publications

Models are often evaluated when their behavior is at its closest to a single, sometimes averaged, set of empirical results, but this evaluation neglects the fact that both model and human behavior can be heterogeneous. Here, we develop a measure, g-distance, which considers model adequacy as the extent to which models exhibit a similar range of behaviors to the humans they model. We define g as the combination of two easily interpretable dimensions of model adequacy: accommodation and excess flexibility. We apply this measure to five models of an irrational learning effect, the inverse base-rate effect (IBRE). g-distance identifies two models, a neural network with rapid attentional shifts (NNRAS) and a dissimilarity-similarity generalized context model (DGCM18), that outperform the previously most supported model (EXIT). We show that this conclusion holds for a wide range of beliefs about the relative importance of excess flexibility and accommodation. We further show that a pre-existing metric, the Bayesian Information Criterion (BIC), misidentifies a known-poor model of the IBRE as the most adequate model. Along the way, we discover that some of the models accommodate human behavior in ways that seem unintuitive from an informal understanding of their operation, thus underlining the importance of formal expression of theories. We discuss the implications of our findings for model evaluation generally, and for models of the inverse base-rate effect in particular, and end by suggesting future avenues of research in computational modeling.

The inverse base-rate effect (IBRE) is an irrational phenomenon in predictive learning. It occurs when people try to generalize what they have experienced to novel and ambiguous events. This irrational generalization manifests as a preference for rare, unlikely outcomes in the face of ambiguity. At least two formal mathematical models of this irrational preference (EXIT, NNRAS) lead to a counter-intuitive prediction: the effect reduces under concurrent load. We tested this prediction across two experiments (N1 = 72, M1age = 20.12; N2 = 160, M2age = 20.88). We confirm the prediction, but only when participants were under an obvious time constraint. This empirical confirmation is as surprising as the prediction itself—irrationality reduces under increased task demands. Further, our data are more consistent with the NNRAS model than with EXIT, the most prominent model of the IBRE to date.

The inverse base-rate effect is a robust irrational bias that arises when people face ambiguity. The most prominent theories of this irrational bias depend on prediction error. In this study, we gradually removed elements of a predictive learning design to test the extent to which error-driven processes underlie this bias. In our first experiment, we removed explicit feedback by implementing the inverse base-rate effect in an observational learning procedure. In our second study, we further removed any causal relationship between stimulus features and category labels by moving towards an unsupervised learning procedure. This removed any information participants could use to identify category labels. In both experiments, the inverse base-rate effect persisted and remained robust. This outcome suggests that this irrational bias is independent of supervised learning procedures. We propose that any theories and models of the inverse base-rate effect must manage information encoding and connection updates without explicit prediction error. We end by proposing two clear paths for future investigations.

There is a growing interest in alternative explanations to thedual-system account of how people learn category structuresvarying in their optimal decision bounds (unidimensional andinformation-integration structures). Recognition memory per-formance and hippocampal activation patterns in these tasksare two interesting findings, which have not been formally ex-plained. Here, we carry out a formal simulation with SUS-TAIN (Love, Medin, & Gureckis, 2004), an adaptive modelof category learning, which had great success in accountingfor recognition memory performance and fMRI activity pat-terns. We show, for the first time, that a formal single-systemmodel of category learning can accommodate recognition per-formance after learning and is consistent with fMRI data ob-tained while participants learned these structures

It is generally assumed that the Rescorla and Wagner (1972) model adequately accommodates the full results of simple cue competition experiments in humans (e.g. Dickinson et al., 1984), while the Bush and Mosteller (1951) model cannot. We present simulations that demonstrate this assumption is wrong in at least some circumstances. The Rescorla-Wagner model, as usually applied, fits the full results of a simple forward cue-competition experiment no better than the Bush-Mosteller model. Additionally, we present a novel finding, where letting the associative strength of all cues start at an intermediate value (rather than zero), allows this modified model to provide a better account of the experimental data than the (equivalently modified) Bush-Mosteller model. This modification also allows the Rescorla-Wagner model to account for a redundancy effect experiment (Uengoer et al., 2013); something that the unmodified model is not able to do. Furthermore, the modified Rescorla-Wagner model can accommodate the effect of varying the proportion of trials on which the outcome occurs (i.e. the base rate) on the redundancy effect (Jones et al., 2019). Interestingly, the initial associative strength of cues varies in line with the outcome base rate. We propose that this modification provides a simple way of mathematically representing uncertainty about the causal status of novel cues within the confines of the Rescorla-Wagner model. The theoretical implications of this modification are discussed. We also briefly introduce free and open resources to support formal modelling in associative learning.

The aim was to quantify ego depletion and measure its effect on inhibitory control. Adults (N = 523) received the letter “e” cancellation ego depletion task and were subsequently tested on Stroop task performance. Difficulty of the cancellation task was systematically manipulated by modifying the text from semantically meaningful to non-meaningful sentences and words (Experiment 1) and by increasing ego depletion rule complexity (Experiment 2). Participants’ performance was affected by both text and rule manipulations. There was no relation between ego depletion task performance and subsequent Stroop performance. Thus, irrespective of the difficulty of the ego depletion task, Stroop performance was unaffected. The widely used cancellation task may not be a suitable inducer of ego depletion if ego depletion is considered as a lack of inhibitory control.