I have two strands of research related to understanding how we learn to categorize and make inductions.
The first strand examines the popular notion that category learning can be explained through the operation of distinct ‘implicit’ and ‘explicit’ systems. In work initiated with David Shanks and David Lagnado we have investigated probabilistic categorization tasks in an attempt to characterize the evidence for the operation of such separate systems. In a similar vein, work with John Dunn and Mike Kalish, goes beyond the ‘systems’ debate by utilizing a technique known as state-trace analysis to uncover the inherent dimensionality of category learning (see
Newell et al., 2011;
Newell & Dunn, 2008;
Kalish et al., 2017). More recent projects in this line have included an exhortation to use more sophisticated methods to model the ‘strategies’ people employ in learning categories (
Donkin et al., 2015) and revisiting a claim for “one of the strongest dissociations” in the literature (
Le Pelley, Newell & Nosofsky, 2019). The over-arching approach has been to consider the viability of single-system cognitive architectures for explaining category learning phenomena. (See also
Newell & Shanks, 2014 and
Newell & Le Pelley, 2018 for other detailed examinations of the value of dichotomizing cognitive processing).
The second strand (with Brett Hayes and Oren Griffiths) examines induction, and specifically the mechanisms underlying induction on the basis of uncertain categorization judgments. This so-called ‘as-if’ reasoning (reasoning as-if a given judgment is true when there is uncertainty associated with it) appears to be prevalent. We have documented some of the factors which affect the tendency to reason in this way (e.g.
Griffiths et al., 2012;
Newell et al, 2011;
Hayes & Newell, 2009).
A new project, also with Brett Hayes, starting in 2022 will examine the reasons why people tend to fall into so-called “learning-traps” – suboptimal methods for exploration which can lead to impaired learning of category structures.