Evidence for core social goal understanding (and, perhaps, core morality) in preverbal infants

Abstract

Spelke's What Babies Know masterfully describes infants’ impressive repertoire of core cognitive concepts, from which the suite of human knowledge is eventually built. The current commentary argues for the existence of a core concept that Spelke claims preverbal infants lack: social goal. Core social goal concepts, operative extremely early in human development, underlie infants’ basic abilities to interpret and evaluate entities within the moral world; such abilities support claims for a core moral domain.

1. What babies know

In What Babies Know (2022), legendary cognitive scientist Elizabeth Spelke reviews decades of evidence supporting the claim that humans possess “core knowledge”: Innate systems of abstract knowledge that are evolutionarily ancient, early-emerging, and invariant over development; within these systems knowledge operates automatically, unconsciously, and independently of belief. Spelke provides a detailed account of the evidence for and against six systems, or domains, of core knowledge: Objects, number, places, forms, agents, and social beings. Each system contains skeletal versions of particular concepts (e.g., how objects behave), allowing inexperienced human and nonhuman animals to reason productively about the entities within it.

Crucial to Spelke's analysis of which concepts are and are not part of core knowledge is the claim that each domain competes with the others for attention, resulting in an initial failure of domains to communicate with each other. This failure precludes thinking about any concept that involves entities from more than one domain, thereby limiting the range of concepts that can be core to the human mind. Spelke argues that these limitations are overcome with the emergence of language around 10 months; language facilitates communication between domains and allows children to begin constructing myriad novel, non-core concepts.

2. What babies don't know: Social goals

After reviewing concepts she believes young infants possess, Spelke devotes significant discussion to one she believes they lack: social goal. Specifically, although two of Spelke's six core domains are for reasoning about the social world (the agent and social being systems), Spelke holds that neither is capable of social goal understanding by itself. On the one hand, the core agent system solely considers agents’ physical and instrumental goals; it does not consider goals underlying agent-directed actions (e.g., social looking, communication), nor instrumental actions undertaken for social reasons (e.g., cooperating, helping, hindering, other prosocial and antisocial acts). On the other hand, the core social being system solely considers engagements between social entities and between those entities and infants themselves; it does not consider the mental states driving those engagements. Given that understanding social goals requires thinking about social beings as agents whose mental states refer to other agents, it requires communication between the agent and social being systems. Because this communication is impossible prior to language, the concept social goal cannot be part of core knowledge.

Although the observation that core knowledge is limited and the provision of an explanation for how its limitations are eventually overcome are crucial to a proper account of the origins of human cognition, it is curious why Spelke spends so much time arguing that core knowledge lacks social goal understanding. Presumably, this is because she wishes to argue against a potential additional domain of core knowledge that has recently attracted attention: core morality. Indeed, a growing number of cognitive scientists have recently argued that human moral systems are supported by evolved, domain-specific mechanisms for thinking about the moral world (e.g., Baumard, André, & Sperber, 2013; Buyukozer Dawkins, Sloane, & Baillargeon, 2019; Hamlin, 2013a; Krebs, 2008; Macnamara, 1991; Mikhail, 2011; Premack, 2007; Woo, Tan, & Hamlin, 2022; Wynn & Bloom, 2014). Among other things, these mechanisms might allow inexperienced humans to identify morally relevant (inter)actions, evaluate those actions and agents who engage in them as positive or negative, and generate expectations for further actions the agents might perform. All such capacities would, at minimum, require an understanding of social goals.

Although not exclusively, much recent argumentation for core morality cites evidence that preverbal infants appear to positively and negatively evaluate agents based on their prosocial and antisocial acts (for review, see, Margoni & Surian, 2018; Woo et al., 2022), and/or possess expectations for the prosocial versus antisocial acts that individuals are likely to perform in distinct contexts (see Buyukozer Dawkins et al., 2019). Spelke acknowledges that this evidence could suggest that young infants understand social goals, but argues that it actually does not, for two reasons. First, much of the evidence involves infants older than 10 months. At 10 months, the core agent and social being systems can communicate, meaning a concept of social goals can be constructed rather than “core.” Second, evidence with infants <10 months has a viable alternative explanation that does not implicate social goal understanding. Specifically, young infants’ preferences for helpers over hinderers (e.g., Hamlin & Wynn, 2011; Hamlin, Wynn, & Bloom, 2007) need not reflect their understanding of (pro- and/or anti-) social goals, as those preferences may stem from mere sensitivity to social beings in states of engagement (or not) with other social beings: Social engagement is handled by the social being system alone.

Specifically, Spelke points out that prototypical helpful/unhelpful acts shown to infants can also be described as imitative/not imitative: In order to help, the helper generally reproduces the actions of a needy protagonist, whereas the hinderer produces opposing actions (see, e.g., Powell & Spelke, 2018). Under Spelke's core social being system, imitation is a powerful cue that one social being is engaged with another, but certainly lacks moral content (see also Powell, 2022). Thus, Spelke holds that much of the evidence used to argue for core moral capacities has no moral content after all.

In what follows, I review evidence that although infants appear sensitive to cues to social engagement like imitation, they can and do reason about social goals prior to 10 months. Indeed, consistent with claims for core morality, preverbal infants may be particularly sensitive to social goals with moral content, including helping/hindering, protection/harm, and fairness/unfairness. Because of space constraints I can only touch on the relevant evidence below; interested readers can find more detailed discussion elsewhere (Hamlin, 2023; Woo, Tan, Yuen, & Hamlin, 2023).

3. Evidence infants’ preference for helpers cannot be explained by imitation

Spelke argues that young infants’ preference for helpers reflects sensitivity to imitation rather than to prosocial/antisocial goals. Indeed, Spelke and her former student Lindsay Powell have demonstrated that infants prefer imitators over non-imitators within scenarios purported to demonstrate infants’ preferences for helpers over hinderers (Powell & Spelke, 2018). Although Powell (2022) argues that infants’ imitator preference itself reflects social goal understanding (e.g., that one agent has adopted another's “utility,” or goal, as its own), Spelke argues that infants could instead prefer imitators without representing utility adoption/goals at all, by inferring that imitators are engaged with their targets.

Inconsistent with Spelke's analysis, several studies now suggest that infants’ preferences focus on helping rather than imitating. For instance, infants’ preferences rely on their understanding of a needy protagonist's goals. Hamlin (2015) manipulated whether or not 6–10-month-olds could recognize the goal of trying but failing to climb a hill, by showing some infants the protagonist's eyes pointing toward the hilltop (suggesting a goal to reach the top) and others the protagonist's eyes pointing away from the hilltop (rendering its goal ambiguous). Critically, in both conditions one character imitated the protagonist (pushed it up), and another character did not imitate the protagonist (pushed it down). Critically, only those infants who saw the protagonist looking toward the hilltop, demonstrating a clear unfulfilled goal, preferred the pusher-upper (here, a helper) to the pusher-downer (a hinderer). Similarly, Tan and Hamlin (2022) showed that infants’ own looking toward the hilltop during the protagonist's failed attempts, arguably indicating goal inference (see Elsner & Adam, 2020), predicted their individual preference for the agent who pushed it to the top: Only those infants who ever looked to the top of the hill preferred the “helper.”

Other studies more directly compare helpers and imitators. For instance, in Hamlin et al. (2007) and Chae and Song (2018), 6- and 10-month-olds were asked to choose between an agent who pushed a needy protagonist uphill, moving like the protagonist and causing it to achieve its goal, and an agent who moved up the hill in exactly the same way, but independently from the protagonist. Here, infants chose between an imitative helper and a mere imitator, and consistently selected the helper. In a study from Spelke's own laboratory led by her former student Brandon Woo (Woo & Spelke, 2023), 8-month-olds were led to infer that an agent's goal was one of the two possible options, either to open a specific box or to obtain a specific toy. Subsequently, one agent facilitated the goal they inferred the agent to have, whereas the other agent facilitated the other goal; critically for the present purposes, in one condition the helpful agent was less imitative. Here again, infants’ choices suggested they consistently preferred helpers, but not imitators.

In each of the above studies, one character's actions always matched the protagonist's more closely than the other's. However, infants only chose those imitators whose actions were also helpful: They preferred helpful imitators over mere imitators, and failed to distinguish differentially imitative characters who were not differentially helpful. Thus, Spelke's claim that young infants’ helper preferences can be reduced to preferences for imitators seems unlikely.

4. Evidence young infants can represent (pro- and anti-) social goals

Although the above work suggests that infants are more sensitive to helping than to imitating, it need not indicate that infants represent that helpers possess the prosocial goal to facilitate another's goal or that hinderers possess the antisocial goal to prevent a goal. Indeed, perhaps young infants merely represent whether or not one agent causes another to achieve its goal (which imitators do not do). Of course, adults’ moral concepts privilege others’ prosocial and antisocial intentions (e.g., Cushman, 2008; Malle, 1999): Do young infants also consider social intentions?

Multiple studies now suggest that infants represent and evaluate prosocial and antisocial intentions before 10 months (see also Hamlin, Ullman, Tenenbaum, Goodman, & Baker, 2013; Kanakogi et al., 2017; Strid & Meristo, 2020; Woo, Steckler, Le, & Hamlin, 2017). First, Hamlin (2013b) demonstrated that 8-month-olds privileged intentions over outcomes in their preferences for pro- and antisocial others; for instance, preferring an agent who tried but failed to help a protagonist achieve its unfulfilled goal over an agent who tried but failed to hinder the protagonist. Second, Woo and Spelke (2022) showed that, remarkably, 8-month-olds preferred an agent who believed it was helping, even though it was not, over an agent who believed it was not helping, even though it was. Finally, Geraci and Surian (2023) and Geraci, Simion, and Surian (2022) demonstrated that 4- and 9-month-olds preferred an agent who tried but failed to distribute resources equally between two recipients over one who tried but failed to distribute resources unequally. Because no resources were ever actually given out, infants’ choices must have been based on intent; further control conditions suggest that it was not that infants simply like agents appearing to have more social partners (for related controls with acts of protection, see Kanakogi et al., 2017). These papers suggest that even young infants understand prosocial and/or antisocial intentions within two morally relevant domains, at times at or before they appear to be able to recognize intention in nonmoral contexts (for discussion, see Woo et al., 2023).

5. Conclusion

In sum, in contrast to Spelke's claims, young infants appear to possess the concept social goal. Although this may seem like a rather insignificant topic of debate, an inability to understand social goals early in development would, in turn, render recent arguments that humans possess core moral capacities (e.g., Baumard et al., 2013; Buyukozer Dawkins et al., 2019; Hamlin, 2013a; Krebs, 2008; Macnamara, 1991; Mikhail, 2011; Premack, 2007; Woo et al., 2022; Wynn & Bloom, 2014) moot, given that many if not most moral concepts are fundamentally rooted in notions of social goals. Of course, there is more to moral concepts than social goals, and any effective claim to core morality will include much more data than reviewed here, including from animals and diverse adult humans (see, e.g., Anderson et al., 2017; Anderson, Kuroshima, Takimoto, & Fujita, 2013a; Anderson, Takimoto, Kuroshima, & Fujita, 2013b; Brosnan, 2023; Cosmides, Guzmán, & Tooby, 2018; Cosmides & Tooby, 1992; Curry, Mullins, & Whitehouse, 2019; Darden, James, Cave, Brask, & Croft, 2020; Isik, Koldewyn, Beeler, & Kanwisher, 2017; Mikhail, 2011). Spelke's What Babies Know (2022) provides an enviable model of what effective arguments for core moral knowledge – and indeed any account of the origins of knowledge more generally – must look like.