Believing is seeing: expectations alter visual awareness

Expectations have been shown to be powerful modulators of pain [1] and emotion [2] in placebo studies. In such experiments, expectations are induced by instructions combined with manipulation of the sensory experience that is unknown to the subjects. After an expectation learning phase where a painful stimulation is surreptitiously lowered following placebo application, the placebo effectively reduces subjective pain intensity in a subsequent test phase [3]. The strength of this placebo effect is closely related to the induced expectation [4]. Here, we asked whether this powerful cognitive bias reflects a general property of sensory information processing and tested whether the contents of visual awareness could be altered by a placebo-like expectation manipulation. We found a dramatic effect of experimentally induced expectations on the perception of an ambiguous visual motion stimulus. This shows that expectations have a strong and general influence on our experience of the sensory input independently of its specific type and content. Ten healthy human observers participated in a behavioral experiment where they viewed a random-dot kinematogram (RDK) that is perceived as a cylinder rotating in depth despite the absence of depth cues (Figure 1) [5]. The rotation direction is ambiguous, resulting in bistable perception with spontaneous direction reversals every few seconds. Participants had to report perceptual reversals, and to indicate the level of perceptual confidence (high-confidence, lowconfidence or uncertain) by pressing different keys. During an ambiguous baseline run, average durations of perceived rightand leftward rotation did not differ (Figure 1A; t(9) = 0.79, p = 0.45, paired two-tailed t-test). In the subsequent expectation learning phase, participants viewed a different version of the RDK in which ambiguity was minimized through the provision of red-green three-dimensional depth cues [5]. The participants now wore redgreen glasses, however, making them unaware of the difference from the previous ambiguous stimulus. They were told they would be presented with the same ambiguous RDK as in the first run, and that just the use of red-green glasses would strongly bias rotation perception towards one direction. The depth cues were consistent with this dominant direction in 80% of total viewing time and reversed unpredictably and unbeknownst to the observers with a frequency based on the reversal rate during the ambiguous baseline run. In a second learning run, the red-green glasses were switched, which reversed the perceptual bias. As expected, perception was strongly biased by the depth cues, resulting in significantly longer average phase durations of rotation in the dominant direction (Figure 1 B,C; t(9) = 9.9, p < 0.0005). While the participants believed that this was just an effect of the redgreen glasses, it was actually due to the stimulus manipulation. During the subsequent test phase, participants again wore the red-green glasses and were told that the two following runs were identical to the preceding runs. While therefore again expecting a strong bias towards one rotation direction, they were now unknowingly presented with completely ambiguous stimuli lacking any depth cues, as in the baseline run. Strikingly, perception was still strongly biased towards the expected direction (Figure 1D,E; t(9) = 3.4, p < 0.008) which, in the absence of depth cues, was purely due to the expectation induced by the learning phase. This expectationinduced perceptual bias was stable across the entire test phase (see Figure S1 in the Supplemental data available on-line with this issue). In a final ambiguous baseline without red-green glasses, no difference was observed between durations of rightand leftward rotation (Figure 1F; t(9) = 0.13, p = 0.90). Did participants truly perceive what they reported during the test phase or were they just reporting what they expected? Such a response bias should have led to an increase in expected percepts at the expense of low-confidence or uncertain percepts. However, the proportions of low confidence and uncertain percepts did not differ from the baseline runs (t(9) = –0.5, p = 0.66 and t(9) = –0.9,