Predicting Preferences: An Examination of Agent Learning

Agent decision making occurs when an individual acts as a purchasing agent for another. Effective agent decision making requires that the agent learn to predict the target's preferences. Two experimental studies demonstrate the impact of providing agents veridical feedback. The results further our understanding of interpersonal prediction and learning from experience. Agents who are given the opportunity to learn from their own successes and failures do not exhibit the false consensus effect, or projection, that has been demonstrated in previous research. Any facilitative effect of similarity in tastes on predictive accuracy disappears when feedback is provided. Information theory is used to establish the informational value of individual instances, as well as an overall distribution of feedback. The results of the studies reported in the present article indicate that agents spend significantly more time considering informative than uninformative feedback, which improves their predictive accuracy. Copyright 1996 by the University of Chicago.