Towards a rational theory of human information acquisition
|University of California at San Diego , USA|
Probabilistic theories of optimal experimental design provide a compelling theoretical framework for information acquisition (Savage, 1954). Psychological experiments suggest that people tend to select pieces of information that are highly useful. Unfortunately, because different authors have used different theoretical models (utility functions, such as Bayesian diagnosticity or information gain) as benchmarks, it has been unclear whether behavior that is optimal with respect to one model would also be optimal if an alternate theoretical model were applied.
Nelson (2005) analyzed several tasks in the literature to identify the predictions of each of 6 theoretical models on each task (Skov & Sherman, 1986; Baron, Beattie & Hershey, 1988; Slowiaczek et al, 1992; Oaksford & Chater, 2003; McKenzie & Mikkelsen, in press). There was high agreement between models on what questions were most (and least) useful. This result supported the feasibility of a rational ( Anderson , 1990) theory of information acquisition, but left unclear what theoretical model best matched human intuitions. Nelson used computer optimization to find limiting cases of disagreement between several theoretical models, to design an experiment for human subjects. Experimental results strongly contradicted predictions of Bayesian diagnosticity. However, methodological constraints precluded definitive test of which of several remaining models (probability gain, impact, and information gain-KL distance) best approximates human intuitions about the value of information.
We describe a new experimental information acquisition task in which environmental probabilities, involving simulated plankton, are learned perceptually over many trials. This scenario shows potential to assess the relative descriptive plausibility of several models, both optimal and heuristic, of information acquisition; to identify preferences of individual subjects, rather than tendencies in groups of subjects; and to facilitate parallel eye movement and neuroimaging studies of the value of information.
Anderson, J. R. (1990). The adaptive character of thought. Hillsdale , NJ : Erlbaum.
McKenzie, C. R. M., & Mikkelsen, L. A. (in press). A Bayesian view of covariation assessment. Cognitive Psychology.
Nelson, J. D. (2005). Finding useful questions: on Bayesian diagnosticity, probability, impact and information gain. Psychological Review, 112(4), 979-999.
Oaksford, M., & Chater, N. (2003). Optimal data selection: Revision, review, and reevaluation. Psychonomic Bulletin & Review, 10, 289318.
Savage, L. J. (1954). The foundations of statistics. New York : Wiley.
Skov, R. B., & Sherman, S. J. (1986). Information-gathering processes: Diagnosticity, hypothesis-confirmatory strategies, and perceived hypothesis confirmation. Journal of Experimental Social Psychology, 22, 93121.
Slowiaczek, L. M., Klayman, J., Sherman, S. J., & Skov, R. B. (1992). Information selection and use in hypothesis testing: What is a good question, and what is a good answer? Memory & Cognition, 20, 392405.