Integration of unreliable visual information

This dissertation investigates how people deal with unreliable visual information. The first chapter focuses on how people integrate information from one source that is contaminated by information from a second source. We found that when viewing simple, uncontaminated stimuli, subjects did not employ robust estimators to discount potential outliers. On the other hand, when instructed to ignore contaminating information in complex-stimuli that are generated from a mixture of two sources, subjects discounted information that had a high probability of belonging to the contaminating source. This suggests that robust methods employed by the visual system are closely tied to mechanisms of perceptual segmentation. The second chapter looks at how people integrate sequential bits of unreliable information. We found that subjects were able to adapt their rule of integration based on the relative precisions of the seven sources of information. While subjects did not learn the optimal weights for the seven sources, they did eclipse the best performance one could possibly obtain from integrating just the best source or even the two best sources of information. So, while subjects did not manage to integrate information optimally, they did derive some benefit from paying attention to the individual precisions of the sources and giving more weight to more precise sources. The third chapter looks at how people gather unreliable information when there is a monetary accrual cost for the information. Subjects were tasked with gathering discrete bits of visual information to help localize a hidden target for a reward. Each additional bit of information reduced uncertainty as to the location of the target, but cost a fixed number of points to get the information. The subject's goal was to balance the costs and benefits of gathering information and try to maximize expected gain. We found that subjects systematically over-sampled and over-paid by gathering more information than needed to maximize expected gain. Taken together, these studies illuminate how people integrate and balance the costs of gathering unreliable information.