Where is the baby in core knowledge?

Abstract

What we know about what babies know – as represented by the core knowledge proposal – is perhaps missing a place for the baby itself. By studying the baby as an actor rather than an observer, we can better understand the origins of human intelligence as an interface between perception and action, and how humans think and learn about themselves in a complex world.

What Babies Know (Spelke, 2022) is a masterful synthesis of what we have discovered about the origins of human intelligence. The use of looking time methods has opened a window into the minds of incredibly young infants, enabling science to weigh in on questions that philosophers have debated for centuries: The content of the human mind and how it develops in the earliest phase of life. In this book, Spelke proposes core knowledge as a way to explain both the remarkable systematicity and striking diversity in human behavior; humans are endowed with a system of knowledge in key domains (systematicity) that provides the foundational concepts that support learning from experience (systematicity and diversity).

The comprehensive review of the empirical evidence for this proposal, however, ironically highlights a glaring gap: Where is the baby (i.e., the self)? As much as it tells a rich, compelling story about what infants know about the external world, it also reveals how the field of infant cognition has focused primarily on infants' representations of their surroundings as onlookers. By characterizing the baby as a learner building a model of the world – and by constraining the scope of “the world” to what is external to the learner – researchers have often overlooked the baby as a key constituent of that world; infants occupy a coordinate in space, are subject to physical principles, and can exert causal power on objects and agents (including themselves). How do infants represent themselves in the physical and the social world, not just as observers, but as actors who navigate in space, manipulate objects, and interact with others? How do infants build a model of the self, both as a physical constituent of the external world and as a mental agent capable of representing their internal world (e.g., affective and mental states)? We cannot answer these questions yet because we still lack a scientific theory of how a baby might represent, model, and learn about the self.

It may be tempting to offload these questions onto the realm of learning from experience, treating them as a topic for How Children Learn rather than What Babies Know. However, if the core knowledge proposal is meant to be a scientific account of the origins of human intelligence – an evolved system of knowledge that is adaptive for survival and learning – it is almost impossible to imagine how this knowledge could be so rich, extensive, and coherent in content yet so passive in nature, detached from the organism itself.

The idea that humans know and learn about the self is far from new. Early philosophers have theorized about the role of introspection (Descartes), first-person sensory experience (Hume), and attribution of self-concept (Locke). William James' distinction between the self as “Me” (the empirical self constructed through interactions with the world) versus “I” (the subjective self capable of representing “Me”; James, 1890; see also Gergely, 2002; Neisser, 1988) and Piaget's emphasis on the self and agentive experience (Piaget, 1952, 1954) have provided grounds for the initial empirical findings that hint at a nascent sense of self-awareness in infancy, especially regarding their own bodies and actions. By 3 months of age, infants not only recognize their own movements (Rochat & Striano, 2000) but also grasp the contingency between their own actions and changes in the physical world (e.g., kicking and the movement of a crib mobile; Rovee & Rovee, 1969) and in their interactions with others (Bigelow & Rochat, 2006; Gergely & Watson, 1999; Nadel, Carchon, Kervella, Marcelli, & Reserbat-Plantey, 1999). These studies suggest that infants at this age can already represent themselves as both physical and social entities that exert influence on the external world. Although questions remain about the richness of such understanding, these findings nonetheless highlight the need to characterize both the nature and the content of the knowledge that could give rise to these behaviors, and the relationship between a baby's representation of the self and their active action planning.

The key functions of the human mind – perception and cognition – are often conceptualized in the context of action. The limited behavioral repertoire of young infants may perhaps explain why infants have often been characterized as observers looking out the window. While limited, however, infants' behaviors are systematic; researchers can use infants' looks to tap into their minds precisely because these behaviors reflect what infants perceive and think. Within a few months of life, infants also begin to reach, vocalize, and make facial expressions in ways that suggest a mapping between themselves and their physical and social surroundings. It is possible that the absence of these behaviors in very young infants reflects little more than their inability to execute motor actions; prereaching infants may want to approach or attain an object but are limited in their capacity to do so. If core knowledge is meant to be adaptive for survival and learning, then it seems reasonable to expect this system of knowledge to serve as the interface between perception and action. Even if this knowledge may not yet manifest as explicit, interpretable actions, it should nonetheless allow infants to use their own experience to learn about themselves and the external world. This critical link between perception and action, which must incorporate at least a crude representation of the self, is curiously absent from the core knowledge proposal.

Spelke's book, in fact, offers many examples of studies where the baby as a participant is more than just an onlooker. Infants visually recognize objects based on the motion patterns they themselves produced via haptic exploration (Streri & Spelke, 1988) and preferentially explore objects that violated their expectations to test for the specific violations they had observed (Stahl & Feigenson, 2015); toddlers use their (shockingly limited) representation of space to orient themselves (Hermer & Spelke, 1994); infants direct their gaze based on the contingency between their own behaviors and another agent (Johnson, Slaughter, & Carey, 1998), and the list goes on. Although some of these findings come from relatively older infants and toddlers, the systematic relationship between what they observe and what they do as a consequence nonetheless raises important questions about the developmental origins of the interface between perception, knowledge, and action.

One particularly striking example comes from studies using the “sticky mittens” training. The “Woodward effect” (Woodward, 1998), indicating an abstract understanding of others' object-directed goals, is observed only when infants view actions that they themselves can perform. Intriguingly, 3-month-olds (who cannot yet reach for objects) show this effect after a short training with a velcro-lined mitten that allows them to entrain objects without performing targeted reach and grasp (Sommerville, Woodward, & Needham, 2005; see also Skerry, Carey, & Spelke, 2013, for the effect of this training on expectation of action efficiency). Why does this training work? Spelke argues against the interpretation that the training allows prereaching infants to learn about goals; they may represent goals but do not yet understand what constitutes goal-directed actions, and the training highlights physical contact as a key feature of causally meaningful actions (Liu, Brooks, & Spelke, 2019). This compelling argument, however, still presupposes something that remains unaddressed by core knowledge; the baby understands the equivalence between its own hand movement and someone else's hand movement (i.e., self-other mapping). Although Chapter 8 (Core Social Cognition) begins to hint at a core system that incorporates the baby itself (Sect. 8.2: Infants' Sensitivity to Social Engagement), perhaps because of the difficulty of directly studying infants' internal, phenomenal experiences, many questions remain open about the nature of knowledge that supports infants' ability to formulate their own goals.

Stepping back, the discrepancy between what babies know and what babies do gives us an opportunity to reflect upon how the field has progressed in the last few decades. Moving on from behaviorism and away from the initial Piagetian emphasis on agentive experience, infant cognition research has been a remarkably fruitful enterprise filled with seminal discoveries on infants' ability to think and represent their surroundings. Although many of these studies have used methods that place infants in a relatively passive position as an observer, others have leveraged self-initiated behaviors such as manual search (e.g., Feigenson & Carey, 2003), crawling (e.g., Denison & Xu, 2010), object-based exploration (e.g., Gweon, Tenenbaum, & Schulz, 2010), and even socially oriented actions such as help-seeking (e.g., Goupil, Romand-Monnier, & Kouider, 2016; Gweon & Schulz, 2011). Although these behaviors primarily serve as “dependent measures” that are meant to inform researchers about the representations and inferential processes that reside in infants' minds, they are not to be taken for granted; they emerge only when babies know enough to understand what is going on, and are motivated enough to act.

More broadly, infants' understanding of the external world has remained a rather separate topic of scientific inquiry from infants' understanding of their own bodies and locomotor ability (e.g., Adolph & Hoch, 2019; Rochat & Striano, 2000) or the motivational drive that underlies their own goal-directed actions (e.g., Dweck, 2017). Yet, infants are clearly motivated to perform actions that exert a systematic influence on the external world (Rovee & Rovee, 1969), are puzzled when their actions fail and driven to figure out why (Gweon & Schulz, 2011), and actively seek information about their own motor abilities (Adolph & Hoch, 2019). Beyond the self as a physical agent, as noted in the book, infants begin to use their own social relationships to reason about potential social partners (Thomas, Saxe, & Spelke, 2022). By preschool years, children expect their parent, but not the experimenter's parent, to know private information about themselves (Chuey, Jara-Ettinger, & Gweon, 2023); they also deliberately seek information about what others think of them (e.g., Zhu, Dweck, & Gweon, 2023) and even try to manage these representations by communicating about themselves (e.g., Asaba & Gweon, 2022; Heyman, Compton, Amemiya, Ahn, & Shao, 2021). These findings suggest that an abstract, socially constructed self is already present by the preschool years, and likely develops earlier.

By imagining how core knowledge could incorporate aspects of the self, perhaps we could inch closer toward a unified theory that explains how humans come to build a model of the world that incorporates themselves. One possibility is that core knowledge in each of the key domains have a placeholder for “the self”; for instance, infants' representations of space might incorporate information about one's own position, orientation, and movement, as suggested by the discovery of place cells, grid cells, speed cells, and head-orientation cells that together may give rise to a sense of the physical self in space and time (Moser et al., 2014). The recent discovery of social place-cells (Danjo, Toyoizumi, & Fujisawa, 2018; Omer, Maimon, Las, & Ulanovsky, 2018) that represent the spatial locations of conspecifics with respect to the self also supports the idea that the core system for spatial navigation has carved out a “place” for the self (note: it also raises a question about a strong version of modularity). Another possibility is that there is a separate domain of core knowledge about the self, which might support the use of proprioceptive senses for haptic feedback and motor coordination. Additionally, core knowledge of agents (Ch. 7) and core social cognition (Ch. 8) might also provide particularly useful grounds for building a model of the self that represents the self as an agent in relation to other agents, and learns about its properties through self-guided exploration of objects and interactions with others. These possibilities are not mutually exclusive, and can offer ways in which core knowledge can make contact with infants' understanding of the self.

By synthesizing decades of work, Spelke's book already offers a rich foundation for new methods and approaches for exploring these possibilities. If core knowledge serves to make learning more tractable, understanding how babies represent themselves may provide key clues that can help explain what makes humans such powerful learners: Humans compete against themselves to improve in ways that are far more efficient than the most powerful machines today, deploy their own intelligence to problem-solve both for themselves and for others, and perhaps most importantly, appreciate themselves as entities that can bond, interact, and communicate with other beings.