In her commentary, Lucia Melloni argues that consciousness unfolds in time: there is a stream of consciousness. What I see now is intricately linked to what I have seen before. And what I see now is what I expect to see—much along the lines of predictive coding. A full understanding of consciousness should not neglect this point. There is even a stronger claim that somehow the process of inference over time is crucial to understanding consciousness.
I appreciate the boldness of linking the framework of Bayesian predictive coding to specific stages in the process of generating consciousness:
One promising framework within which the influence of previous experience can be understood is the Bayesian framework. When applied to perception, each mathematically-formulated ingredient of this framework can be assigned a perceptual counterpart, with previous experience referring to the prior, the current moment referring to the likelihood, unconscious inference referring to Bayes rule (which combines the prior with the likelihood in an optimal way), and the result—our perception—referring to (the peak of) the posterior distribution. (Melloni this collection, p. 4)
To my knowledge, this is the first time this has been so explicitly laid out—writers on predictive coding thus far have always stayed a little vague on where exactly consciousness sits in the Bayesian framework.
Yet at the same time, there is the suggestion of long temporal range integration being the key ingredient of consciousness:
Event-objects of the conscious mind thus per definition unfold in time and we also act in time: we make use of current and previous input to figure out the most appropriate response predicting their consequences. There is thus a continuum of interdependencies along the time dimension whereby every past moment is integrated with the present and projected into the future, giving rise to the flow of consciousness. The same way we have been thinking about the integration of multiple sources of information within a given moment of time, such as multiple features of a single object, there is thus integration across time. (Melloni this collection, pp. 7-8)
This makes intuitive sense, particularly in the case of moving objects, such as the tennis ball Melloni uses as an example. Indeed it is hard—if not impossible—to pinpoint the exact now of conscious experience of such a ball.[1]
Yet the two points seem contradictory. In the Bayesian predictive coding framework, consciousness is the result of the unconscious inferential processes. Previous knowledge and experience (the priors) play an important role, but they are combined with current input to produce the posterior, which is conscious sensation. In the second account, however, consciousness seems to be something that is stretched out over time, so that both prior and posterior are smelted into a “flow” of consciousness. I find it hard to reconcile these two views.