The Bayesian Mind and the Nature of Consciousness

Abstract

The article assesses the potential of the program of predictive coding and Bayesian models of cognitive processes to contribute to contemporary con­sciousness studies. The author starts from the question poised in the recent works of philosophers A. Clark and N. Block, namely, if the mechanisms of perception, as suggested by supporters of Bayesian models of cognitive processes, are prob­abilistic, then why does consciousness not reflect the nature of the probabilistic perceptual representations that underlie it? Contrary to Clark and Block, the author points out that there is no reason to believe that consciousness, as a high-level phenomenon, should in principle reflect the characteristics of some more basic processes and representations. Accordingly, preference is given to the question of the prospects and limitations of Bayesian models in cognitive sci­ence, psychology, and neuroscience to help shed light on the problem of human experience. In order to answer this question, the article primarily examines the status of the Bayesian models of cognitive processes, concerning which there is significant disagreement among researchers. It is argued that Bayesian models in cognitive science, psychology, and neuroscience can be formulated at all three of D. Marr’s levels of analysis of information processing systems. At the same time, the most important level for Bayesian models is Marr’s computational level which gives the description of the computational task and the information in the environment available to the organism. Nevertheless, it is pointed out that concerning consciousness studies models related to the Marr’s algorithmic level have been predominantly developed. In conclusion, it is suggested that in order to characterize consciousness at Marr’s computational level the program of pre­dictive coding will probably need integration with the theory of global workspace.