A recent surge of research in cognitive developmental psychology examines whether human learners, from infancy through adulthood, reason in ways consistent with Bayesian inference. However, when exploring this question an important first step is to identify the available inference mechanisms and computational machinery that might allow infants and young children to make inductive inferences.
A number of recent studies have asked if infants may be "intuitive statisticians," making inferences about the relationship between samples and populations in both looking-time and choice tasks. In this dissertation, I present three sets of empirical experiments in support of infants' ability to make probabilistic inferences. The first empirical chapter examines the age at which infants begin making probabilistic inferences. I present an experiment suggesting that 6- but not 4-month-old infants can make generalizations about the likely composition of a large population after observing the contents of a small sample drawn randomly from that population. The second empirical chapter presents seven experiments that compare infants' and adults' abilities to make more complex probabilistic inferences. These experiments suggest that infants can integrate both probabilistic and deterministic physical constraints in probabilistic inference, and that in some cases, they show greater competence in doing so than adults. The final chapter exploring infant probabilistic inference presents evidence from three experiments with 10- to 12-month-old infants. In these experiments, results suggest that infants can use single-event probability computations to make predictions about where to direct their search to locate a desired object. These experiments also suggest that infants use proportional reasoning and not a simple heuristic based on comparisons of absolute quantity to make probabilistic inferences.
In the fourth empirical chapter I present data from a series of four experiments conducted with preschool-aged children. My colleagues and I use a causal learning task that is broadly similar in structure to the series of infant experiments reported in Chapter 4: Children are asked to use single-event probability computations to make causal inferences. The key motivation of this chapter is to reconcile divergent findings in the literature suggesting that, on the one hand, children reason in ways that are consistent with rational inference and, on the other hand, children tend to produce responses that are quite variable in nature. This chapter outlines a proposal termed the "Sampling Hypothesis", which suggests that the variability in young children's responses may be part of a rational strategy for inductive inference. In the reported experiments, we find evidence to suggest that children sample responses from the distribution of possible hypotheses that explain the observed data, weighting the different hypotheses according to their probability. This chapter provides an illustration of one of the ways in which a learner's ability to engage in probabilistic inference, which comes online early in infancy, can provide the foundation for more complex inductive inferences later in development.
In the final chapter I discuss the implications of this work, point to a number of remaining open questions, and consider some future directions for this line of research. I conclude with the suggestion that infants and young children are much more sophisticated at making probabilistic inferences than was previously thought. The competences demonstrated by infants and young children in the reported experiments appear to draw on an intuitive probability notion that is early emerging and unavailable for conscious reflection. Moreover, I suggest that young learners are capable of making rapid inductive inferences by capitalizing on their ability to compute probabilities, in order to acquire knowledge in a variety of domains, including causal reasoning.