Attention in Associative Learning

Indiana University, Bloomington IN

Investigators

Abstract

Medical clinicians learn to associate certain symptoms with particular diseases. Seafarers learn to associate certain patterns of wind, wave and animal activity with particular future weather conditions. Stock traders learn to associate certain levels of value, earnings and market activity with increases or decreases in price. Law enforcement agents learn to associate features of people and the environment with particular crimes. Consumers learn to associate various products with different feelings. These examples illustrate that associative learning underlies much of human behavior, with potentially significant consequences for decisions in the real world. People are adept at learning many arbitrary associations very quickly, without entirely forgetting previously learned associations. Despite this proficiency, people also show numerous learned behaviors that are suboptimal or irrational from a normative statistical perspective. The goal of the present research is to investigate one likely cause of these irrational behaviors, namely, selective attention. When people are learning to associate a complex array of information with an outcome, a subset of the information can be selectively attended to. By attending to distinctive aspects of the situation, the person can learn quickly without confusing it with previous knowledge. Selective attention, because it excludes some aspects of the situation, also implies incomplete knowledge. This side-effect of selective attention can then lead to numerous suboptimal behaviors in later circumstances. The research involves (a) numerous experiments in human learning and (b) computer simulation of mathematical models of attentional shifts. A primary goal of the proposed research is to explain a wide spectrum of seemingly disparate and irrational effects in associative learning as the natural consequence of attention shifting. The attention shifting itself is actually an efficacious and adaptive response to the goal of learning quickly. The mathematical model reflects this goal by shifting attention to whatever features reduce predictive error most rapidly. The model mimics human learning in detail, and makes new predictions to be tested by the research. The model also accounts for several previously unexplained or disparately explained phenomena. The research will improve our understanding of the basic principles of associative learning. Future research can build on this understanding to develop interventions and applications in real-world environments.