Category Learning as a Use Case for Anticipating Individual Human Decision Making by Intelligent Systems

Abstract

Anticipating human behavior requires a model of the rationale how humans acquire knowledge while solving a problem. The rational aspects of decision making needs to be taken into consideration for improving computational models that currently fail to fully explain behavioral data in rule-based category learning. Compared to reinforcement learning models that assume gradual learning, cognitive modelling allows to implement selection rules and instance based learning for decision making to allow more flexible behavior. Here we use ACT-R to model behavioral data of auditory category learning. By systematically changing the probabilities of rule selection, capturing individual preferences of auditory features, we first improve the original model regarding the average learning curves of subjects. The aim is then to generate a version of the model that explains learning performance of individual subjects. Neuroimaging data will allow to test the predictions of the model by analyzing the dynamics of activation of brain areas linked to the model processes.