Abstract
Bayesian models describe precision (inverse variance) as a key determinant of perception. However, there is limited evidence on the behavioural effects of precision. The default assumption is that higher precision leads to greater surprise (or perceived change) from otherwise equivalent sensory changes. Four human experiments investigated the influence of precision on perceived salience of systematic changes in auditory stimulus streams. Participants reported Perceived Salience of Change (PSC) in the mean value of Gaussian sequences of pure tones varying in either frequency or intensity, with sequences differing in precision. We hypothesised that PSC, for a particular absolute mean change, would positively correlate with stimulus precision. Surprisingly, we observed multiple instances of the opposite effect, where PSC was rated as higher in low-precision conditions. The conditions under which we found evidence for a counter-Bayesian strategy was under extreme values of individual stimuli within sequences, and mostly in experiments where frequency rather than intensity was the varied parameter. Further scrutiny of the specific conditions for these surprising results showed that low precision could be associated with worsened, unaffected or improved correct reporting of the direction of sound frequency change. These results raise the intriguing possibility that certain circumstances, particularly those characterised by low signal-to-noise, human perception may adopt a counter-Bayesian strategy, and we discuss the potential mechanisms, evolutionary benefits, and clinical implications for future work to further test this falsifiable hypothesis.