R1. Preliminaries

I begin by clarifying some theoretical commitments. Basso and Nolan object to my claim that human minds (including infant minds) process information. However, decades of research strongly support the view that humans – including infants – and other animals recognize patterns, solve problems, and learn across diverse domains. This information processing helps us respond to diverse environments, innovate, and build cultural knowledge. My use of the terms “computation” and “information” is not meant to equate human beings with current technology. Instead, information can come from any process that is not random, and computation refers to the transformation of information. This does not mean that humans are maximally efficient, amoral, or self-interested, nor does it imply that people think about human relationships in the same way they think about bank accounts. I remain open to the possibility that people are intentional beings and that information can be processed not only within individual minds but also within bodies or between bodies, as when an orchestra collectively creates music (Vélez et al., Reference Vélez, Christian, Hardy, Thompson and Griffiths2023).

Following Susan Carey’s definition, cognitive representations are states of our nervous system that process information and have content referring to concrete, abstract, or fictional entities, properties, or events (Carey, Reference Carey2009). Importantly, the representations are not equivalent to the things to which they refer: relationships are not the same as cognitive representations of them. For example, just because actual relationships have fuzzy boundaries (as Lavrador and Pietraszewski correctly point out), it does not mean that our understanding of them is not built from primitives that are categorical. Cognitive representations can, for example, change someone’s behavior, allow them to make predictions, or help them imagine alternative possibilities. One type of learning occurs when cognitive representations change or become newly integrated with others, altering how we process information.

I am committed to the idea that looking-time measures are valid indicators of infant cognition. These measures yield reliable and replicable results across various study designs and stimuli, grounded in widely accepted theories of learning (Stahl & Kibbe, Reference Stahl and Kibbe2022). Available neural data suggest that infants as young as two months process information in ways similar to adults, in both cortical and subcortical regions (Kosakowski et al., Reference Kosakowski, Cohen, Herrera, Nichoson, Kanwisher and Saxe2024; Liu et al., Reference Liu, Raz, Kamps, Grossmann and Saxe2023). This evidence supports the claim that looking-time measures can indicate higher-order cognition, challenging arguments that infants lack abstract representational capacity.

I am committed to the possibility that genes and natural selection could have endowed us with learning mechanisms, including representations, making such representations innate. This, of course, does not mean that all infants are identical at birth – a child may be born missing a limb – but that does not mean that humans are not genetically predisposed to have two arms and two legs. While Aydin and Carpendale raise important questions about the complex interplay of genes and experience, this complexity does not preclude the possibility of innate representations (refer also to Nettle & Scott-Phillips, Reference Nettle and Scott-Phillips2023).

That said, my article did not deeply engage with learning. I covered a relatively large amount of developmental time (over 2 years), and many commentators were justified in questioning whether the evidence I provided supports the claims about innate representations. Any proposal regarding early representations of relationships must be placed in the context of the incredible amount of learning that occurs during infancy and childhood. Furthermore, any proposal about how infants come to know about social relationships should also consider how infants and children themselves shape the social interactions in which they participate (not just how they understand them). With these preliminaries out of the way, I will now address the specific issues raised.

R1. Is relationship understanding the result of embodiment?

In the target article, I claimed that infants have abstract cognitive representations. Basso and Nolan, along with Aydin and Carpendale, challenge this claim from an embodiment perspective, suggesting that we understand relationships from physical experiences. However, these embodied accounts face a problem: the concepts they invoke – such as “obligations,” “in/out,” and “balance” – are themselves abstract. If infants can represent obligations or containment relations, they already possess abstract representations. Moreover, how do infants generalize from their embodied experience to novel situations without an abstract representation? Consider my saliva-sharing study: when an infant observes a woman placing her finger in a puppet’s mouth and then in her own, she must determine which features are relevant. An embodiment explanation may be that the infant understands the scene through her body’s response and connections to past experiences. However, how does the infant know which experiences to connect it to in order to make predictions about what will happen next? Even if infants understand scenes through physical reactions, they must still organize these reactions; otherwise, we would not observe the systematic behavior that we do. Abstract representations are particularly well-suited to organize diverse input (Goodman, Ullman, & Tenenbaum, Reference Goodman, Ullman and Tenenbaum2011). These questions also apply to Kanakogi’s argument that the foundation of moral cognition is somatic responses of infants (refer to the “Innate Moral Core?” section later in this document).

R1. Relationship representations and attachment

In the target article, I proposed that good candidates for primitives of relationship representations are Fiske’s three models (communal sharing, asymmetrical relationships, and equality matching). Several commentaries show how this proposal contrasts with proposals about early relationship knowledge in attachment theory (Nandrino and Gynberg, Groh and Luo, Rodriguez et al., Vornlocher and Shiota). The first way this proposal contrasts with attachment theory, on the surface, is the concept of an attachment figure, which is defined as individuals with whom infants form emotional bonds and from whom they seek proximity, comfort, safety, and protection in times of stress or threat. In the target article, I discussed the possibility that the representation, “caregiver” (similar to “attachment figure”), could be the result of an integration of representations of communal sharing relationships and asymmetrical ones. However, I also proposed that infants likely have representations that apply to a larger set of relationships than the specific definition of attachment figure in attachment theory. It is not that, as Groh and Luo suggested, that I misunderstood what attachment relationships are theorized to be. Instead, the target article was not clear enough about the idea that cognitive representations of relationships do not need to map onto “real relationships perfectly.” For example, while it may be true that attachment figures provide protection for infants, it does not necessarily follow that infants perceive these relationships in this way (albeit an interesting empirical question).

Second, according to attachment theory, infants do not have innate representations. Instead, the theory argues that human infants are born with predispositions for behavior (or “sensitivities”; refer to Groh and Luo’s commentary) that give rise to emotional bonds with attachment figures. Specifically, Bowlby proposed that infants are not born with the internal working models, central to attachment theory, which are theorized to give rise to ideas about how relationships work. Instead, these working models are gradually built through experiences with their attachment figures (Bowlby, Reference Bowlby1982)Footnote ¹ . The nature of these relationships and the corresponding internal working models is proposed to come from caregiver responsiveness and synchrony between caregivers and infants. This last point is supported by correlational and experimental evidence, which shows that caregiver responsiveness and synchrony predict behavior in the strange situation task, a method used to categorize the quality of infants’ attachment to their caregivers (Juffer et al., Reference Juffer, Struis, Werner and Bakermans-Kranenburg2017). Thus, while the commentaries that discuss attachment vary in the extent to which they agree that innate representations of relationships are possible, each of them argues that early experiences with caregivers significantly affect or give rise to representations of relationshipsFootnote ² .

Our disagreement centers not on whether early experiences affect representations – they clearly must – but when this happens in development, and on what level of abstraction. In other words, do infants’ experiences with their caregivers create representations of relationships, or do such experiences update existing ones? And at which level of abstraction? Do they create the representation “relationship,” “communal sharing relationship,” “caregiving relationship,” or “my relationship with this particular person”? I am inclined to think that infants learn a lot from interacting with and observing their primary caregivers, as well as the many other people they encounter. I am also inclined to think that, like other domains, this learning updates representations at different levels of abstraction (refer to Ullman & Tenenbaum, Reference Ullman and Tenenbaum2020, for a more detailed discussion on learning at hierarchical levels in development). However, the idea is that the starting state representations enable infants to organize the vast amounts of information they are bombarded with in everyday life, allowing them to learn an incredible amount over the first years of life.

Consider studies often cited as evidence of internal working models: securely attached infants, but not insecurely attached onesFootnote ³ , expect large characters to respond to small ones who display distress. Attachment theorists may take this finding as evidence that infants have learned through experience with their caregivers, the representation of “relationship,” and what it entails. However, it is not the only reasonable explanation of these findings. For example, infants may differ in terms of innate inclinations or temperaments that influence their behavior in the strange situation task, their synchrony with their parents, their parents’ behavior toward them, and their expectations in the experiment (Kagan, Reference Kagan2009; Kleppesto et al., Reference Kleppesto, Czajkowski, Vassend, Roysamb, Eftedal, Sheehy-Skeffington, Ystrom, Kunst, Gjerde and Thomsen2024). Alternatively, my preferred explanation is that infants’ representations are updated, rather than created by experience with their caregivers. In other words, even if individual differences arise from experience with caregivers, it does not rule out the possibility of universal, innate, abstract representations. The proposal in the target article is that innate representations are likely to exist, not that they are static. Similarly, Rodriguez et al. raise questions about innate representations of relationships based on findings suggesting that prenatal experiences affect later relationships. Changes to emotional regulation, communication abilities, or physiology can certainly affect how humans interact and may create individual differences in relationships and people’s perceptions of them. However, again, none of this precludes innate cognitive representations that allow infants to recognize and categorize relationships. For example, in other domains, such as quantity approximation, there are known individual differences in accuracy, as well as extensive evidence that we are evolutionarily endowed with the ability to distinguish between quantities (Carey, Reference Carey2009; Sarnecka, Reference Sarnecka2016; Spelke, Reference Spelke2022b).

The proposal that infants learn about relationships primarily through experience with their caregivers faces an additional challenge: infants demonstrate expectations about other kinds of relationships. For example, infants as young as 6 months expect that individuals with more allies will “win” a conflict. It is unclear how caregiver responsiveness, or synchrony with caregivers, would create such expectations. Moreover, in response to Groh and Luo’s proposal that infants’ behavior in the experiments is the result of sensitivities rather than cognitive representations, the question becomes: how do infants transfer experiences with caregivers to novel individuals and situations? What organizing principle(s) guide these sensitivities to the relevant social cues during their experiences with their caregivers? Finally, if the “attachment figure” is proposed as the representational framework guiding infants’ understanding of relationships, more empirical work is needed to show that infants distinguish this concept from broader categories of intimate, family-like, or communal sharing relationships.Footnote ⁴

A third way the target article did not engage with attachment theory is the central claim regarding the evolutionary origins of attachment – namely, that attachment behaviors evolved as adaptive mechanisms to ensure infant survival by promoting proximity to caregivers in the face of environmental threats. Groh and Luo ask, what evolutionary pressures would lead to innate abstract representations of relationships?

R1. What are the evolutionary origins of these representations?

In the target article, I only briefly covered the evidence that other species share our ability to represent relationships. However, I did not discuss the potential evolutionary origins of early cognitive representations of relationships. The commentaries by Seyfarth and Grossman consider these potential evolutionary origins. Seyfarth provides evidence that other species possess abstract representations of relationships. He also discusses how these representations improve fitness. For example, baboons have hierarchical (not in the social sense) representations of dominance in their social groups, where rank differences occur both across kin groups and within them. Importantly, these representations cannot have a 1:1 correspondence with perceptual features such as size, because rank is inherited. The ability to track specific individuals and their relationships increases fitness: recognizing social relationships informs animals’ decisions to interact with specific individuals. It has significant implications for survival and reproductive success. It is no surprise, then, that there are many pieces of evidence that several other species are endowed with abstract, implicit knowledge of social relations.

However, as Seyfarth points out, the question is how, and whether, ecological niches across species have led to different starting states. For example, would a gorilla, dropped into a chimpanzee group, come to understand the chimps’ social relationships? My prediction is that just as humans can watch chimpanzees and infer aspects of their social relations, so too could a gorilla observing a chimpanzee group (though perhaps not as enthusiastically as a bright-eyed graduate student). Ultimately, it is an intriguing empirical question. While putting a gorilla in a chimp group would be unethical, one open and answerable question is the extent to which other animals perceive abstract animated shapes in a similar way to humans (Gergely & Csibra, Reference Gergely and Csibra2003; Heider & Simmel, Reference Heider and Simmel1944). If they do, it would provide evidence that individuals across species could understand one another’s social interactions. This is not to say that there are no significant differences between what human infants and adults represent, and between those of other species and humans. The idea is that, like in other domains, human infants and other species share skeletal abstract representations of relationships that are starting states, from which learning is built.

Another question raised in these commentaries is how to think about human infants’ ecological niche. Grossman proposes a human and infant-specific evolutionary pressure: human infants are more likely to survive if they can find reliable caregivers in addition to their mothers, arising from many specifics of human reproduction and child rearing, including short interbirth intervals, early weaning, and hairless bodies that prevent human infants from clinging to their mothers (Hrdy & Burkart, Reference Hrdy and Burkart2022). This, he argues, may underlie infants’ ability to represent relationships and may shape which relationships they represent. Grossman uses kiss-feeding as a case study, which is prevalent across human cultures and is a valid cue of caregiving, because, as mentioned above, infants are weaned relatively early and depend on the provisioning of adults to survive. There are a few issues I see with this argument. First, non-cooperative breeders, such as baboons discussed above, seem to have abstract relationship representations. Moreover, non-cooperative breeders engage in similar behaviors. While cooperative breeders, such as marmosets and wolves (Fentress & Ryon, Reference Fentress, Ryon, Harrington and Paquet1982), do engage in kiss-feeding, non-cooperative breeders, including chimpanzees, orangutans, and bonobos, also exhibit mouth-to-mouth contact and lip-smacking as affiliative acts that help relieve social tension. Some monkey species put fingers in the mouths of conspecifics, possibly indicating social trust (one could lose a finger!; (Perry et al., Reference Perry, Baker, Fedigan, Gros-Louis, Jack, MacKinnon, Manson, Panger, Pyle and Rose2003). In fact, Lameira and colleagues suggest that human kissing is “left over” from grooming practices found in other primates, which often end with a lip smack. When humans lost hair, we lost grooming, but we retained kissing as part of their bonding interactions (Lameira, Reference Lameira2024). Of course, the extent to which other species see these practices as cues of intimacy, as well as how much the concepts related to them overlap with those that may arise when infants understand kiss-feeding, are empirical questions, the answers to which could support or refute this proposal.

In summary, while I agree that cooperative breeding has shaped human inclinations, I am hesitant to conclude that it is the sole driver of our ability to represent relationships broadly. I think it is more likely that other proclivities of human infants were shaped by cooperative breeding. For example, infants’ ability to detect potentially human-specific cues of social attention (e.g., infant-directed speech, smiling, direct eye gaze) may be better candidates for human-specific mechanisms to ensure attention from mothers and others. Alternatively, human infants may be especially attuned to the social interactions of their caregivers in order to learn about the social relations of their caregivers, which may be a good cue about what other adults would be reliable caregivers.

R1. Underlying computations

I now turn my attention to the commentaries that disagreed with the specific underlying computations for relationship representations that I proposed in the target article. Powell and Tompkins agreed with the target article that infants may have core knowledge of social relationships. They disagreed with the target article that the primitives are likely to be Fiske’s relational models. Powell and Tompkins’ reasoning centers on my treatment of cues. In the original paper, I argued that infants use specific cues to recognize different relational models. For example, I proposed that imitation was a cue of communal sharing because it makes people unitary. Powell and Tompkins were correct to point out that, in everyday life, many social behaviors, including imitation, occur across the three models: Aside from communal sharing relationships, imitation can be indicative of asymmetrical relationships where low-ranking people often imitate high-ranking people, or equality matching relationships, where people may take turns doing the same thing, in essence imitating one another.Footnote ⁵ While this may be true in actual relationships for adults or older children, young infants may understand imitation as especially indicative of communal sharing. This is an empirical question for future studies (But refer to Bas et al., Reference Bas, Sebastian-Galles, Csibra and Mascaro2022 for some evidence that 15-month-olds may perceive imitation as a cue of asymmetrical relationships).

That said, two commentaries, Papeo, Mello, and Hochmann and Chen, Chandra, and Saxe propose an alternative to the behavioral cues that I focused on in the target article: roles (refer also to Jara-Ettinger & Dunham, Reference Jara-Ettinger and Dunham2025; However, my treatment of them here is different than theirs, there is more below in the discussion of Kaufman’s commentary). Papeo et al. suggest that the primitives of relationships could include concepts of agent and patient that loosely map onto linguistic concepts. Take imitation – infants may distinguish between interactions in which two individuals assume different roles (e.g., imitator and imitated) or the same roles (e.g., imitating one another). Both of these would be distinguished from non-interactions (e.g., people who happen to do the same thing without intending to interact, or perhaps for an instrumental purpose). The studies covered in their commentary, in which infants track the roles of individuals, support this argument. If so, infants should also be able to distinguish between social interactions that are asymmetrical and those that involve an agent and a patient, a question that is being actively tested in my laboratory. That being said, roles are unlikely to be the only way that infants make sense of social interactions. For example, infants distinguish across similarly structured interactions. Infants seem to see vocal imitation, as opposed to mere vocal reactions, as a cue indicating who will respond to distress. Likewise, infants seem to perceive saliva sharing, rather than ball passing, as an indicator of who will respond to distress (Thomas, Woo, et al., Reference Thomas, Woo, Nettle, Spelke and Saxe2022). Across conditions within studies, the structure of the interactions between agents and patients remains the same; yet, infants distinguish between the two conditions. What allows infants to do so?

One possibility is statistical learning. For example, Powell and Tompkins argue that infants have learned a statistical association between physical intimacy and responses to distress. Fabio et al. propose that, while some innate predispositions may exist, statistical learning and cultural scaffolding account for much of infants’ social understanding through repeated observation of social regularities and contingencies. Son more radically argues that infants need not have any innate representations, but could, through statistical learning alone, acquire both relationship representations and knowledge of network structure. I agree that humans, including infants, are prodigious statistical learners. The question is not whether statistical learning informs relationship reasoning – it clearly must – but rather, what minimal starting state enables infants to organize the overwhelming complexity of social information they encounter. Ultimately, as Chen et al. argue, computational models that systematically vary initial constraints could help determine which combination of innate structure and statistical learning best explains the developmental trajectories observed (Griffiths & Chater, Reference Griffiths and Chater2024; Lake et al., Reference Lake, Ullman, Tenenbaum and Gershman2017; Ullman & Tenenbaum, Reference Ullman and Tenenbaum2020).

Another commentary critiques the target article’s proposal that Fiske’s models could be primitives from a different standpoint. In the target article, I proposed that equality matching is a distinct relationship representation. Lavrador and Pietraszewski’s commentary argued that equality matching is not a type of relation but rather underpins all successful social exchanges. The argument that all relationships involve reasoning about equality matching logic comes, at least in part, from the extensive evidence that, for example, humans are sensitive to free riders. However, this reasoning has most often been observed in studies involving strangers who are unlikely to meet again, as well as in games where the primary goal is to earn money. Thus, Lavrador and Pietraszewski’s idea may be truer for interactions outside of interpersonal relationships. Moreover, while pressures to ensure mutual benefits in relationships may make evolutionary sense, it need not follow that our proximate representations, especially the primitives, compute costs and benefits. Indeed, people expect reciprocal generosity in equal relationships, but not asymmetrical ones (Chen & Saxe, Reference Chen and Saxe2025). Shore et al. also engage with the equality matching section of the target article, proposing that intuitions about positive and negative reciprocity may follow distinct developmental pathways due to the different evolutionary pressures that may have led to these behaviors. Their proposal implies that negative reciprocity intuitions may not be about relationships at all. Instead, negative reciprocity might operate more as a general response to harm or threat – potentially rooted in self-defense or status management rather than relational expectations per se. This may be true of direct negative reciprocity, but indirect reciprocity, including negative reciprocity, can certainly reveal something about relationships (refer to Lavrador and Pietraszewski’s commentary and Pietraszewski, Reference Pietraszewski2022). It would be interesting to expand on existing work on friendship (Afshordi & Liberman, Reference Afshordi and Liberman2021) to investigate whether young humans employ both direct and indirect reciprocity to infer relationships that they are not explicitly informed about.

In the target article, I proposed that infants may also recognize the “strength” of relationships, which I proposed was graded or continuous. Tatone argues that more specific relationship categories, such as “family” or “caregiver,” are more likely. He argues that adults compute strength by tracking repeated interactions over time, further suggesting that infants lack both the capacity and the opportunity to do so. First, I disagree that adults necessarily compute strength through direct evidence of repeated interactions. For example, as I argued in the paper, responses to distress may be especially indicative of strong relationships compared to responses to laughter or vocalizations. Second, infants can make graded distinctions (at least ordinally). There is empirical evidence that 10-month-olds track whether an agent incurs a high versus a low cost (Liu et al., Reference Liu, Ullman, Tenenbaum and Spelke2017). There is also empirical evidence that 15-month-old infants use transitive inferences when reasoning about prosocial interactions (Pepe et al., Reference Pepe, Woo, Thomas and Powell2025): when a protagonist helped A over B, and B over C, infants predicted the protagonist would help A over C, despite never having seen them interact directly. Thus, while I acknowledge that the evidence is limited, I disagree with the argument that infants cannot, in principle, track interactions over time or represent them in a graded fashion.

R1. Groups

In the target article, I argued that representations of groups and relationships are potentially distinct, and that many studies initially interpreted as evidence that infants represent ingroups and outgroups could also be interpreted as evidence that infants represent relationships. Shutts and Kinzler question whether it is reasonable to assume that a clear distinction exists between the two representations, ultimately arguing that social relationships are embedded within groups. First, I absolutely agree that relationships are embedded in groups. However, the key issue is not whether these representations may influence one another, but what the underlying representations are and when, in development, they become available to humans. In this paper, I argued that the distinction between a relationship representation and a group representation is the extent to which one generalizes to new people and how.

To expand on what I mean, consider an observer who represents two groups (the Montagues and the Capulets) and specific members of each group (Romeo, who is a Montague, and Juliet, who is a Capulet). If the observer learns the groups are at war, they may be surprised to learn Romeo and Juliet are in love. This is because, as Shutts and Kinzler point out, adults understand that relationships are embedded in groups. However, the observer can understand two things about this situation: first that the groups hate each other, and second, that individual members of each group love each other. In other words, to imagine the consequences of love across group lines, we need both group representation and dyadic representation. Adults can see how they are intertwined: after learning about Romeo and Juliet’s affair, the audience could, in principle, update their expectations of whether other members of the warring groups will get along, but it need not be the case that they do. The tragedy of Romeo and Juliet, and the intuition that the audience can presumably easily grasp, is that Romeo and Juliet believe their families will forbid their love affair and may even deepen the conflict. Our ability to understand this play is only possible because we can hold these two representations separately. My argument is not that in the adult mind, these two kinds of representations operate with entirely different assumptions. For example, I think that relational models also apply to how we characterize groups. My argument is that the extent to which you generalize and individuate people implies different computations. There may also be special assumptions that go with one over the other. Ultimately, it is an empirical question whether infants have access to both representations. I stand by the claim that it is unclear from the current data whether infants represent “social group.”

In asking this empirical question, we must also distinguish between ultimate and proximate causes of behavior. For example, let us assume that human infants are predisposed to notice when people speak the same or different languages, and this predisposition is the result of people being stratified by language groups in our ancestral past. It is still possible that the proximate representation that drives an infant’s behavior in an experiment reflecting that predisposition is not a group representation. It may be limited to the two individuals in the interaction they see. Some evidence for this is the work showing that 6-month-olds only predict friendliness after learning that two people prefer the same foods when those people appear together eating the same foods (Liberman, Kinzler, & Woodward, Reference Liberman, Kinzler and Woodward2021). In other words, the initial social interaction is important. McCaffree makes a stronger claim, arguing that the evidence for group categorization in infancy is “overwhelming.” I obviously disagreeFootnote ⁶ . While it is a worthy empirical question, it remains unsettled. For example, if infants have group representations, they should, at a minimum, make predictions about individuals they see coordinate beyond infants’ attentional capacity to represent and track individual entities (i.e., four entities). To the best of my knowledge, this has not yet been tested, although my lab is currently investigating it.

R1. Social networks

In the target article, I focused on evidence that infants represent dyadic relationships; three of the commentaries (Mascaro; Parkinson and Krems; Ozlu, Denison and Friedman) show how this reasoning may also enable infants to represent social structures that extend beyond dyads. First, I agree with Mascaro when he argues that there is evidence that infants situate dyads into larger social structures. For example, he discusses his work wherein infants use transitive reasoning to infer dominance relationships (i.e., if A is higher ranked than B and B is higher ranked than C, the inference is that A is higher ranked than C) and wherein infants expect pyramidal hierarchies as opposed to a tree structure (i.e., it is easier for infants to recognize that A is higher ranked than B and C, but not that B and C are higher ranked than A; (Mascaro et al., Reference Mascaro, Goupil, Pantecouteau, Depierreux, Van der Henst and Claidière2023; Mascaro & Csibra, Reference Mascaro and Csibra2012). I agree with Mascaro that these abilities cast doubt on whether relationship representations can be reduced to core agency (c.f. Powell and Tompkins’ commentary). While Mascaro does not argue for this himself, I believe that the evidence that infants represent larger structures also raises questions about whether infants reason about groups. If infants were primarily generalizing based on group markers, we might expect them to be less sensitive to structures inferred from individuals and the relationships between them. Of course, they may do both.

I also agree with Parkinson and Krems when they emphasize that more work needs to be done to understand the development of network representations. Recent work from my laboratory suggests that, by early childhood, children are accurately representing parts of their social networks. For example, 3- to 5-year-old children seem to know that their mother has a more intimate relationship with her own mother (the child’s maternal grandmother) compared to the mother’s mother-in-law (the child’s paternal grandmother; (Steele & Thomas, Reference Steele and Thomas2025). Preschool-aged children, as young as three years old, are better at describing the social networks of their classroom than a randomly generated network, even when their own relationships are removed from the networks (Chuey, Gweon, Thomas, in prep). Finally, a recent small study conducted in my laboratory demonstrated that 24- to 36-month-old toddlers were nearly perfect at identifying the parents of their peers in a daycare classroom (Thomas, in preparation). While these studies are starting to shed light on what young children learn about their own social networks, very little is known about infants’ representations of their social networks.

Moreover, as Ozlu, Denison, and Friedman highlight, another question is how young humans use general attributes of social networks to reason about relationships. In other words, do young humans use general properties of social networks to understand relationships without directly observing them?Footnote ⁷ For example, while children infer that people with more mutual connections are more likely to be friends, adults are also sensitive to other attributes of social networks. And, while there is some existing evidence that infants are sensitive to principles such as triadic closure, much more work is needed to investigate whether infants reason using assumptions based on the general attributes of social networks.