Predictive processing is standardly combined with views of perceptual experience as the outcome of psychological processes of unconscious inference. Perceptual experience is taken to be the product of guesswork about what is out there in the external world. The brain forms its guess by a process that approximates probabilistic reasoning, combining current multi-modal sensory signals with states that encode “prior beliefs” or “expectations” formed in part through learning. Perceptual experience is thereby understood as the hypothesis that the brain settles upon given the available evidence and its prior beliefs. The processing that culminates in perceptual experience is thus a matter of striking a delicate balance between what is expected given our past learning, and what is currently sensed.
Predictive processing has recently been used to argue for a constructivist account of perceptual experience as “controlled hallucination” (Clark 2013; Seth 2016; c.f. Grush 2004 for an early statement of this idea). The brain forms a sort of virtual reality simulation of what is most likely to be going on in the world. The process of building such an inner simulation is not wild and chaotic as it is in full-blown hallucination or psychosis. The brain must construct an inner simulation that makes the best overall sense of the available sensory evidence given what is already known about the world. It must settle upon a hypothesis based on priors that do the best job of minimising prediction error in the long-term. The contribution of the world to perceptual experience is however simply to “reign in” and control the brain’s predictions so that the agent, at least for the most part, avoids surprising sensory encounters with the world (Seth 2017).
This picture of perceptual experience as controlled hallucination doesn’t seem to sit very well with our contention that predictive processing can be taken to provide support for extended consciousness. Predictive processing seems to entail a view of perceptual experience as the product of processes of active construction that takes place wholly inside of the brain of the organism. The involvement of the external world in this process is to keep in check the worst excesses of the brain’s inner fantasising so that the agent doesn’t become unhinged from the world. Feedback from the world allows the brain to distil the statistical regularities from its environment required for predicting its own sensory input. However, once the regularities are well-learned there is no longer any work for the world to do. We can thus compare the agent’s relation to the world to the 1980s arcade game Pac-Man. Pac-Man explores his maze-world consuming all of the dots he can find. Once the dots are consumed the world of the maze no longer serves any purpose, and he must move on elsewhere to find his dots. The controlled hallucination view leads to a similar conclusion. The agent only needs the world in perceptual processing in order to consume the information it provides. Once the information has been fully harnessed, the world can safely be thrown away and perceptual states can be generated wholly from inside of the brain.
We argue by contrast that perception and action form a circular causal process that couples the agent to the environment. The agent’s active and sensory states form a closed circuit (a Markov blanket – see post 3) only in and through the agent’s coupling with the environment. Thus action can do the work of maintaining and tuning the generative model only in the agent’s coupling with the environment. There is no end-point after which the organism can throw away the world and rely solely upon the generative model installed in its brain for the construction of perceptual experience. The necessity of coupling to the environment through cycles of perception and action for the tuning and maintaining of the generative model is ongoing.
Still the internalist about the realisers of perceptual experience may respond that the ongoing tuning of the predictive machinery is insufficient to establish the constitutive dependence of the realisers of experience on the world. We’ll have more to say about this line of attack in our final post, but note by way of an initial response how there is something question-begging about this move. Such an objection only seems compelling if we suppose that the generative model is located in the brain, issuing its predictions from inside of the head. We argue (see post 3) that the whole embodied agent is the model of its environment. The whole embodied agent is the author of its own sensations in active inference. We are in agreement with Gallagher and Allen (2016) when they write that active inference is “not something happening (exclusively) in the brain, and is not just providing new input for the brain; it’s what the whole organism does in its interaction with the environment.” (Gallagher & Allen 2016, p. 12) The flow of sensation the agents generates through its movements is bound up with and embedded in an ongoing extended dynamic – an ongoing web with feedback loops running across internal and external orbits. The flow of sensation itself is predictable only given a pattern and regularity in the flow, reflecting meaningful structure in the environment and the needs and activities of the individuals themselves. The patterns the agent predicts are what Hurley called sensorimotor dynamics (Hurley 2010). They form as the extended dynamic singularity forms, in the coupling of the agent with its environment.
The third-wave view of predictive processing we develop thus leads to a different account of perceptual experience to dominant interpretations that take perceptual experience to be controlled hallucination. On these more standard interpretations, action is thought of as testing out the brain’s best guess of what is out there in the world. The work of constructing experience is done inside of the brain. However, such an interpretation of perceptual experience won’t fly once we keep in view the extended dynamic singularity that forms through the coupling of the agent and the environment. The controlled hallucination picture of perception rests on a view of the generative model as working in the same way as Pac-Man eating up the information it gathers from the environment so that in the end the world can be thrown away. But conscious subjects are not like Pac-Man – they need the world not only to keep in check the inner fantasising of their brains. Conscious subjects need the world in experience because their experience forms in their sensorimotor coupling with the world.
Clark, A. (2013). Whatever next? Predictive brains, situated agents, and the future of cognitive science. Behavioral and Brain Sciences, 36, 181–253.
Gallagher, S., and Allen, M. (2016). Active inference, enactivism and the hermeneutics of social cognition. Synthese (SI: Predictive Brains, M. Kirchhoff (Ed.)), pp.1-22.
Grush, R. (2004). The emulation theory of representation: motor control, imagery and perception. Behavioural and Brain Sciences 27(3): 396-442
Hurley, S.L. (2010). The varieties of externalism, in Menary, R. (ed.), The Extended Mind (pp. 101–154). Cambridge, MA: MIT Press.
Seth, A. (2017). Your brain hallucinates your conscious reality. TED talk. https://www.ted.com/talks/anil_seth_how_your_brain_hallucinates_your_conscious_reality?language=en