Abstract
After brief exposure to multiple object arrays observers seem able to use summary statistics to make generalized judgments of object attributes. Such ensemble perception is thought to aid visual cognition of complex scenes by overcoming tight capacity limitations through extraction of 'gist'. However, even when a scene is rich with perceptual consistencies, some stimuli within it may have greater motivational salience than others. Previous work from our lab has shown that value-associated stimuli are processed faster and maintained in WM better than neutral stimuli, suggesting that such stimuli may bias ensemble perception. To investigate we asked participants to adjust a response circle to match the average size of 12 heterogeneously sized circles viewed simultaneously for 200 ms (test array). In an initial value-learning phase, all circles had the same color and responses were rewarded with either a high or a low reward depending on array color. In a second test phase, test arrays comprised circles of three different colors and no rewards were forthcoming. On different trials, the smallest four or the largest four circles were presented in the previously high or low reward-associated color. In the first experiment, average size estimates were biased towards the mean size of circles with the high reward-associated color, especially when they comprised the largest circles. This value-biasing effect was not found for low reward-associated colors. To investigate whether this effect occurred late in processing, we repeated the experiment, this time inserting a pattern mask immediately after the test array in test phase trials only. In addition to value-biasing being eliminated, test phase (but not value learning phase) performance accuracy was significantly better than in the first experiment. Our results demonstrate that previously rewarding stimuli bias the extraction of summary statistics and we suggest that such biases result from slow recurrent processes.
Meeting abstract presented at VSS 2017