Bidirectional Influences of Information-Sampling and Concept Learning

Contemporary models of categorization typically tend to sidestep the problem of howinformation is initially encoded during decision-making. Instead, a focus of this work hasbeen to investigate how, through selective attention, stimulus representations are contortedsuch that behaviourally-relevant dimensions are accentuated (or “stretched”), andrepresentations of irrelevant dimensions are ignored (or “compressed”). In high-dimensionalreal-world environments, it is computationally infeasible to sample all availableinformation, and human decision-makers selectively sample information from sourcesexpected to provide relevant information. To address these and other shortcomings, wedevelop an active sampling model, Sampling Emergent Attention (SEA), whichsequentially and strategically samples information sources until the expected cost ofinformation exceeds the expected benefit. The model specifies the interplay of twocomponents, one involved in determining the expected utility of different informationsources and the other in representing knowledge and beliefs about the environment. Thesetwo components interact such that knowledge of the world guides information sampling,and what is sampled updates knowledge. Like human decision-makers, the model displaysstrategic sampling behaviour, such as terminating information search when sufficientinformation has been sampled and adaptively adjusting the search path in response topreviously sampled information. The model also shows human-like failure modes. Forexample, when information exploitation is prioritized over exploration, the bidirectionalinfluences between information-sampling and learning can lead to the development ofbeliefs that systematically differ from reality.