Abstract
In perception, action, and decision-making, participants exhibit automatic, apparently suboptimal sequential effects (SQE): they respond more rapidly to a stimulus if it reinforces local patterns in stimulus history. We investigate whether SQE are driven by mechanisms critical for adapting to a changing (i.e. dynamic) world. Mere presence of SQE suggests an emphasis on the recent over the distant past in predicting upcoming events, a rational response in dynamic environments. Previously reported biases in SQE toward repetition versus alternation may be due to prior beliefs about what environments look like after change. Subjects completed a modified version of "Whack-a-Mole". Sesame Street’s Elmo popped up either to the right or left of fixation. Participants were instructed to press a spatially congruent button as fast as possible but not at the expense of accuracy. In Session 1 and 3, repetitions and alternations in Elmo’s location were equally likely (p(repetition) = 0.5). In Session 2, p(repetition) was repeatedly sampled from a Beta-distribution with B(6,12) for "alternation-training" and B(12,6) for "repetition-training". Resampling occurred at p(resample) = 0.18 and was signaled to the subject to allow them to learn the environment’s "change-rate". Participants’ reaction times (RT) showed SQE in all sessions. Prior to training, participants were biased towards alternations. After training, repetition-trained participants were biased towards repetitions while alternation-trained participants remained biased towards alternations. Modeling shows that this training-induced bias in SQE can only be due to prior beliefs about the probability to observe a repetition versus alternation. Importantly, participants’ belief in a dynamic environment was sufficient to produce these biased SQE. This belief may act as a self-fulfilling prophecy: if probability estimates are derived with an emphasis on the recent past, then stable environments seem dynamic. Pervasive SQE on perceptual, motor, and decision tasks should surprise no one.
Meeting abstract presented at VSS 2013