ABSTRACT
Intelligent machines that are able to learn from data have become more and more common. To be efficient, such machines need to represent uncertainty in the data, be adaptive and robust. In neuroscience, the idea has emerged that the brain might work in the same way. The brain would represent beliefs in the form of probabilities, and would have developed an internal model of the world in the form of prior beliefs and that can be consulted to predict and interpret new situations. The brain would then combine new evidence with prior beliefs in principled way, through the application of Bayes theorem. When combining multiple sources of integration, the brain does take into account the reliability of each source of information. Moreover, it is clear that the brain works by using prior beliefs in situations of strong uncertainty. The existence of these beliefs can explain variety of visual illusions, such as the 'hollow mask illusion' or the 'Ames room illusion'.
