As above, so below
by Axel Cleeremans
A central aspect of the entire hierarchical predictive coding approach, though this is not readily apparent in the corresponding literature, is the emphasis it puts on learning mechanisms. In other works (Cleeremans, 2008, 2011), I have defended the idea that consciousness is itself the result of learning. From this perspective, agents become conscious in virtue of learning to redescribe their own activity to themselves. Taking the proposal that consciousness is inherently dynamical seriously opens up the mesmerizing possibility that conscious awareness is itself a product of plasticity-driven dynamics. In other words, from this perspective, we learn to be conscious. To dispel possible misunderstandings of this proposal right away, I am not suggesting that consciousness is something that one learns like one would learn about the Hundred Years War, that is, as an academic endeavour, but rather that consciousness is the result (vs. the starting point) of continuous and extended interaction with the world, with ourselves, and with others. The brain, from this perspective, continuously (and unconsciously) learns to anticipate the consequences of its own activity on itself, on the environment, and on other brains, and it is from the practical knowledge that accrues in such interactions that conscious experience is rooted. This perspective, in short, endorses the enactive approach introduced by O’Regan and Noë (2001), but extends it both inwards (the brain learning about itself) and further outwards (the brain learning about other brains), so connecting with the central ideas put forward by the predictive coding approach to cognition. In this light, the conscious mind is the brain’s (implicit, enacted) theory about itself, expressed in a language that other minds can understand.
The theory rests on several assumptions and is articulated over three core ideas. A first assumption is that information processing as carried out by neurons is intrinsically unconscious. There is nothing in the activity of individual neurons that make it so that their activity should produce conscious experience. Important consequences of this assumption are (1) that conscious and unconscious processing must be rooted in the same set of representational systems and neural processes, and (2) that tasks in general will always involve both conscious and unconscious influences, for awareness cannot be “turned off” in normal participants.
A second assumption is that information processing as carried out by the brain is graded and cascades (McClelland, 1979) in a continuous flow (Eriksen & Schultz, 1979) over the multiple levels of a heterarchy (Fuster, 2008) extending from posterior to anterior cortex as evidence accumulates during an information processing episode. An implication of this assumption is that consciousness takes time.
The third assumption is that plasticity is mandatory: The brain learns all the time, whether we intend to or not. Each experience leaves a trace in the brain (Kreiman, Fried, & Koch, 2002).
The social roots of consciousness
by Axel Cleeremans
How does this ability to represent the mental states of other agents get going? While there is considerable debate about this issue, it is probably fair to say that one crucial mechanism involves learning about the consequences of the actions that one directs towards other agents. In this respect, interactions with the natural world are fundamentally different from interactions with other agents, precisely because other agents are endowed with unobservable internal states. If I let a spoon drop on a hard floor, the sound that results will always be the same, within certain parameters that only vary in a limited range. The consequences of my action are thus more or less entirely predictable. But if I smile to someone, the consequences that may result are many. Perhaps the person will smile back to me, but it may also be the case that the person will ignore me or that she will display puzzlement, or even that she will be angry at me. It all depends on the context and on the unobservable mental states that the person currently entertains. Of course, there is a lot I can learn about the space of possible responses based on my knowledge of the person, my history of prior interactions with her, and on the context in which my interactions take place. But the point is simply to say that in order to successfully predict the consequences of the actions that I direct towards other agents, I have to build a model of how these agents work. And this is complex because, unlike what is the case for interactions with the natural world, it is an inverse problem: The same action may result in many different reactions, and those different reactions can themselves be caused by many different internal states.
Based on these observations, one provocative claim about the relationships between self-awareness and one’s ability to represent the mental states of other agents (“theory of mind”, as it is called) is thus that theory of mind comes first, as the philosopher Peter Caruthers has defended. That is, it is in virtue of my learning to correctly anticipate the consequences of the actions that dIirect towards other agents that I end up developing models of the internal states of such agents, and it is in virtue of the existence of such models that I become able to gain insight about myself (more specifically: about my self). Thus, by this view, self-awareness, and perhaps subjective experience itself, is a consequence of theory of mind as it develops over extended periods of social intercourse.