Abstract
Two prominent ideas in the study of decision-making have been that organisms behave near-optimally, and that they use simple heuristic rules. These principles might be operating in different types of tasks, but this possibility cannot be fully investigated without a direct, rigorous comparison within a single task. Such a comparison was lacking in most previous studies, because the optimal decision rule was simple, no simple suboptimal rules were considered, it was unclear what was optimal, or a simple rule could closely approximate the optimal rule. We used a visual search task in which the optimal decision rule is well-defined and complex, and makes qualitatively distinct predictions from many simple suboptimal rules. Set size was 4. Each search display contained one target and three distractors. The distractors shared the same orientation. The target orientation and the common distractor orientation were independently drawn from the same distribution. The observer reported the direction of tilt of the target. All simple rules tested fail to describe human behavior. The optimal rule accounts well for the data, and several complex suboptimal rules are indistinguishable from the optimal rule. In a second experiment, we completely withheld trial-to-trial feedback and found that the conclusions remained unchanged, suggesting that visual computation in this task is not only optimal but also relies on an implicit representation of uncertainty. In both experiments, we found evidence that the optimal model is close to the (unknown) true model: first, the better the trial-to-trial predictions of a suboptimal model agree with those of the optimal model, the better that suboptimal model fits; second, an estimate of the Kullback-Leibler divergence between the optimal model and the true model is not significantly different from zero. Beyond the specifics of the task, these novel "absolute goodness-of-fit" tests could likely benefit many areas of vision science.
Meeting abstract presented at VSS 2016