Abstract
In motor tasks with explicit rewards and penalties, humans choose movement strategies that maximize expected gain (Trommershauser et al., 2003, JOSA, 20, 1419). Here, we investigate an analogous perceptual task. On each trial, subjects saw a briefly presented (1 sec) texture followed by a test display. The texture comprised 32 line segments (.7 deg in length) in a 4.6 deg diam circular window. Segment orientations were drawn from a von Mises distribution whose mean and standard deviation were varied from trial to trial. The mean was uniformly distributed and the standard deviation was 2.6, 8.0 or 27 deg. Subjects then saw a test display containing two green arcs (the reward region) at opposite sides of the circular window and two red arcs (the penalty region) also at opposite sides. All arcs were 22 deg and the penalty region either overlapped half of the reward region or abutted it. Subjects rotated this display using key presses until satisfied. If the mean texture orientation fell within the reward region the subject won 100 points. If the mean orientation fell within the penalty region, the subject lost a fixed number of points (0, 100 or 500). There were 9 blocks of 80 trials, one for each combination of orientation standard deviation (noise) and penalty. There were 5 subjects. We compared each subject's performance across conditions to a decision strategy that maximizes expected gain (MEGaVis: Maximize Expected Gain in a Visual task). MEGaVis takes into account the magnitudes of penalties, the relative placement of the reward and penalty regions, and the subject's uncertainty in estimating mean orientation. In most conditions, subjects compensated for their variability and the reward/penalty structure in making settings. Their performance was close to optimal. However, in the most difficult (high noise/high penalty) conditions, observers were consistently sub-optimal: They made settings insufficiently far away from the penalty region.
NIH EY08266, HFSP RG0109/1999-B, DFG Emmy-Noether-Programm, EPSRC GR/R57157/01