Bayesian vs psychological approaches to cognitive development

Endress, A. 1, 2

1 Universitat Pompeu Fabra, Barcelona
2 City University, London

In recent years, Bayesian learning models have been applied to an increasing variety of domains, and have been proposed to account for many important phenomena in cognitive development. However, such models are often constructed to fit specific cognitive phenomena, but, with few exceptions, little attention is given to novel predictions, or to the question of whether the principles they implement are psychologically plausible. Here, I take two important case studies that have recently been proposed to conform to Bayesian principles - rule learning and statistical learning. I spell out the central assumptions of the Bayesian models in non-mathematical terms, and show that they are implausible. Further, I derive novel predictions of these models, and show that experimental data does not adhere to them. I propose alternative accounts in terms of basic psychological mechanisms, and show that they fit the data better and are easier to falsify.