0 The Infant’s Problem

Piaget, core knowledge, and compositional worlds

Long before a child can explain the world, the child must make the world explainable. Sensation does not announce its own ontology. A moving patch of color does not say that it is one object, that it persists when hidden, that it can be grasped, or that another person is reaching for it. A stream of speech does not arrive with reliable spaces between words, labels for parts of speech, or a grammar. A smile does not carry a typed declaration of attention, affiliation, or intent. Nevertheless, children rapidly begin to organize these changing fragments into objects, actions, places, quantities, utterances, agents, and relationships.

This is the motivating problem of the book. An intelligent learner does not merely fit a response to a fixed task. It discovers a world in which many tasks become expressible. The discovered world is compositional when local pieces can be connected into larger wholes, when the result of a connection can be reused, and when incompatible connections can be detected and repaired.

The purpose of this chapter is empirical and conceptual. Piaget supplies a dynamic account of construction through assimilation and accommodation. Spelke supplies evidence that construction begins with a small collection of domain-specific systems of core knowledge rather than an undifferentiated blank slate [ Piaget , 1952 , Spelke , 2022 , 2023 ] . Modern machine learning supplies concrete systems against which these ideas can be tested. The categorical language begins in Chapter 1; here we first establish what that language must explain.

Guiding question.

How can a learner equipped with a small repertoire of organizing capacities turn fragmentary interaction into persistent structures that support prediction, action, correction, and transfer across an open-ended lifetime?