At the core of Thomas’ proposal is the claim that, beginning in infancy, humans are prepared to interpret the gamut of social relationships by appealing to three distinct models: communal sharing, authority ranking, and equality matching (as originally laid out by Fiske, Reference Fiske1992). Expanding upon previous attempts to frame early relational concepts within a Fiskeian framework (Thomsen & Carey, Reference Thomsen, Carey, Banaji and Gelman2013; Tatone, Reference Tatone2017), Thomas expertly reviews a wealth of developmental evidence to argue that infants spontaneously form expectations consistent with the normative standards of these models (i.e., social equivalence, power asymmetry, and even balance). Ancillary to this proposal is the suggestion that, beyond categorizing a connection by model, infants may also compute its strength. Despite sharing the classificatory framework adopted by Thomas, I argue that the introduction of this second component into her account of naïve sociology is problematic for multiple reasons.

Thomas defines strength as a continuous representation of obligations (the extent to which certain actions are expected), and commitment (the likelihood that people will continue the relationship). Critically, strength does not define a continuum along which to organize relational models, but a parameter specifying the range of social expectations within a certain model. In other words, relational strength is computed only after interactions have been categorized under one of the three available models. Relationships falling under each model can consequently be weaker or stronger depending on their inferred strength. There are a few unstated implications of this proposal that are worth unpacking.

This account assumes that the observation of episodic interactions, like the ones that infants are presented with in the studies here reviewed, is sufficient not only to determine a model, but the extent of an individual’s commitment to their social partner. This is an assumption that Thomas, albeit tacitly, is forced to make. If strength may be plausibly inferred via bookkeeping asymmetries in social investment over repeated observations (e.g., by tracking the delay and magnitude of reciprocity between partners: Xue & Silk, Reference Xue and Silk2012), neither the brief snapshots of social interactions that infants are provided within the course of an experimental session nor the limited opportunities for social monitoring in their everyday life appear suited for this kind of extended eavesdropping.

There must be, then, local proxies of relational strength that even isolated encounters may bear. The example that Thomas chooses to illustrate this point with is her saliva-sharing study (Thomas et al.): under her reading, infants expect comforting to occur between individuals shown to share saliva, rather than passing a ball, because, altough both actions prime a communal frame, only the latter indexes a strong enough bond to license expectations of social support. As Thomas herself acknowledges, however, it is unclear why infants would categorize these two conspicuously different interactions under the same model. Considering that model identification is the first inferential goal that infants need to secure, and that the type of interaction observed is often the only stimulus feature differing between events in these experiments, it seems more likely that this information would be leveraged to classify interactions with respect to the relationships they token rather than to compute their strength. Corroborating this point, infants have been shown to categorize even superficially similar actions differing only in the direction of resource transfer (giving vs. taking) as instantiating qualitatively distinct exchange models (Tatone & Csibra, Reference Tatone, Csibra, Denison, Mack, Xu and Armstrong2020). A second example that Thomas uses as evidence of strength computation in communal relation is Spokes and Spelke (Reference Spokes and Spelke2017), which showed that toddlers infer an unobserved bond between two agents if comforted by the same “adult”, but not if comforted by different ones. It is unclear how this finding could support graded differences in strength, since the interaction patterns that infants compared were structurally different (one with a common partner, licensing transitive tie formation, and another without). It seems more plausible that infants interpreted these comforting actions as indicative of caregiving relationships, which, when combined in an open triadic arrangement, licensed inferences about siblinghood. Rather than exemplifying continuous strength computations, these findings instead suggest that, within the broader domain of communal relationships, infants may hold specific assumptions about kinship-informative acts and their combination (Lieberman, Tooby, & Cosmides, Reference Lieberman, Tooby and Cosmides2007).

Furthermore, the proposal that social interactions may bear two distinct classes of cues (i.e., model and strenght identifiers), which infants are prepared to attend to, entails an implausibly large database of hierarchically organized cues, which sits at odds with the explanatory appeal of Relational Models Theory for cognitive development. Fiske’s theory gained traction in the field due to providing a parsimonious solution to the “learnability problem”: against the backdrop of infinite inductions about the meaning of interpersonal actions, a few relational primitives (and a corresponding sensitivity to their perceptual indicators) were proposed to suffice to constrain and bootstrap learning in the social domain (Fiske, Thomsen, & Thein, Reference Fiske, Thomsen and Thein2009). In this model, however, the in-built cognitive springboard is drastically expanded (to capture distinctions within each model based on connection strength), making its internal structure harder to justify on developmental grounds. This is not to say that cue integration does not occur in early relational mapping, but it does so primarily to disambiguate models, not to compute the strength of their tokening interactions (cf. Steele et al., Reference Steele, Richardson, Taylor, Tatone and Thomas2024).

Finally, in embedding the concept of strength within that of “interchangeability,” Thomas introduces yet another component in her model of naïve sociology which, while parcel of adult social reasoning, is unlikely to be an inferential target for infants. From the perspective of a naïve observer, it is unclear how observing an isolated interaction may engender considerations about the replaceability of the partners involved. For an observer to entertain said inferences, they would have to appreciate the trade-offs in bond formation that an agent faces so that their future social investments can be appropriately understood as redirected (i.e., A wants to be friends with B instead of C), rather than duplicated (i.e., A wants to be friends with B and C). This type of computation requires a network-level assessment of an agent’s standing, which is presumably beyond infants’ capabilities.

In lieu of graded computations of strength, I believe that narrow relational schemas (e.g., caregiving) may better account for the latitude of social expectations documented in the studies reviewed by Thomas without having to resort to implausible cue architectures.