Bayesian models of child development

Abstract

Abstract[Bayesian models have been applied to many areas of cognitive science including vision, language, and motor learning. We discuss the implications of this framework for cognitive development. We first present a brief introduction to the Bayesian framework. Bayesian models make assumptions about representation explicit, and provide a detailed account of learning. Furthermore, they can provide an account of developmental transitions and other phenomena in development, such as curiosity and exploration. Drawing on recent work bridging empirical developmental data and modelling, we show that these features of the Bayesian approach provide solutions to problems that elude traditional accounts of learning and raise new areas of investigation.] Somehow every child solves the baffling and intractable philosophical problem of induction. Children take the plethora of ambiguous information coming in through their senses and turn it into meaningful, abstract, structured representations. Despite centuries of philosophical thought and empirical study, we still don't fully understand how this is possible. How can children's mental representations both support wide-ranging new inferences and change in the light of experience?This fundamental problem has led to a tension in cognitive science. How can we learn abstract representations from concrete data? The classic nativist response to this question is to say that learning is an illusion. We come equipped with abstract representations. The alternative empiricist response is to say that abstraction is an illusion. Knowledge is just a collection of associations derived from the statistics of our environment.Nether of these options seems completely right to most empirically-minded developmental psychologists. That's because we see evidence for both abstract representation and sophisticated learning even in infancy and early childhood. Piaget proposed "constructivism" as a way to bridge the nativist/empiricist divide. Piagetians appealed to an ever-enriching developmental process that employed mechanisms of accommodation and assimilation.The "theory theory" [1][2][3][4] , which was a theoretical offspring of constructivism, was also intended to be an alternative to the nativist and empiricist extremes. It proposed that children's beliefs are rich, structured and abstract but defeasible representations. Even if children are equipped with innate (or rapidly developed) "starting-state" theories, those theories can always be changed in the light of new evidence.However, both the Piagetian and theory theory accounts essentially restate the problem. They fail to provide a precise account that describes the representational details of children's beliefs and specifies the mechanisms that support learning.In the last ten years or so, however, probabilistic models have begun to promise a more precise solution to the nativist-empiricist tension. These approaches are based on ideas from the philosophy of science, machine learning, and artificial intelligence. The probabilistic modelling ...

Naive psychology depends on naive physics

“…If our early-emerging knowledge of physics and psychology is best described as intuitive theories (Gopnik & Wellman, 1992), then they should be subject to revision (Bonawitz & Ullman, 2024;Carey & Spelke, 1996;Gerstenberg & Tenenbaum, 2017;Ullman, 2005). Recent work shows that adults' expectations about agents are at least temporarily and locally defeasible with evidence (R. Liu & Xu, 2023), but it remains unclear whether these are genuine cases of belief revision, and how durable and general they are.…”