Core knowledge of social relationships: a proposal

When humans, including young children, reason in domains such as physics, navigation, and psychology, we use intuitive theories, which are interconnected systems of knowledge that allow us to make sense of varied and unfamiliar situations and plan appropriate actions (S. Carey, Reference Carey1987; Gelman, Reference Gelman2005; Gerstenberg & Tenenbaum, Reference Gerstenberg and Tenenbaum2017; Gopnik & Wellman, Reference Gopnik, Wellman and Hirschfeld1994). These intuitive theories are proposed to be built on very early developing or innate knowledge called ‘core knowledge’. Core knowledge serves as inductive biases: it supports infants’ ability to organize the vast amounts of information in their environment and direct infants’ attention to relevant information (S. Carey, Reference Carey2009; Kemp & Tenenbaum, Reference Kemp and Tenenbaum2008; Spelke, Reference Spelke2022, p. 20; Spelke & Kinzler, Reference Spelke and Kinzler2007; Tenenbaum et al., Reference Tenenbaum, Kemp, Griffiths and Goodman2011; Ullman & Tenenbaum, Reference Ullman and Tenenbaum2020; Xu, Reference Xu2019). Core knowledge therefore shapes intuitive theories. What are intuitive theories of social relationships and what core knowledge shapes them?

First, consider the core components of intuitive physics and psychology. The core components across these domains are analogous. When we reason about physics, we reason about entities (things in the world), properties of the entities (e.g., heavy, smooth), and forces (how entities change properties); when we reason about psychology, we reason about agents (a specific type of entity), states (psychological or physical), and actions (how agents change states) (refer to Spelke, Reference Spelke2022; Ullman, Reference Ullman2015 page 200; and Liu, Reference Liu2022; Liu et al., Reference Liu, Outa and Akbiyik2024. I propose that our intuitive theory of relationships has analogous core components: connections (who is socially connected?), characteristics (what type of connection is it and how strong is it?), and communication (the way that connections are established, maintained, and changed). That is, much the same way we reason about properties of entities and states of agents, we reason about the characteristics of relationships. Much the same way that we represent forces to be the vehicle by which entities change or maintain their properties (e.g., a force causes a change in the location of a ball) and we represent actions as the vehicle by which agents change or maintain states (e.g., a person moves their hand to reach for an object), we represent communication as the vehicle by which a social connection changes or maintains its characteristics (e.g., accepting a hug from someone to communicate social closeness). This communication can be explicit (e.g., saying, ‘I declare my allegiance and loyalty to you’) or implicit (e.g., expressing comfort with using the same spoon as a family member). Like the core components in physics and psychology, each of these components have foundations in infancy. In this paper, I focus on the second component: characterizing relationships. I propose that two basic computations determine the way humans represent characteristics of social relationships: categorizing the connection by model and computing the strength of the social connection.

While the core components of each domain are analogous, and the domains interact, reasoning about social relationships goes beyond the domains of physics and psychology (for a discussion about the interdependence of intuitive psychology and physics, refer to Liu, Reference Liu, Outa and Akbiyik2024). Consider first that intuitive physics helps us understand the actions of people (or agents more generally). We can use our knowledge of agents as physical bodies to know it is unlikely that a person will walk through a wall. However, we need more than an understanding of the physical world to make sense of many actions of agents: when we see a person turn left to reach a door, we cannot only use intuitive physics to understand the action—we also need a representation of the person’s goal. Likewise, some behavior cannot be understood by only representing a single agent’s mental state. Take the notion from intuitive psychology, in which people expect agents to act efficiently toward achieving goals (S. Carey, Reference Carey2009; Gergely & Csibra, Reference Gergely and Csibra2003; Jara-Ettinger et al., Reference Jara-Ettinger, Gweon, Schulz and Tenenbaum2016; Liu et al., Reference Liu, Ullman, Tenenbaum and Spelke2017; Spelke, Reference Spelke2022). If we observe someone bowing and moving out of the way for someone else and then moving forward on their path, without an understanding of relationships, the action seems irrational. We could explain this action as evidence that the bower is prosocial, but it would not tell the whole story. If we know that relationships are at play, we understand that bowing communicates a certain role in a relationship (A. P. Fiske & Schubert, Reference Fiske, Schubert, Gillath, Adams and Kunkel2012; Kajanus et al., Reference Kajanus, Afshordi and Warneken2019; Thomas, Radkani, et al., Reference Thomas, Radkani and Hung2022 refer to Box 2).

What do core representations of relationships allow us to do, and how are they distinct from representations of traits or groups?

Like representations of traits and social groups, representations of social relationships constrain and affect the way we explain and evaluate behavior. The difference between representations of traits, relationships, and groups is that they each lead to different generalizations. To illustrate, imagine you are on the subway, and you see someone return a dropped wallet. Without other social cues, you may attribute the behaviors to traits—for example, the wallet-dropper is forgetful, and the wallet-returner is kind. This could lead you to change predictions about the behaviors of the actors toward most people. For example, the wallet-returner may open a door for someone, and the wallet-dropper may be late for a meeting. Now imagine that the two people are wearing Red Sox hats, and the wallet-returner says, ‘Go Red Sox’, as they return the wallet. Here you may explain the interaction in terms of social groups, leading you to predict, for example, that the wallet-returner will extend prosocial behaviors to other people wearing Red Sox hats. The wallet-returner might even be less likely to help a person wearing a Yankees hat. In other words, the representation of groups would afford predictions concerning people who belong to the group of Red Sox Fans. Finally, imagine you see the same interaction, but first, the two people kiss. Here you would likely infer that the pair is a romantic couple. This could lead to predictions that are constrained to the couple. For example, the wallet-returner might hold open a door for their lover but may not be more likely than others to hold open a door for a stranger. You would be surprised to see them hug or kiss seemingly unrelated people on the subway.Footnote ¹ One way of visualizing the distinction between these representations is to imagine a social network, where nodes are individuals, and connections between nodes are interpersonal relationships. Trait inferences change your expectations about how nodes with act toward most other nodes, group inferences update your expectations about different aggregates of nodes and how nodes from these aggregates interact, while relationship inferences change your expectations about the lines between nodes and how those two nodes will interact (refer to Box 3). Thus, infants’ representations of social relationships would lead them to make different predictions than representations of traits and groups. Next, I will describe in detail how humans, including infants, may recognize relationships.

How do we recognize relationships?

So far, I have described the components of a relationship representation: connections, characteristics, and communication. Next, I proposed their function—they allow humans to make sense of interactions between individuals, updating predictions about whether and how people will interact. What follows is that relationship representations, require us to track individuals. The ability to track individuals has deep phylogenetic roots. For example, chimps recognize group mates even after decades of separation and they better remember those with whom they had closer relationships (Lewis et al., Reference Lewis, Wessling, Kano, Stevens, Call and Krupenye2023). This ability has also been found in elephants, horses (Proops et al., Reference Proops, McComb and Reby2009), sheep (Kendrick et al., Reference Kendrick, Da Costa, Leigh, Hinton and Peirce2001), and even wasps (Sheehan & Tibbetts, Reference Sheehan and Tibbetts2008). Human newborns recognize their mother’s face (Field et al., Reference Field, Cohen, Garcia and Greenberg1984), scent (Vaglio, Reference Vaglio2009), and voice (DeCasper & Fifer, Reference DeCasper and Fifer1980). Humans as young as four months (Kosakowski et al., Reference Kosakowski, Cohen, Takahashi, Keil, Kanwisher and Saxe2022, Reference Kosakowski, Cohen, Herrera, Nichoson, Kanwisher and Saxe2024), have specialized face regions in the brain whose connectivity to other regions suggest that this specialization is driven by social value, perhaps associated with forming relationships with individuals (refer to Powell et al., Reference Powell, Kosakowski and Saxe2018).

But recognizing relationships goes beyond tracking individuals. How do we know that individuals are connected? It may depend on recognizing cues that people are mutually aware.Footnote ² Newborn infants recognize many cues of social engagement: they distinguish direct gaze from averted gaze (e.g., Farroni et al., Reference Farroni, Csibra, Simion and Johnson2002), prefer infant directed speech (e.g., Cooper & Aslin, Reference Cooper and Aslin1990), are aware of contingency, (e.g., Bigelow, Reference Bigelow and Rochat1999; Bigelow et al., Reference Bigelow, Power, Mcquaid, Ward and Rochat2008; refer to Csibra, Reference Csibra2010) and Spelke, Reference Spelke2022 for an in depth review). As observers, infants as young as 12 months differentiate when others look at each other compared to when they don’t, and expect people to look at one another when communicating (Beier & Spelke, Reference Beier and Spelke2012). Thus, infants may use cues of mutual awareness, such as mutual gaze, contingency, or joint attention, to infer whether a connection exists at all (refer to Wolf & Tomasello, Reference Wolf and Tomasello2023).

How do we characterize relationships?

Computing a relationship’s strength

While models can tell us what to expect even in fleeting social interactions, computing a relationship’s strength would tell an infant what to expect over many interactions, across time and contexts. This idea overlaps with proposals in many theories of social relationships, including proposals about pull (A. P. Fiske & Fiske, Reference Fiske, Fiske and Kitayama2007), attachment (M. D. S. Ainsworth et al., Reference Ainsworth, Bell, Stayton and Richards1974), interdependence (Rusbult & Van Lange, Reference Rusbult and Van Lange2008), thickness (Margalit, Reference Margalit2017, p. 2), and strength (Granovetter, Reference Granovetter1973). I propose that each of these proposed concepts relate to one underlying representation which I call “relationship strength”. Relationship strength is a continuous computation: it informs the extent to which we are committed to someone (i.e., the strength of our intention to maintain the connection) and/or the extent we feel obligated toward someone (i.e., the extent to which certain actions are normatively expected). Relationship strength does not necessarily describe the ‘strength’ of the model (e.g., how highly ranked we see someone or how much overlap we have with someone, though this may be correlated). The computation occurs in relationships for each model. While people can be highly committed to many things, the extent to which they are committed to single individuals over timeFootnote ⁴ and across contextsFootnote ⁵ determines the relationship’s strength. For example, a highly devoted teacher may feel committed to teaching because she is generally motivated to impart knowledge on the next generation. However, if this commitment is not directed toward specific individuals the relationships would be relatively weak. By contrast, a teacher could have strong relationships with her students, if she is committed to individuals across contexts and intends to maintain relationships with them even after the course is over. Likewise, you might feel highly obligated to help victims of a hurricane, but if your obligation is not directed toward individuals, or if your obligation is not caused by a relationship, then it wouldn’t be evidence of a strong relationship.

Relationships that fall under each of the three relational models can be relatively weak or strong. However, the likelihood of very strong or very weak relationships may vary depending on the governing relational model people use in a single relationship. Below are examples of relationships in each relational model that are relatively strong or weak. For example, you might feel much closer and more committed to your spouse than a teammate but use the logic of communal sharing in both relationships. The logic of communal sharing can even happen in fleeting relations, such as when people feel as one with strangers at a concert, sports, or religious event. Likewise, authority ranking relations can also be strong or weak: one mentee could be devoted to their mentor, while another may see their mentor as a temporary boss to whom they feel no allegiance and do not expect their relationship to continue after the employment has ended. Authority ranking can also occur in fleeting interactions, such as when people feel they owe deference to a judge in court. Interestingly, Graeber suggests that equality matching relations encourage weak relationships since when ‘debts’ are fully paid, there is no necessity to interact in the future (Graeber, Reference Graeber2012). Thus, communal sharing relationships and authority ranking relationships may have a larger range of ‘strength’ as opposed to equality matching. Even so, equality matching relations can sometimes be strong and long-lasting, such as between neighbors who have long been doing favors for one another, or weaker, such as children taking turns at a playground. Each model may have different emotions tied with strength, which could serve as one cue to infer a strong relationship (e.g., attachment or ‘kama muta’ (A. P. Fiske et al., Reference Fiske, Schubert, Seibt, Cassaniti and Menon2016) in communal sharing relationships; allegiance in authority ranking relationships; and trust in equality matching relationships).

One untested hypothesis about relationship strength is that the stronger a relationship (i.e. the stronger you feel attachment, allegiance and loyalty, the more you expect the relationship to continue, and the stronger you imagine the bond), the less replaceable the relationship and person. If this is true, then it might be one reason why we mourn more when stronger relationships end (Harris, Reference Harris2018). Take, for example, one’s relationship with their children. In almost all situations, a person’s child is completely non-interchangeable. One would certainly notice if their spouse brought home the wrong child from school. The immense number of resources parents put into raising children is directed toward specific individuals. At least in most times and places, and possibly even for non-human primates (Goldsborough et al., Reference Goldsborough, van Leeuwen, Kolff, de Waal and Webb2020; Monsó & Osuna-Mascaró, Reference Monsó and Osuna-Mascaró2021), the death of a child is often considered one of the worst things that can happen to someone, and parents do not feel that the death of their child can be solved by having a new child. Or consider that when people describe ending a relationship with one’s child via other means, they often do so in terms of death (e.g., ‘he is dead to me now’). This may be because death feels like the only way one can end a relationship with one’s child. On the other extreme, you may not care much about which of your colleagues accompanies you to lunch or who helps you with an error in your code. If the colleague gets fired, and another person takes that colleague’s place then the person wouldn’t think it’s the same relationship, but it is conceivable that they would think the new relationship could replace the old one. The owner of a factory may see workers as individuals, but individuals who can easily be replaced by any worker who has the same skills. One can also think of friendships or acquaintance relationships that can easily be swapped or replacedFootnote ⁶ (refer R. M. Carey & Markus, Reference Carey and Markus2017 for evidence that the extent to which people think friendships can/should be replaced varies with social class in the United States).

Is this intuitive theory built on core knowledge of social relationships? If so, what would this core knowledge be?

I propose that our intuitive theory of social relationships is built on core knowledge. Elizabeth Spelke describes core knowledge as something between belief and perception; Susan Carey theorizes it to be iconic in formFootnote ⁷ (S. Carey, Reference Carey2009). Furthermore, core knowledge is innate, skeletal, largely automatic, and evolutionarily old. Core knowledge is abstract and cannot be computed from perceptual input alone. It persists throughout the lifespan and serves as the foundation of later developing representations, which are heavily influenced by cultural input, statistical learning, and explicit teaching.

Core knowledge of social relationships allows infants to track individuals and their social connections. Like other core knowledge, these representations act as inductive biases that enable learning. I further propose that infants recognize relationship models. These models are analogous to the ones found in Relational Models Theory: communal sharing, where individuals are like ’one’, authority ranking, asymmetrical relationships where individuals are ranked, and, equality matching, reciprocal relationships where individuals are separate. They may be substantiated as edges of graphs, with nodes representing individuals and edges representing relationships that are undirected for communal sharing, directed for authority ranking, and bidirectional for equality matching (refer to Figure 1). These model representations could be innate. Another possibility is that they could be learned from the combination of more basic social relationship representations (i.e., connections between people: two people are aware of one another) and abilities to recognize general relations (e.g., nominal, ordinal, interval, and ratio).

Figure 1.

Intuitive Theory of Relationships (A) The components of our intuitive theory of relationships are connections (which individuals are connected), characteristics (what is the relationship model, and what is the strength of the connection?), and communication (how are connections formed, maintained, or changed?). (B) People represent the characteristics of relationships along two dimensions: relational model and strength. (C) People (black dot) must locate themselves in relationships and learn about potential new social connections (dotted line). (D) This is a process of asking how people relate, what behaviors follow from different types of relationships, and what behaviors or actions might change these relationships or groups.

One way research could help us disambiguate these possibilities is to consider the literature on object recognition. When infants categorize objects they better individuate and remember them. A six-month-old infant is not surprised when a toy duck is replaced by a toy truck, but is surprised when a doll’s face is replaced with a ball(Kibbe & Leslie, Reference Kibbe and Leslie2019). Finally, adults categorize objects at the same time they recognize them (Grill-Spector & Kanwisher, Reference Grill-Spector and Kanwisher2005). Thus, understanding how young infants distinguish relationships may help us understand whether they categorize them from an early infancy.

Whether the models are innate or early developing, infant representations of relationships are simpler than later-developing representations: they have the structure without the details. A key difference between infant and adult representations is likely that infants do not fully represent the minds of individuals within a relationship (Favre & Sornette, Reference Favre and Sornette2015). This is also supported by the fact that species with simpler representations of others’ minds recognize relationships analogous to the three relational models (Haslam, Reference Haslam2013; Verbeek, Reference Verbeek, Killen and Smetana2006).

Using Authority Ranking as an example, young infants likely recognize asymmetrical relationships but may not understand the various ways rank is cued or construed. Relational Models Theory describes Authority Ranking relationships as those where one person is motivated to protect and guide, while the other is motivated to follow. These relationships are distinguished from dominance relationships, which occur when weaker individuals defer to avoid harm. By contrast, the analogous representation in infants is unlikely to be dominance or legitimate authority. As outlined below, infants do recognize relevant cues that occur cross-species, such as physical size that are associated with dominance, and do use these cues to predict behavior. However, it is unlikely they fully represent the motivations of the individuals until the second year of life. For example, toddlers distinguish between different types of asymmetrical relationships, which may include different motivations of high- and low-ranking individuals, a distinction that would require representing the minds of the individuals involved.

The skeletal nature of the infant representations allows learning of the vast cross-cultural variation in how relationships are established and marked. For example, gender, tattoos, hairstyles, clothing, and race have all been valid indicators of relative social rank in different cultures and times. Yet, they are not valid cues in all cultures, and different instances have meant different things. Or take an even more specific example. In some cultures, taking off a head covering in front of someone means that you have a family or intimate relationship with them. In other cultures, head coverings are only worn for fashion or utility and give little to no information about relationships. Infants in some cultures would need to come to know that head coverings can tell you something about social closeness, while those in other cultures would need to ignore it. Of course, this presents a learning challenge—if infants are born with abstract knowledge about relationships, they would need some way to connect perceptual input to this knowledge.

Some cues might be recognized with either very little learning or no learning (refer also to Fiske & Schubert, Reference Fiske, Schubert, Gillath, Adams and Kunkel2012; Thomsen & Carey, Reference Thomsen, Carey, Banaji and Gelman2013). Cues that can be learned with very little or no input may be related to evolutionarily old mechanisms. Physical closeness, including touch and exchanging bodily fluids, correlates with communal sharing relationships and strength of relationships in many cultures and species, and it is also related to mothers caring for their young. Physical size tends to highly correlate with formidability, which is what determines dominance rank in many species (Huntingford, Reference Huntingford2013; Smith & Price, Reference Smith and Price1973). Infants may need very little or even no input to form expectations about the relation between differences in size and asymmetrical relationships (refer to Asymmetrical Relationships section below). Finally, reciprocal relationships are often, though not always, based on the exchange of goods; thus, equal resource distribution may be a salient cue of equality matching relationships. If there are innate inductive biases that relate to specific cues, then we would expect most humans to recognize them. Humans may take advantage of this to communicate about relationships (Astuti et al., Reference Astuti, Solomon and Carey2004). Many cultures have ceremonies that feature saliva-sharing and physical contact, for example, dances where people hold hands or ceremonies where people drink from the same ceremonial cup. Likewise, many cultures make high-ranking individuals, whose bodies are not necessarily large, appear larger using thrones, headdresses, crowns, monuments, etc. (A. P. Fiske & Haslam, Reference Fiske, Haslam and Baldwin2005; Thomsen & Carey, Reference Thomsen, Carey, Banaji and Gelman2013). On the other side of the spectrum, humans recognize and use many cues that are arbitrary and must be learned. Examples include some military rank symbols, specific labels (e.g., ‘madame’, ‘sister’), and wedding rings.

Below, I review evidence that supports the hypothesis that humans possess innate or very early representations of social relationships. I organize the paper by relational model type and consider any evidence that infants infer the strength of connection within these types. While some of this proposal is supported by currently available evidence, other parts of the proposal have yet to be tested. Specifically, several studies support the proposal that infants represent relationships analogous to communal sharing and authority ranking relationships. In contrast, fewer studies support the idea that infants recognize equality matching relationships. Finally, existing studies only hint that infants represent the strength of relationships. This paper does not cover the vast amount of literature on children’s and adults’ ideas of social relationships, nor on how ideas about social relationships develop over the lifespan through cultural learning and first-person experience. Instead, it describes evidence that infants can recognize and categorize relationships. This knowledge lays the foundation for our adult intuitive theory of relationships.

Core Knowledge of Relationships: Evidence from Infants

Are infants sensitive to cues of communal sharing?

Observing Social Interactions

During the first year of life, humans recognize cues of people acting as one, including when individuals imitate one another, act in rhythmic synchrony, respond to distress, eat together, or share bodily fluids. For example, in one set of studies, 8-month-old infants saw animations in which two triads of animated characters moved their bodies in rhythmic synchrony (Powell & Spelke, Reference Powell and Spelke2013) (refer to Figure 2). When observing a new scene, infants expected that the characters would imitate the actions of the members of their triad as opposed to the other triad. Importantly, infants did not match ‘same’ with ‘same’: infants expected imitation among triads when the characters had different physical characteristics (i.e., were different colors and had different styles of hair). Likewise, when the characters looked the same but failed to coordinate their actions, infants did not expect imitation. Infants could not have used trait inferences to make predictions. The characters do the same actions in the familiarization events, what differs are their social partners, thus infants likely see relationships. Why might these studies be evidence for representations of communal sharing? In both the familiarization and the test events, the characters made themselves appear to be one unit. To adults, this coordinated action looks like a dance: each character does the same thing at the same time, moving in a circle. In the test events, infants expected that individuals would again make themselves alike by imitating each other’s actions. It is unlikely that the 8-month-old infants from the United States had much experience with people dancing, much less had the opportunity to compare subsequent behavior of individuals who dance together or not. Yet, infants recognized a cue that, in many cultures, identifies communal sharing relationships.

Figure 2.

Stills of stimuli (green background) and diagram depicting the hypothesized underlying representation in Powell and Spelke, Reference Powell and Spelke2013. (A) What infants observed in the study. (B) hypothesized cognitive representation (C) What infants observed during the test trial and the prediction that they made. Diagrams were added by me.

Expectations that arise from recognizing similarity may emerge even earlier. Four- and 5-month-old infants expect a character to approach another character it had previously imitated rather than one it had merely responded to (Powell & Spelke, Reference Powell and Spelke2016). However, these young infants did not expect imitation to predict mutual responses: they did distinguish scenes where a character approached someone who had previously imitated the character. Like young infants, twelve-month-old infants anticipated that a puppet who imitated an actress would respond to her distress. Unlike young infants, they also expected the targets of imitation to respond, suggesting expectations of mutual behaviors (Kudrnova et al., Reference Kudrnova, Spelke and Thomas2024; refer also to Pepe & Powell, Reference Pepe and Powell2023). Moreover, their expectations did not extend to individuals outside the initial interaction or to scenes where the actress laughed, suggesting they saw imitation as a cue of relationships. While there are several reasons why infants may have shown different patterns of results across the two studies, one explanation is that responses to distress are a better measure of relationship representations than approach behavior (for example, we approach objects we like but we can only comfort another person). It is also possible there is a developmental change such that younger infants see imitation as evidence that “A likes B” which would only affect predictions about A’s behavior toward B, while older infants see it as evidence that “A is connected to B,” which would also change predictions about B’s behavior.

The results reviewed thus far suggest that young infants use social cues in which individuals make themselves more like someone else to predict subsequent behavior. Converging evidence comes from another study, 9-month-old infants expected members of synchronized triads to help each other, rather than members of other triads, in a right-of-way conflict (Pun et al., Reference Pun, Birch and Baron2021). Infants and toddlers also use other cues in which people make themselves more alike to predict future affiliative behavior. For example, toddlers (aged 2 1/2 years and 17 months) expect that individuals who refer to themselves by the same label will support each other (Jin & Baillargeon, Reference Jin and Baillargeon2017; Ting et al., Reference Ting, He and Baillargeon2019; Ting & Baillargeon, Reference Ting and Baillargeon2021). Infants as young as 6 months expect people who announce the same preferences for food and express positive emotions when eating together to be friendly toward one another (Liberman et al., Reference Liberman, Woodward, Sullivan and Kinzler2016, Reference Liberman, Kinzler and Woodward2021).

Interestingly these behaviors can be construed, at least by adults, as intentional—they are things people do to make themselves more like others. In the study with the dancing groups, intentional similarity was compared to happenstance similarity (i.e., individuals who happen to look alike). This suggests that infants use intentional actions to recognize communal sharing relationships. Infants also reach for puppets who express the same preference as the infant, but only when the puppet expressed their preference after the infant, not when the puppets expressed their preference first. This suggests that the communicative act of choosing the same mittens as the infant influenced the infant’s evaluation(Mahajan & Wynn, Reference Mahajan and Wynn2012; Spelke, Reference Spelke2022).Footnote ⁸ Likewise, 12-month-old infants expect that people who wear the same clothing will support one another (Ting et al., Reference Ting, He and Baillargeon2019). However, they no longer expect support when similar clothing has an instrumental purpose (Bian & Baillargeon, Reference Bian and Baillargeon2022). The distinction between intentional similarity and happenstance similarity is central to the way that many scholars in fields such as Anthropology and Sociology conceive of social relationships: relationships are created through intentional communication and actions (Ingold, Reference Ingold2018). While I do not propose that young infants need to represent the minds of the individuals to infer a communal sharing relationship, they may be more sensitive to actions that have cues of intentionality, perhaps because those cues have been more useful over evolutionary time.

In contrast to the proposal that intentional similarity matters, there is evidence that 9-month-old infants expect friendly interactions between people who speak the same language, which for people who speak one language could be construed as unintentional (Liberman et al., Reference Liberman, Woodward and Kinzler2017). Likewise, infants’ first-person evaluations are affected by whether others share characteristics with their caregivers. For example, very young infants look more at faces who share the race of their caregivers(Bar-Haim et al., Reference Bar-Haim, Ziv, Lamy and Hodes2006; Quinn et al., Reference Quinn, Uttley, Lee, Gibson, Smith, Slater and Pascalis2008). Infants seem to prefer to interact with those who speak the same language as their caregivers (Kinzler, Reference Kinzler2021). Interestingly, while infants look longer at same-race individuals, they do not prefer to take toys from them, suggesting that their visual preference might be based more on lower-level perceptual preferences than social preferences (Bar-Haim et al., Reference Bar-Haim, Ziv, Lamy and Hodes2006; Kinzler & Spelke, Reference Kinzler and Spelke2011; Quinn et al., Reference Quinn, Uttley, Lee, Gibson, Smith, Slater and Pascalis2008). In contrast to race, but agreeing with the results about language, 5-month-old infants preferentially look toward someone who had previously sung a familiar song (Soley & Sebastián-Gallés, Reference Soley and Sebastián-Gallés2015). (In this U.S. American sample, this was ‘Twinkle Twinkle Little Star’; they used parental report to confirm the infants had heard the song before). Eleven-month-old infants also prefer people who sing a familiar song that had only recently become familiar because their parents are assigned to sing it to them, but not when their parents were assigned to play the song from a toy. This is true even after several weeks of not hearing their parents sing the song (Mehr & Spelke, Reference Mehr and Spelke2018). Because language and music are learned over social interactions, it may be a good cue of a person’s history of social interactions (Kinzler, Reference Kinzler2021). In line with the other studies covered earlier in the section, this body of work could reflect representations of communal sharing relationships. That is, singing the same songs or speaking the same language could be seen as evidence that individuals have made themselves more alike in the past. They could also represent representations of groups (refer to Box 3)

Do infants think about relationship strength in communal sharing relationships?

So far, I have argued that infants recognize communal sharing relationships. A series of studies suggest that infants may also recognize the strength of communal sharing relationships. Both infants (8 to 10 months old) and toddlers (16 to 18 months old) use interactions that include saliva-sharing, as opposed to other cooperative interactions, to predict responses to distress (Thomas, Woo, et al., Reference Thomas, Woo, Nettle, Spelke and Saxe2022). Across many cultures actions that require sharing bodily fluids through actions like kissing, drinking from the same cup, etc., are distinctively found in, and expected to be in, especially strong communal sharing relationships (e.g., Fiske, Reference Fiske and Haslam2004; Hung et al., Reference Hung, Thomas, Radkani, Tenenbaum and Saxe2022; Miller et al., Reference Miller, Rozin and Fiske1998; refer to supplementary materials of Thomas et al. Reference Thomas, Mitchell, Sumner, Terrizzi, Piff and Sarnecka2022e) In this study, an actress and a puppet ate from the same orange slice. Then, the same puppet and a different woman passed a ball back and forth. After the puppet expressed distress, infants and toddlers looked first and longer at the food sharer, suggesting they anticipated that the orange-sharer would respond. Importantly, these expectations were not driven by inferring the traits of the characters: participants did not anticipate the orange sharer to respond to an unfamiliar individual’s distress. Moreover, toddlers did not expect the orange-sharer to respond to requests for balls. This was true in a subsequent experiment that did not include food: infants saw a woman put her finger in her mouth, put her finger in a puppet’s mouth, and then back in her mouth. The same woman was then shown touching her and another puppet’s forehead. Here, infants and toddlers expected the puppet who had been the recipient of the ‘saliva-sharing’ action to respond to the woman’s distress, suggesting that their inferences did not depend on sharing food and showing that they had expectations that responses would be mutual (refer to Figure 3).

Figure 3.

Stimuli (from experiment) and diagram of hypothesized underlying representation driving responses in Thomas et al. (2022e). (A) What infants observed in the study. (B) hypothesized cognitive representation (C) What infants observed during the test trial and the prediction that they made.

Why might this be evidence that infants represent the strength of communal sharing relationships? As mentioned above, in many cultures, sharing bodily fluids is found in especially strong communal sharing relationships (refer to supplementary materials of Thomas et al., Reference Thomas, Radkani and Hung2022), while the other actions may be found in many communal sharing relationships, both weak and strong. On the other hand, it is possible that the participants categorized saliva-sharing as communal sharing but categorized the other actions as evidence of some other kind of relationship. Future studies will be needed to disambiguate whether participants see the same model with ball-passing and saliva sharing, or different strengths. These findings leave open the possibility that infants represent relationship strength.

Toddlers make inferences about triadic closure when observing cues of communal sharing relationships (i.e., if A and B are family, and A and C are family, then B and C are family; Spokes & Spelke, Reference Spokes and Spelke2017). When 15-month-olds saw the same large character respond to the distress of two small characters, they expected the small characters to be socially connected. They also expected two large characters who responded to the distress of the same small character to be socially connected. However, when the characters responded instead to laughter, toddlers no longer had these expectations. This suggests representations of strength: Responses to distress, unlike laughter, may be associated with stronger communal sharing relationships, and the logic of triadic closure may be especially true in networks of strong communal sharing relationships, such as in families. This idea is integral to Attachment Theory, where an infant’s propensity to go to their attachment figures when distressed, but not in other situations, is a hallmark of a secure attachment (Bowlby & Ainsworth, Reference Bowlby, Ainsworth, Goldberg, Muir and Kerr2013).

Do infants represent their own communal sharing relationships? In one study, infants used knowledge about their social relationships with their parents to infer whether they were socially connected to new individuals (Thomas, Saxe, et al., Reference Thomas, Saxe and Spelke2022) (refer to Figure 4). In this study, 12-month-old infants watched digitally edited videos of their parents interacting with two puppets. The parent showed affiliation toward one puppet by vocal imitation and had a friendly but non-imitative interaction with the other puppet by vocally responding. The question was whether the infant expected social engagement from the puppet their parent had imitated. During the test event, both puppets moved their mouths in synchrony while a friendly voice called the infant’s name. Infants looked longer at the imitated puppet, suggesting they expected it to be the source of the voice. This inference was not due to attributing a friendly trait to the imitated puppet, as infants did not look more at the puppet when it appeared to have a friendly interaction with someone else off-screen. Crucially, infants’ inferences were specific to interactions that involved their parents. When infants saw an unfamiliar adult (another infant’s parent) interact with another set of puppets they no longer looked more at the imitated puppet. These results suggest that infants combine their knowledge of pre-existing relationships with observations of social interactions to discover new social connections. The study’s details hint that infants may use cues of communal sharing relationships, such as imitation, to make these inferences. However, it remains an open question whether the strength of the infant-parent relationship was essential to their learning about new social connections.

Figure 4.

Stimuli from experiment and diagram of hypothesized underlying representations from Thomas, Saxe, Spelke, Reference Thomas, Saxe and Spelke2022. (A) What infants observed in the study. (B) Hypothesized cognitive representation (C; Top) What infants observed during the test trial and the prediction that they made. (C; Bottom) Data from one experiment in Thomas, Saxe, Spelke Reference Thomas, Saxe and Spelke2022, infants spent more time looking at the imitated puppet after they saw their parent’s interactions (blue) but not after they saw another infant’s parent’s interactions (gray).

Communal Sharing: Conclusions and Open Questions

In this section, I have argued that infants reason about observed interactions by using cues that occur in communal sharing relationships. They do this in situations they have likely not encountered before and about unfamiliar individuals, suggesting that they make predictions in these situations using abstract knowledge. The evidence here is suggestive that infants also represent the strength of communal sharing relationships. Many open questions remain.

First, many of these studies were originally interpreted as evidence that infants recognize groups. Why should we think about these studies as evidence of representations of relationships instead? In short, these studies do not provide any direct evidence that infants are generalizing beyond the interacting social partners (refer to Box 3 for a longer discussion). Second, in several of these experiments, infants may have used narrower representations of caregiver-child relationships, which are intimate and asymmetrical (refer to Thomas, Steele, Gopnik & Saxe, Reference Thomas, Steele, Gopnik and Saxe2024), instead of a broader communal sharing representation. If caregiving relationships are the innate primitive it would agree with the argument in attachment theory that the quality of an infant’s relationship with their caregivers acts as a model for all other relationships (M. D. Ainsworth, Reference Ainsworth1989). This is supported by findings that an infant’s attachment to their parents affects their expectations about responses to distress (Biro et al., Reference Biro, Alink, Huffmeijer, Bakermans-Kranenburg and van IJzendoorn2015; Johnson et al., Reference Johnson, Dweck, Chen, Stern, Ok and Barth2010). Fiske argues that models can be used as a lexicon, so it is possible that caregiving relationships are those that take on both communal sharing and authority ranking depending on the context. Third, what are the computations that infants use to infer the category and strength of communal sharing relationships? For example, do infants use naïve utility calculus, where one computes costs and rewards to infer the value someone places on things (Powell, Reference Powell2021)? In short, I am skeptical that this is the only computation that infants use to represent relationships as the logic does not easily apply to all of these findings. For example, saliva-sharing has a less than straightforward ’cost’ (refer to Box 2). Relatedly, how should we interpret the current argument, considering the claim that humans have an innate moral core? (refer to Box 1). Moreover, it is unclear whether infants think about obligation or commitment that goes beyond the time frame of the experiments, which may be necessary components of strong relationships. Far more work could be done to address these any many other questions.

Asymmetrical Relationships

In Relational Models Theory, asymmetrical relationships are authority ranking relationships, which occur in social hierarchies when people are ordered along a dimension (e.g., based on things such as physical size, age, skill, and military rank). In these relations, people have asymmetrical roles and often express the relation by accentuating asymmetries. For example, higher-ranking individuals are often made to appear larger (e.g., sitting on a throne, wearing a headdress, or sitting above others, such as a judge in a courtroom), brighter (e.g., wearing gold), or more important (e.g., being first or last in a procession, having a larger office).

Do non-human animals think about hierarchical relationships?

Many animals form dominance hierarchies, where rank is usually defined by priority access to resources, such as food, territory, and mates. Sometimes, dominance rank is established by one’s ability to inflict harm, which correlates with factors such as body size or coalition size. Other times it is inherited or arises from group dynamics (refer to Tibbetts et al., Reference Tibbetts, Pardo-Sanchez and Weise2022 review; refer to Mandalaywala, Reference Mandalaywala2022 about whether prestige-based hierarchies are present in non-human primates). Many dominance hierarchies form because it is in the interest of weaker individuals to yield in conflicts that they would otherwise lose. Therefore, in dominance hierarchies, patterns of yielding in conflicts could be reliable cues about social rank (Smith & Price, Reference Smith and Price1973; van Vugt & Tybur, Reference van Vugt and Tybur2014).

Individuals in several species reason about relative rank when they observe patterns of yielding. Experiments show fish, birds, mice, and rats use transitive inference about dominance rank to make decisions about how to interact with new individuals (Grosenick et al., Reference Grosenick, Clement and Fernald2007; Paz-Y-Miño C et al., 2004). That is, after seeing that A is lower ranked than B, and B is lower ranked than C, the animals in the experiment are more likely to approach A than C. Field studies reveal that many other species, including hyenas, lions, dolphins, horses, and several species of primates, represent the dominance ranks of conspecifics in their group, for example, using dominance rank to predict support in conflicts (Seyfarth & Cheney, Reference Seyfarth and Cheney2012). Chimpanzees also seem to connect spatial position with social rank (Dahl & Adachi, Reference Dahl and Adachi2013). Finally, bonobos seem to prefer (they reach for) dominant individuals, even after observing interactions between simple characters (shapes with googly eyes) (Krupenye & Hare, Reference Krupenye and Hare2018). While it is unclear how other species come to recognize cues of dominance, this work leaves open the possibility that learning mechanisms that lead to representations of dominance relations are evolutionarily old and prevalent even in human infants.

Evidence that infants are sensitive to asymmetrical relationships.

Human infants do seem to represent dominance relations, using physical size and coalition size to predict who will yield in conflicts (Thomsen, Reference Thomsen2019). In one study, 10-month-old infants from the United States expected smaller animated characters to yield to larger ones in a conflict. In this study, two characters (rectangles with eyes) had conflicting goals of crossing a platform in opposite directions, blocking one another’s pathway. Then, one of the characters lowered its body (i.e., bowed) and moved aside, allowing the other to pass (Thomsen et al., Reference Thomsen, Frankenhuis, Ingold-Smith and Carey2011). Infants looked longer when the larger individual yielded to the smaller individual, as though it violated their expectations. In another study, Canadian 6-month-old infants were shown similar animations as described above, but the characters were the same size, here they expected a character with more allies to win (Pun et al., Reference Pun, Birch and Baron2016). In another study, 12- to 16-month-old Japanese infants expected those who appeared lower in space to yield in conflicts over a resource (Meng et al., Reference Meng, Nakawake, Nitta, Hashiya and Moriguchi2019). Why might these studies suggest that infants have abstract, and possibly, innate representations? Infants at these ages have likely seen conflicts. However, it is unlikely that infants at this age have observed a member of a dyad and a member of a triad in a right-of-way conflict and were able to compare the two outcomes. Together, these studies suggest that before the first year of life, infants compare the relative power of individuals (i.e., infants track who defers and who is larger, stronger, or has more allies). However, these studies do not provide evidence that infants understand these interactions in terms of relationships. Their predictions are based on the relative traits (larger, higher, or more-allied) of the individuals. In other words, there isn’t evidence that infants use these interactions to update their belief about whether these individuals will interact again. Rather these studies provide evidence that infants have expectations about the outcome of an interaction given the individual traits.

Do infants use this information to inform whom they approach, as other species do? In one study 10- to 16-month-old infants were shown the right-of-way conflict described above, but this time the two characters were equal in size. Here, infants reached for the individual who yielded (Thomas & Sarnecka, Reference Thomas and Sarnecka2019). Infants’ preferences in these studies did not seem to be based on preferences for helpers: infants reached for the yielder even after a ‘foot race’ which differed only in that the characters moved in the same direction. While there are several interpretations of why infants reach for yielders, these results provide further evidence that they recognize these conflicts as social interactions potentially relevant to themself.

Other studies provide some evidence that infants do represent asymmetrical relationships: they expect hierarchical relations to be stable across situations. In one study, 12- and 15-month-old infants in Hungary watched two animated characters compete for the same object, with one character yielding the object to the other. Later, during a conflict over territory between the same characters, 15-month-olds—and possibly 12-month-olds—expected the character who had previously given up the object to also give up the territory. Importantly, infants in this study did not expect past winners to prevail in a conflict with a new individual, suggesting that their expectations were not based on the traits of the individual but rather the relationships between the individuals (Bas & Sebastian-Galles, Reference Bas and Sebastian-Galles2021; Mascaro & Csibra, Reference Mascaro and Csibra2012) (refer to Figure 5). At 17 months, toddlers expect rank to be related to resource distribution. When all else is equal, Italian and U.S. American infants expect equal distribution—looking longer when an individual gets more than another (Geraci & Surian, Reference Geraci and Surian2011; Schmidt & Sommerville, Reference Schmidt and Sommerville2011; Sommerville et al., Reference Sommerville, Schmidt, Yun and Burns2013). However, 17-month-old toddlers from the US seem to expect that those who prevail in conflicts will get more resources (Enright et al., Reference Enright, Gweon and Sommerville2017).

Figure 5.

Stimuli used in Mascaro, 2012 and schematic (added here) of the proposed representations driving the inferences. (A) What infants observed in the study. (B) Hypothesized cognitive representation (C) What infants observed during the test trial and the prediction that they made.

Infants also use transitive inference to infer asymmetrical relations. Since transitive inference is useful only if there are stable relations, this provides further evidence that infants represent individuals and their asymmetric connections. For example, in forceful conflicts over a toy, 10-month-old infants from the United States use transitive inference to predict who will defer (Gazes et al., Reference Gazes, Hampton and Lourenco2015). When infants first saw a cow puppet forcibly take a toy from a bear puppet, then the bear forcibly take a toy from a pig, infants looked longer when the cow puppet yielded to the pig puppet, suggesting it violated their expectations (Mascaro & Csibra, Reference Mascaro and Csibra2014). Relatedly, in another study, French 14-month-olds expected dominance relations to generalize across allies, learning ‘pyramid’ structures of dominance relations but not ‘tree’ structures (refer to Mascaro et al., Reference Mascaro, Goupil, Pantecouteau, Depierreux, Van der Henst and Claidière2023 for a longer discussion about why pyramid structures occur more often than tree structures).

In summary, in the first year of life, infants use cues of formidability to predict how individuals will behave in conflicts. Around the end of the first year, they seem to make sense of social interactions by representing asymmetrical relationships: using prior behavior in social interactions to predict behavior between those individuals in new and different social interactions. However, none of these studies ask whether these interactions lead infants to predict whether people will interact with one another. For example, would they expect the individual who deferred over a resource to be more likely than someone who had not been in the interaction to help the high-ranking individual? Thus, it is not clear that they are representing relationships or relations between individuals.

Dominance or Authority Ranking?

So far, I have reviewed evidence that young humans understand cues of dominance found in other species and that at least around the age of one, infants may recognize asymmetrical relationships. However, especially in human relationships, rank is often based on more factors than the ability to inflict harm by being larger or by having more allies. In Relational Models Theory, authority-ranking relationships (A. P. Fiske, Reference Fiske1992; Henrich & Gil-White, Reference Henrich and Gil-White2001), higher-ranked individuals are expected to provide benefits, such as cultural knowledge, protection, and leadership. In these relationships, people are often made to appear larger by sitting on a throne or wearing a crown, which could be seen as ways to communicate formidability (Schubert et al., Reference Schubert, Waldzus, Seibt, Semin and Smith2008). However, a person’s rank does not need to be the result of true physical formidability. While the evidence so far suggests that infants recognize dominance relations, do they distinguish between dominance and other types of rank?

Some studies suggest that during the second year of life, humans begin to distinguish between legitimate authority and dominance. First, consider the finding described above where 10- to 16-month-old infants reach for a puppet who yields in a conflict. When toddlers aged 21 to 31 months (drawn from the same U.S. American population) were presented with the same puppet show, they reached for the winner (Thomas et al., Reference Thomas, Thomsen, Lukowski, Abramyan and Sarnecka2018). This reaching did not depend on competence: they did not reach for a winner when it won a ‘foot race’ instead of a face-to-face conflict. While there are several explanations for this developmental change (e.g., refer to Thomas, Reference Thomas2018 for a discussion), participants may have understood the conflict differently. Infants see these conflicts in terms of dominance and toddlers see them in terms of authority or prestige. If so, the change in representation could be the result of a better ability to represent motivations. In support of this view, toddlers no longer reach for the winner when the winner uses coercive force to win.

Seventeen-month-old toddlers also expect leaders to provide benefits: they expect that a puppet whose directions were previously followed will be the one to intervene in conflicts and punish those who did anti-social actions (Stavans & Baillargeon, Reference Stavans and Baillargeon2019). U.S. American and Italian toddlers form different expectations about future interactions, depending on the way that a high-ranking individual treats subordinates. Twenty-one-month-old toddlers saw one of two scenes. In the leader scene, a high-ranking character directed low-ranking characters, in the bully scene, the high-ranking character enforced its demands by hitting the subordinates on the head. When the leaders were absent, toddlers expected the subordinates to follow the requests of the leader to go to bed, but not the bully (Margoni et al., Reference Margoni, Baillargeon and Surian2018). However, when the bully was present, toddlers expected the subordinates to obey. Interestingly, this study hints at inferences about the strength of an authority ranking relationship—the subordinates follow the leader’s requests in the leader’s absence, and thus seem to feel obligated or committed to the leader across time.

Equality Matching

In Relational Models Theory, an equality matching relationship is symmetrical and occurs when people keep track of what they give or receive and are motivated to maintain an even balance. Equality matching relationships can be out of balance, but people are aware of the imbalance and are motivated to return the relationship to balance. Examples of stronger equality matching relationships would be people in a long-term business relationship or neighbors who have been doing favors for one another for many years. Examples of weak equality matching relations include children taking turns at the playground and people waiting in a queue for coffee.

Evidence of Equality Matching in Non-Human Species

Some behaviors in non-human species indicate that there might be phylogenetic precursors to the logic underlying these relationships. Vampire bats share more of their blood meal with bats who have shared with them in the past (Denault & McFarlane, Reference Denault and McFarlane1995). There is reciprocal grooming among bonded females of white-faced capuchins, chacma baboons, and bonnet macaques (Manson et al., Reference Manson, David Navarrete, Silk and Perry2004; Muroyama, Reference Muroyama1991; Perry et al., Reference Perry, Baker, Fedigan, Gros-Louis, Jack, MacKinnon, Manson, Panger, Pyle and Rose2003; Ventura et al., Reference Ventura, Majolo, Koyama, Hardie and Schino2006). Experiments also show that captive brown capuchins may use the logic of reciprocity: previous sharing predicts whether food-taking is tolerated (de Waal, Reference de Waal2000). While it is unclear how these behaviors arise, it leaves open the possibility that this logic is evolutionarily old.

Evidence that infants are sensitive to cues of equality matching.

Do infants represent equality matching relationships? I will first review studies about infants’ expectations about resource distribution, though of course, each relational model can be applied to situations in which resources are shared or distributed. U.S. American and Italian infants expect that resources will be distributed equally when they are given no prior information about the relationship between the interactants. For example, at 19 months, 9 months, and 4 months, infants expect people to divide two resources equally rather than unequally between two animated puppets or characters (Buyukozer Dawkins et al., Reference Buyukozer Dawkins, Sloane and Baillargeon2019; Sloane et al., Reference Sloane, Baillargeon and Premack2012; Surian & Margoni, Reference Surian and Margoni2020; Ziv & Sommerville, Reference Ziv and Sommerville2017). Importantly, across these studies, these expectations depend on the recipients being agents as opposed to inanimate objects.

By 16 months, infants also seem to prefer those who distribute equally. They reach more often for a photo of a puppet who had distributed resources equally than for a photo of a distributor who had given all the resources to one of two characters (Geraci & Surian, Reference Geraci and Surian2011). Moreover, 16-month-olds expect a puppet who observes uneven versus even distributions to approach the even distributor. Likewise, 13- and 16-month-old infants choose to accept a toy offered by a person who allocated equally to two others as opposed to a person who allocated unequally (Lucca et al., Reference Lucca, Pospisil and Sommerville2018). By 25 months, toddlers are more likely to help a person who has distributed resources equally to two agents than a person who distributed resources unequally (Surian & Franchin, Reference Surian and Franchin2017). Each of these studies indicates an expectation or positive judgment of equal distribution. That is, given no information about relationships, infants seem to expect that two individuals will be given the same amount. But do these studies shed light on how representations of equality-matching relationships? While they do provide evidence that infants have early expectations about resource distribution, they do not tell us whether seeing equal resource distribution prompts infants to predict future interactions. Rather infants seem to infer traits of the distributor. (Mascaro et al., Reference Mascaro, Goupil, Pantecouteau, Depierreux, Van der Henst and Claidière2023) By contrast, an equality-matching relationships should be marked by even balance over time and across contexts between individuals.

Other studies suggest that infants have the cognitive tools to recognize these relationships. There are two studies in which toddlers use interactions in one situation to make predictions about interactions in another situation. In one study, 21-month-old toddlers were shown two characters who were asked to clean up toys (Sloane et al., Reference Sloane, Baillargeon and Premack2012). When both characters helped, they expected equal distributions, but when one helped, they expected the more helpful character to get more. However, this study does not directly test the relationship between the recipients of the resources, since the requester/distributor is a third party.

In another study with 17-month-olds, they saw that one character worked harder on a task, and then the characters distributed resources to themselves (Wang & Henderson, Reference Wang and Henderson2018). Here infants expected the one who worked harder to get more. Thus, toddlers tracked the extent of equality in one context (working) to make predictions in another context (resource distribution), suggesting that they may have inferred a relationship.

Denis Tatone and colleagues have more directly tested whether infants have the cognitive tools to represent equality matching relationships. They reasoned that infants should have different expectations about giving actions, where someone intentionally gives objects, as opposed to ‘tolerated’ taking, where someone still loses an object, but the intentions of the interactants are more ambiguous. In a series of studies, this was shown to be true. Here, Hungarian 12-month-old infants were shown interactions in which one character either gave an object to another character or took an object away from another character. The question was, would infants be more likely to encode the identities of the actor and the receiver when the actor gave compared to when the actor took? After watching giving interactions, infants seemed to expect that the ‘giver’ would continue to give to their social partner: they looked longer when the giver took something from, as opposed to when they gave something to, the original recipient (Tatone et al., Reference Tatone, Geraci and Csibra2015). Moreover, infants did not seem to encode the interactions by representing traits (e.g., ‘generous’) because they did not expect the giver to give to an individual who had not previously been in the interactions. The pattern of results was different after seeing that an actor took an object. Infants were first shown scenes where an actor took something from a character. Here, infants did not distinguish between scenes where the actor took again from the character and scenes where the actor gave to the character (Tatone et al., Reference Tatone, Hernik and Csibra2021). Together, these studies suggest that infants pay special attention to giving actions. They encode what object was given, suggesting that they can keep track of equalityFootnote ¹⁰ . But they also track the identities of the interactants and the direction of these actions, suggesting they could encode the relationship. Since equality matching relationships require tracking equality, these are necessary components to recognizing these types of relationships. But do infants expect that the balances will be evened out?

In another set of studies, infants not only encoded these aspects of the interaction but also expected reciprocation of giving actions. For example, in one study, they saw one character (A) give an object to another (B). Then, they saw a third character (C) take an object from B. They then expected B to give to A, as opposed to give to C. They also expected B to give to A, compared to B taking from A (Tatone, Reference Tatone2017; Tatone & Csibra, Reference Tatone and Csibra2020)Footnote ¹¹ . Further studies suggest that infants distinguish between characters who are given an item, as opposed to when those characters engage in ‘tolerated taking’ (i.e., take an item without protest from the other character). In both actions, a character ends up with one more resource, but infants seem to see the interactions in different ways. When infants observe giving actions, they encode the direction of giving and receiving. However, when they observe tolerated taking, they encode neither. Tatone and Csibra argue that tolerated taking is more diagnostic of communal sharing relationships, in which are mutual, not symmetric and in which people do not track the direction or value of shared resources. Similarly, forceful taking should be more diagnostic of authority ranking, where one needs to track the direction but not necessarily the value of the object (since it will not be repaid). Equality matching is different—the interactants must keep track of the direction of giving and the object that is transferred, since part of the relationship is keeping an even balance (refer to Tatone, Reference Tatone2017; Tatone & Csibra Reference Tatone and Csibra2020; and footnote for a longer discussion).

Taken together, these studies provide evidence that infants encode the necessary parts of social interactions that would enable them to represent equality matching relationships. They have at least rudimentary expectations of even balance. Importantly, these expectations are not the result of infants inferring traits of individuals, as they do not expect individuals to perform the same actions toward individuals not involved in the interaction. However, future studies could more directly test whether infants represent equality matching relationships, in which equality matching is maintained over time and across contexts. For example, one could ask whether infants expect even balance in one situation (e.g., giving resources) to generalize to even balance in another situation (e.g., taking turns). Moreover, that they expect that two individuals who maintain even balance will be more likely to interact with one another than those who had not previously interacted or interacted in a different way (refer to Figure 6).

Figure 6.

Logic of a possible study asking whether infants recognize equality matching relationships and use this representation to predict future interactions.

Do infants compute strength for equality matching relations? So far, there is no evidence that this is the case. For equality matching relations, trust may be the important dimension: the larger the imbalance is allowed, the stronger an equality matching relation might be. For example, if I do many favors in a row for you and we are in an equality matching relationship, it could signal that I trust you will eventually ‘repay’ the debt. In many cultures, it is customary to give gifts whose value cannot be directly computed. This allows a sort of never-ending chain of gift-giving in which the debt can never be perfectly evened out (Carmichael & MacLeod, Reference Carmichael and MacLeod1997; Graeber, Reference Graeber2012; Mauss, Reference Mauss1925; Fiske personal correspondence). Future studies could ask whether young humans make different inferences depending on the size of the inequality allowed in repeated interactions, which would imply that they are computing something about the size of the imbalance in an equality matching relation.

Box 1: Innate Moral Core or Innate Representations of Relationships?

How should we reconcile the claims I have made in this paper with the claim that infants have an ‘innate moral core’ (Woo et al., Reference Woo, Tan and Hamlin2022)? This claim is supported by studies which have found that infants preferentially look toward and physically reach for helpers over hinderers, imitators over non-imitators, and those who distribute equally over those who do not (Hamlin, Reference Hamlin2013; Woo et al., Reference Woo, Tan and Hamlin2022). Arguments for an innate moral core implies that infants infer traits of the individuals (i.e., using the logic of, ‘if they are nice to them, they will be nice to me’). But what traits do infants prefer? One way to interpret these findings is that infants reach for those who initiate or maintain prosocial relationships. Infants prefer those who help—which could be seen as one way to begin a communal sharing relation or an equality matching one; those who imitate, which could be interpreted similarly; those who distribute resources equally, which could be seen as establishing an equality matching relationship; and those who defer in a conflict, which could be seen as an individual who is establishing an asymmetrical relationship.

Consider that infants prefer imitators. As discussed previously, at least 12-month-old infants seem to see imitation as evidence of relationships. We had replicated the finding that 12-month-olds reach for imitators over non-imitators and unexpectedly found that infants reach for those who are not imitated over those who had been (Thomas et al., Reference Thomas, Saxe and Spelke2020). Given this asymmetry in preferences and the asymmetry in expectations of 4-month-olds (remember, 4-month-olds expect imitators to approach the target of their imitation but do not expect the target of imitation to approach those who imitate them), we investigated whether older infants’ expectations matched their reaching preferences.

First, we found that infants expected imitators to respond to their social partner’s distress, as opposed to those who did not imitate. However, these expectations could be the result of either an inferred trait (the puppet who imitated is more responsive) or a relationship (the puppet imitated that individual and so will respond to that individual’s distress). To disambiguate these possibilities, in the next experiment, rather than showing an imitating puppet and a non-imitating puppet, we showed infants a scene in which one puppet was imitated by the central actor and one was not. Being imitated does not give direct evidence of one’s traits (one could be imitated without knowing it). To our surprise, infants also expected that the imitated puppet would respond to their partners’ distress. In a third study, we found that infants did not have these expectations when a person uninvolved in the original interactions was in distress, suggesting that they did not infer traits. Finally, we found that infants do not expect imitative partners to respond to laughter, again suggesting they do not see imitation as a cue of who is more responsive, even toward their social partner. Taken together, these results present a puzzle: why do infants reach for imitators if they do not expect them to be more responsive to everyone? While infants may not expect the puppet to respond to all people’s distress, they may still expect the puppet who imitates another individual to be more willing to create relationships. This would also explain the preference for non-imitated puppets, as the social initiation was not done by the imitated puppet and the non-imitated puppet is more socially available.

Admittedly, it is unclear why infants would prefer those who initiate relationships without inferring some sort of prosocial trait. These people might be seen as ‘friendly’, ‘helpful’, etc. If so, then it is unclear why infants seemingly generalize to themselves but not others. Another possibility is that infants generalize prosocial behavior toward themselves because of the way that experiments are set up. In many experiments, the people and puppets direct their attention toward the infant to keep their attention. Infants therefore may feel they are part of the social interaction. Their evaluations may, therefore, depend on these signs of awareness. There are two exceptions—one is in the hill paradigm in which the puppet’s eyes are directed up the hill (Hamlin, Reference Hamlin2014; Scarf et al., Reference Scarf, Imuta, Colombo and Hayne2012). In this paradigm, however, the puppet may still appear to be looking at the infant, since both eyes are present (of note, reaching preferences of 6 month olds was not replicated in a multi-lab replication attempt; Lucca & Henderson, Reference Lucca and Henderson2021). The other exception is in the right-of-way conflicts, in which infants reach for those who defer in a conflict. Here the puppets’ faces appear to be looking away from the infant and toward its goal. While these two experiments raise questions about this proposal, it may be worth testing empirically. If so, some of the experiments reviewed in this paper only rule out broad generalizations. Of note, there are studies that suggest toddlers do generalize from a prosocial behavior in one context to a prosocial behavior in another context with new targets (Gill & Sommerville, Reference Gill and Sommerville2023; Surian et al., Reference Surian, Ueno, Itakura and Meristo2018). For example, 14- to 27-month-old toddlers expect that a helper, but not a hinderer will distribute resources fairly to two new individuals. They also expect that someone who distributes things unfairly will hinder. Thus, it is possible that the ability to infer traits (which seems to be driven by tracking antisocial behavior) happens sometimes during toddlerhood or that different actions lead to different generalizations.

Box 2: Naïve Utility Calculus and Representing Relationships

Are relationships most basically understood in terms of shared or adopted utility (Powell, Reference Powell2021; Tatone, Reference Tatone2017)? In a recent proposal, Powell argues that representations of affiliation compute whether people adopt the utility of others. This can happen when someone helps another person (i.e., takes on a cost so that another person can achieve their goal) or imitates another person (i.e., takes on a cost to achieve the same goal as someone else). One straightforward connection to the current proposal is that relationship strength is computed by adopted utility (Powell does not make this claim). While the connection between strength and utility is appealing, some examples call the connection into question. First, humans, including toddlers (Warneken & Tomasello, Reference Warneken and Tomasello2006; though see Barragan & Dweck, Reference Barragan and Dweck2014) are happy to adopt the utility of strangers. These toddlers’ expectations of future interactions is unknown, however adults help people whom we never expect to see again (e.g., helping a stranger pull their car out of a ditch). Young adults report taking on similar costs for strangers, acquaintances, and close relations (McGuire, Reference McGuire1994). Perhaps one way of reconciling this tension is that as adult observers, if we find out that a person adopts the utility of a stranger, then we shift our inference to be about traits instead of relationships (i.e., a parent who donates their kidney to their child is a good parent, but a person who donates their kidney to a stranger is a good person). The extent to which infants adjust their evaluations of prosocial behaviors based on different types of evidence could be tested.

We also fail to adopt the utility of even our closest relations. Take a parent who fails to adopt the utility of their toddler who wants another bowl of ice cream or a parent who fails to adopt the utility of an infant who reaches for a hot cup of coffee. An adult observer would assume that the parent’s long-term goal is to increase the utility of the younger human. For example, the parent wants to help their child make healthy eating choices or avoid a burn. But do toddler or infants understand this? If not, it would lead to the uneasy hypothesis that infants and children do not see themselves in a strong relationship with their caregivers. Of course, infants and children are excellent statistical learners (Saffran, Reference Saffran2020), so it could be that unhelpful behavior from parents are overpowered by more helpful behavior. As a parent, I find this hypothesis to be less than straightforward, but it is an empirical question.Footnote ¹² Interestingly, children do not seem to expect more altruism between family members than between friends. When asked whether someone will share with either a family member or friend, children do not expect that others will preferentially share with or help family members compared to other affiliates (Marshall et al., Reference Marshall, Gollwitzer, Mermin-Bunnell, Shinomiya, Retelsdorf and Bloom2022; Spokes & Spelke, Reference Spokes and Spelke2016; Thomas et al., Reference Thomas, Garcia, Sommerville and Saxe2023; Thomas, Woo, et al., Reference Thomas, Woo, Nettle, Spelke and Saxe2022).

Perhaps a better candidate to infer relationship strength could be intimacy, such as physical or emotional intimacy (refer to Steele & Thomas; Reference Thomas, Steele, Gopnik and SaxeThomas, Steele, Gopnik, Saxe for longer arguments). These cues may have become valid because of the costs and benefits associated with sharing bodily fluids and close physical contact. However, I think it is unlikely that infants recognize this. Disgust, especially social disgust, seems to develop relatively late (Rottman, Reference Rottman2014), suggesting young humans don’t recognize overcoming ‘disgust’ as a cost. Infants differentiate actions that are difficult to distinguish based on utility alone (saliva-sharing vs. forehead touching, for example). Likewise, adults may see responses to distress costly because of emotional vulnerability, but it seems unlikely that infants see it this way. Ultimately these are open questions. To sum up, it seems that adopted utility could be one way we compute strength, but perhaps not the only way and it is unclear whether utility computations occur in all representations of relationships.

Can adopted utility calculus be used in the computations that allow us to recognize models? One can describe the models in terms of adopted utility: a communal sharing relation is one where we adopt each other’s utility, an asymmetrical relationship is one where utility is adopted for one person but not the other, and a reciprocal or equality matching relationship is one where we adopt each other’s utility exactly (this is not Powell’s proposal but refer to Tatone Reference Tatone2017).

However, some examples also call this description into question. Another proposal is that the communication of utility is more important. A distinctive feature of communal sharing relations is that individuals do not keep track of relative costs and rewards. People’s skills, resources, and ‘rewards’ are pooled, and people don’t communicate about how much people are contributing or how much people are taking. In fact, in some situations, when people keep careful track in what ‘should be’ a communal sharing relation, it can seem cold or even evidence that the relationship is not a communal sharing one. For example, in the novel The Joy Luck Club, a married couple tries to keep everything strictly equal, and the implication is that they are not intimate.

Second, in asymmetrical relationships, it is not whether one person adopts the other’s utility that is the key distinction. At least with adults, helping itself can be interpreted as communicating both high and low status (Nadler & Chernyak-Hai, Reference Nadler and Chernyak-Hai2014). Subordinates often adopt the utility of leaders, but ‘mansplaining’ is critiqued because it implies that the man thinks they are higher status than the target of this explaining. Perhaps a better indicator is who decides whose utility is adopted. For example, take an academic mentor and mentee—both parties adopt the utility of one another in different circumstances. The mentee does not expect to get the largest office and so ‘takes a cost’ for their mentor. However, the mentor is expected to adopt the utility of the mentee in providing guidance. However, the mentor could probably give up their office and could decide guidance would do more harm than good with no penalty.

In equality matching relationships, a key marker is symmetry which requires that people keep track of whose utility is being adopted and communicate about imbalance. This is apparent when someone buys a coffee during one meeting and the other insists it is their turn to buy during their next meeting. There is an agreement, often explicit, that the utilities of both parties will be adopted at equal rates. Likewise in groups where voting occurs, the utility is matched in a way: each person has the same amount of influence. However, the utility can end up being asymmetrical in which some people’s goals are adopted. Yet, this wouldn’t necessarily change the relationship to be asymmetrical.

To sum up, it is possible, and perhaps likely, that the logic of naïve utility calculus plays a role in our representations of relationships in some contexts. However, it is less than clear that such computations could cover all relationship representations. An open questions is what computational framework could.

Box 3. Groups and Relationships

In this paper, I have argued that there are related but distinct representations of traits, relationships, and groups, each leading to different generalizations. Representations of relationships are constrained to the individuals involved, while representations of groups allow for generalizations beyond the individuals. Much of the work used to argue that infants represent communal sharing relationships was originally taken as evidence that they represent ingroups (Powell & Spelke, Reference Powell and Spelke2013; Spelke & Kinzler, Reference Spelke and Kinzler2007). This work was inspired by decades of research in social psychology showing that people and young children are prone to thinking about others in terms of ingroups and outgroups (refer to Dunham, Reference Dunham2018; S. T. Fiske, Reference Fiske2018; Shutts & Kalish, Reference Shutts and Kalish2021; Tomasello, Reference Tomasello2020). Indeed, the term ’folk sociology’ was coined to argue that humans, including young children, are prone to thinking about aggregates of people (L. Hirschfeld, Reference Hirschfeld, Buchanan-Barrow and Barrett2005).

However, in the work with infants that I review, there is little direct evidence that infants represent groups. Infants see two or three individuals in the same physical space doing something that makes them more alike. Then, infants’ expectations about those same individuals are tested. There is no evidence that infants generalize beyond the individuals in the scenes. While these studies leave open the possibility that infants’ expectations are based on representing groups, it is just as likely that they make predictions based on inferred relationships. Take, for example,(Powell & Spelke, Reference Powell and Spelke2013): in this study, three individuals coordinate their actions. We don’t know whether infants see the three individuals as an aggregate or whether they see three relationships. Would infants expect affiliation when they get evidence that people speak the same language but not during their social interaction? Furthermore, would infants expect people who are part of an aggregate will speak the same language?

There are reasons to think that infants’ ecological niche has led infants to represent relationships and social networks, as opposed to groups. Throughout our evolutionary history, care has mostly come from ‘mothers and others’ – that is, your primary caregivers and those with whom they have relationships (refer to Hrdy Reference Hrdy2009). While it is true that these ‘others’ are also usually part of your mother’s group, there are many people in the group who do not provide care. Therefore, it may be more important for infants to track individuals and their relationships. As discussed in the introduction, human infants, as well as other animals track individuals, a prerequisite in tracking relationships, but not necessarily a prerequisite in tracking group membership.

Thinking about larger coalitional groups may become important as children are exposed to more people in and outside of their social networks. There are several studies with older children in which they do seem to represent groups. They are sensitive to the ‘minimal group effect’ in which they generalize based on arbitrary cues (Baron & Banaji, Reference Baron and Banaji2006; Chalik & Dunham, Reference Chalik and Dunham2020; Dunham et al., Reference Dunham, Baron and Carey2011; Tajfel, Reference Tajfel1982; Yang & Dunham, Reference Yang and Dunham2019). Importantly, these expectations and biases occur without direct evidence of interpersonal relationships (Chalik & Rhodes, Reference Chalik and Rhodes2020; Yang & Dunham, Reference Yang and Dunham2019) suggesting that children represent groups as I have defined them. Representations of groups may arise from ’scaling up’ from representations of relationships, but this process of scaling up would involve generalizations beyond the direct observation of behavior of or between individuals (refer to Figure 7). We may start by mapping out individuals and social networks, and over time, we notice or are told that certain shared characteristics help delineate these networks. This would be similar to the hierarchical learning theorized to be important to early learning in other domains (Ullman & Tenenbaum, Reference Ullman and Tenenbaum2020).

Figure 7.

Here I illustrate what I mean when I propose that groups inferences may arise from making inferences about interpersonal relationships. One could map out relationships and networks, learn ‘borders’ of social networks, then learn (through explicit or implicit teaching) that characteristics delineate networks, including how people interact within and across networks (refer to Pietraszewski, Reference Pietraszewski2021 and accompanying commentary as well as Cikara, Reference Cikara2021for more detailed proposals about how that may happen.) More people could be added to the network until you are no longer representing nodes and lines but instead an aggregate with fuzzy borders.

Exactly what leads people to infer group membership, and what kinds of predictions follow and when, is not a trivial questionFootnote ¹³ (Cikara, Reference Cikara2021; L. A. Hirschfeld, Reference Hirschfeld2001; Pietraszewski, Reference Pietraszewski2021). I add to the problem by proposing that people’s representation of social groups includes more than group membership; they also represent the social structure of groups and the relative strength of those groups (Lickel et al., Reference Lickel, Rutchick, Hamilton and Sherman2006) Children recognize group structure and use these structures to generalize about the behavior of group members without any direct evidence of their interpersonal relationships or the behavior of individuals (Thomas, Mitchell, et al., Reference Thomas, Mitchell, Sumner, Terrizzi, Piff and Sarnecka2022). People may also think about relationships between groups, such as some groups being higher ranked than others (refer to work on social dominance theory and system justification theory (Jost et al., Reference Jost, Banaji and Nosek2004; Pratto et al., Reference Pratto, Sidanius and Levin2006), as well as evidence that children connect identity and status e.g., Mandalaywala et al., Reference Mandalaywala, Amodio and Rhodes2018) or some groups being more likely to work together while others are more likely to be in conflict (refer work on coalitional psychology; Cikara Reference Cikara2021;Gershman & Cikara, Reference Gershman, Cikara and Learning2020). People, including children, see aggregates as though they are entities with single intentionality (Bloom & Veres, Reference Bloom and Veres1999; Sheikh & Hirschfeld, Reference Sheikh and Hirschfeld2019), suggesting that we may call on our intuitive theory of relationships when considering relations between aggregates.

Box 4: Navigation, Space, and Social Relationships

Many scholars have invoked the metaphor of navigation when discussing how we reason about social relationships (Banaji & Gelman, Reference Banaji and Gelman2013). Considering recent research into the hippocampus (Montagrin et al., Reference Montagrin, Saiote and Schiller2018; Tavares et al., Reference Tavares, Mendelsohn, Grossman, Williams, Shapiro, Trope and Schiller2015) as well as theories from linguistics (Jackendoff, Reference Jackendoff2009), this may be appropriate. Our reasoning about navigation allows us to represent where we are, where we can go and gives us an anchor for memories (Spelke, Reference Spelke2022). When I think about my hometown, I don’t think of ‘a small town in the central, south part of the west coast’ I think about ‘Ojai’. Likewise, when I reason about my daughter, I don’t reason about a small child with blond hair, blue eyes, and an impish smile, I think ‘Ursula’ (Spelke, personal correspondence). There may be overlaps in the way we reason about nodes as locations in space and nodes as locations in social networks. First, in many languages, people use spatial terms to describe social relationships and social interactions. We can locate ourselves or others in a family tree or a social network. You can feel ‘close’ to someone or ‘distant’. Someone can stand ‘above you’ or be on ‘equal footing’. You can even be in the ‘same circle’ as someone. None of these descriptions need people to be in the same physical space to have meaning, yet they are easily understandable. In this paper, I have been vague about the formal computations that are involved in our representations of social relationships. However, graph theory may be one avenue of research that could be fruitful in describing our cognition of social relationships with formal computational models (Kemp & Tenenbaum, Reference Kemp and Tenenbaum2008).

Another way that our reasoning about navigation may connect to our reasoning about social relationships is the computational modeling work done about planning. A recent proposal connects intuitive psychology and spatial navigation (Ho et al., Reference Ho, Saxe and Cushman2022). There may also be similarities in how we think about relationships. The proposal is that sometimes when we think about relationships, we think about them as objects in the world. Other times, when the status of our relationship is uncertain, or when we are trying to change a social relationship, we might go through the computationally costly process of simulating possible social actions, their consequences, and how this might affect the minds of our social partners and subsequently the relationship. For example, when someone knows that they are in a committed, communal sharing relationship it might mean that they go through actions the same way one does when they are in ‘auto-pilot’ mode in the car. You kiss your spouse goodbye without thinking twice about it. Errors such as calling your granddaughter the name of your daughter may be evidence of this kind of ‘auto-pilot’ mode where we represent the relationship with ‘just enough’ detail. On the other hand, when we think about changing or creating relationships, we might simulate many possible social actions or conversations that could take place. These ideas are tentative and speculative. More work could explore how the cognition involved in spatial navigation, planning our movements in space, and social ‘navigation’ may be connected.