The visual system as an imperfect detective 

A good detective uses all available clues when making inferences. An even better detective will weigh each of these clues by their reliability, such that, for example, the testimony of an unreliable witness matters less than that of a reliable one.

In this talk, I will argue that the visual system works in very much the same way. First, I will present a variety of experimental evidence showing that the human brain indeed weighs visual clues by their reliability, even when the weights change from trial to trial in an unpredictable manner. Thereafter, I will present findings from a more recent study, showing that despite this remarkable feature, there are notable imperfections in the visual system’s detective work. I will speculate on possible causes of these imperfections and discuss what our findings mean in light of the "Bayesian brain" hypothesis.