Abstract
Wiener (1958) showed that one could use noise to analyze the behavior of a black box, suggesting a way to study the brain. Ahumada and Lovell (1971) adapted the technique to psychophysics: white noise is added to a signal and the observer's detection responses are classified into hits, false alarms, misses, and correct rejections. To depict the representation used to detect the signal, a discrimination image is computed by subtracting two classes of stimuli: those in which observers detected the signal (hits plus false alarms) minus those in which observers did not perceive the signal (misses plus correct rejections). The discrimination image can depict a memory representation of a simple object. We went back to Wiener's original proposal to derive a purer approach to visualize object representations from noisy stimuli. We instructed subjects to detect a black “s’ letter on a white background, extending across the whole stimulus, embedded in white noise, and present on 50% of the trials. We classified the stimuli into those in which subjects perceived the letter, and those in which they did not. We then derived the discrimination image as explained earlier. As expected, it revealed the ‘s’ template subjects used to detect the signal. However, there was no signal in any of the stimuli, only white noise. During the experiment, subjects had “superstitious perceptions” of a letter that was never presented. The superstitious perceptions could only be driven top-down, from memory, not bottom-up, from a signal embedded in noise, because there was no signal. The depicted discrimination image is a purer representation, because the standard reverse correlation technique does not always tease apart the respective contributions of a top-down representation and a bottom-up noisy signal.