational theory of groups. In fact, there are specific component mechanisms involved in social processes that may not fit in the categories proposed in the target article. Specifically, motivations and emotions are crucial for different aspects of group functioning – generally, and when it comes to conflict within and between groups. For example, a key social motivation for group functioning is the desire to build a positive reputation in the eyes of other people (cf. Faber, Savulescu, & Van Lange, Reference Faber, Savulescu and Van Lange2016). This motivation critically shapes prosocial behaviour (Ariely, Bracha, & Meier, Reference Ariely, Bracha and Meier2009; Nowak & Sigmund, Reference Nowak and Sigmund2005) and also group-decision making (De Dreu, Nijstad, & van Knippenberg, Reference De Dreu, Nijstad and van Knippenberg2008; Faulmüller, Mojzisch, Kerschreiter, & Schulz-Hardt, Reference Faulmüller, Mojzisch, Kerschreiter and Schulz-Hardt2012). Regarding emotions, empathy is an exemplary social emotion that is crucial. Empathy allows us to understand each other from a first-person experiential perspective (e.g., Lamm, Rütgen, & Wagner, Reference Lamm, Rütgen and Wagner2019). Although this can have beneficial effects on prosocial behaviours, such as intergroup cooperation, there is also a potential “dark side” to empathy. This may come out in competitive contexts (when we use our understanding of others to better compete against them), as well as when considering that empathy and the ensuing behaviour is prone to ingroup biases (Bloom, Reference Bloom2017). Although we have only started to understand the neurocomputational processes that underpin motivations and emotions at an individual level, even less is known on how a group context may alter or amplify these processes – or vice versa, how these processes determine membership in groups.

In summary, neuroscience provides a complementary approach that may enrich the proposed computational theory of social groups. It may help determine the precise – social and non-social – component mechanisms underlying group computations and provide a scaffold to incorporate additional component processes relating to motivations and emotions as well as their interaction into a computational theory of social groups.