This is the second of three posts about the brain having a singular purpose of prediction error minimization. PEM literature has many contributors. Karl Friston is probably the strongest idea man, but Andy Clark and Jakob Hohwy are more understandable. Hohwy’s papers include: Hohwy, J. (2015). “The Neural Organ Explains the Mind”. In T. Metzinger & J. M. Windt (Eds). Open MIND: 19(T). Frankfurt am Main: MIND Group. Hohwy, J., Roepstorff, A., & Friston, K.(2008). “Predictive coding explains binocular rivalry: an epistemological review.” Cognition 108, 687-701. Hohwy, J. (2012). “Attention and conscious perception in the hypothesis testing brain.” Frontiers in Psychology/Consciousness Research, April 2012, Volume 3, Article 96. Paton, B., Skewes, J., Firth, C., & Hohwy, J(2013). “Skull-bound perception and precision optimization through culture.” Commentary in Behavioral and Brain Sciences (2013) 36:3, p 42.
Both Clark and Hohwy use “explaining away” to illustrate the concept of cancelling out sensory prediction error. Perception thus involves “explaining away” the driving (incoming) sensory signal by matching it with a cascade of predictions pitched at a variety of spatial and temporal scales. These predictions reflect what the system already knows about the world (including the
body) and the uncertainties associated with its own processing. What we perceive depends heavily upon the set of priors that the brain brings to bear in its best attempt to predict the current sensory signal.
Hohwy uses this example:
Perhaps we can describe what goes on in terms of “explaining away”. Imagine, for example, that one night the electricity in your house cuts out. You consider two hypotheses: that a possum has torn down the power line to your house, or that the whole neighborhood has blacked out due to the recent heat wave. Out in the street you see other people checking their fuse boxes and this evidence favors the second hypothesis. Importantly, this evidence considerably lowers the probability of the possum hypothesis even though the two hypotheses could be true together. There is debate about what explaining away really is, but agreement that it exists. Part of what grounds this notion is that our background knowledge of the frequency of events tells us that it would be rather an unusual coincidence if, just as the overall power goes out due to the heat wave, a possum caused the line to go down (unless possums are known to take to power lines during heat waves).
(Although I am not familiar with fuse boxes being outside, I have been known to look for lights in similar situations.) Hohwy thus uses a judgment and decision making example to illustrate hierarchical predictive processing.
This appeal to “explaining away” is important as it reflects the key property of hierarchical predictive processing models, which is that the brain is in the business of active, ongoing, input prediction and does not (even in the early sensory case) merely react to external stimuli. This model is also attractive for its coding efficiencies, since all that needs to be passed forward through the system is the error signal, which is what remains once predictions and driving signals have been matched. Hohwy suggests that we should not, however, overplay this difference. In particular, it is potentially misleading to say that: “Activation in early sensory areas no longer represents sensory information per se, but only that part of the input that has not been successfully predicted by higher-level areas.” It is potentially misleading because this stresses only one aspect of what is actually depicted as a kind of duplex architecture: one that at each level combines quite traditional representations of inputs with representations of error. According to the duplex proposal, what gets “explained away” or cancelled out is the error signal, which (in these models) is depicted as computed by dedicated “error units.” These are linked to, but distinct from, the so-called representation units meant to encode the causes of sensory inputs. By cancelling out the activity of the error units, activity in some of the laterally interacting “representation” units (which then feed predictions downward and are in the business of encoding the putative sensory causes) can actually end up being selected and sharpened.
What is most distinctive about this duplex architectural proposal (and where much of the break from tradition really occurs) is that it depicts the forward flow of information as solely conveying error, and the backward flow thus achieves a rather delicate balance between the familiar (there is still a cascade of feature-detection, with potential for selective enhancement, and with increasingly complex features represented by neural populations that are more
distant from the sensory peripheries) and the novel (the forward flow of sensory information is now entirely replaced by a forward flow of prediction error). This balancing act between cancelling out and selective enhancement is made possible, it should be stressed, only by positing the existence of “two functionally distinct sub populations, encoding the conditional expectations of perceptual causes and the prediction error respectively”. Superficial pyramidal cells are depicted as playing the role of error units, passing prediction error forward, while deep pyramidal cells play the role of representation units, passing predictions (made on the basis of a complex generative model) downward. However it may (or may not) be realized, some form of functional separation is required. Hohwy sees such separation as “a central feature of the proposed architecture, and one without which it would be unable to combine the radical elements drawn from predictive coding with simultaneous support for the more traditional structure of increasingly complex feature detection and top-down signal enhancement.”