Gipps vs Seth: The Muddle of Predictive Processing

        “without a constant misuse of language, there cannot be any discovery, any progress”

        Paul Feyerabend, Against Method

This blogpost is written in response to a blogpost by Richard Gipps in which he critiques the account of perception as a form of prediction and “controlled hallucination” as presented in Anil Seth’s book Being You. Gipps takes a Wittgensteinian approach here and is concerned with the ways in which Seth fails to define crucial terms and argues that the account as presented by Seth is philosophically confused and muddled to the point of being not even false. Gipps is very incisive in his analysis and I would encourage readers to read his post in detail. He particularly zeroes in what is ambiguous and murky in Seth’s descriptions and shows how this murkiness leads to philosophical problems.

When I read Seth’s book last year, I enjoyed it considerably, and found it intelligible in an intuitive way that Gipps apparently does not. While I have little to no scientific or philosophical expertise when it comes to neuroscientific accounts of predictive processing and active inference, I find myself quite sympathetic to them given what I know and understand. So I am perturbed by the fact that Gipps finds the whole thing confused to the point of meaninglessness. In this blogpost I will try to unpack this and show why predictive processing accounts are still intelligible. I have to clarify: I am not defending Seth’s book per se or the exact way he describes things in it. I am more interested in the general scientific ideas and questions tackled by the book with regards to the question of perception.
#1. Appearance and Reality
Anscombe: ‘I suppose, because it looked as if the sun went round the Earth.’
Wittgenstein: ‘Well, what would it have looked like if it had looked as if the Earth turned on its axis?’
Much is made of this exchange between Anscombe and Wittgenstein, and Gipps discusses how Seth misunderstands what the exchange is intended to convey. Gipps is right that Seth misunderstands the intended lesson that Wittgenstein was offering, but it does seem to me that there is a discussion to be had regarding appearance and reality.
Gipps writes: “Recall that the ordinary use of either 'seems as though' or 'looks as if', in the context of perceptual judgement is either i) to distinguish veridical perception from perceptual illusion or hallucination, or ii) to express self-conscious caution.”
I think there is another sense in which we can use the term. It seems as if the sun goes around the earth. According to Wittgenstein and Gipps, here 'seems as if' is not being deployed with any meaning at all. Perhaps so, but perhaps there is more to be said. If we were to extrapolate parsimoniously from what we observe, what we see, what it appears, what it seems to us (i.e. sun moving in the sky from east to west), we would hypothesize a scientific model in which the sun is in orbit around the earth. It is easy to see how the sense in which “it seems as if the sun goes around the earth” referring to our observation of sun’s movement in the sky can be conflated with the scientific model according to which sun is in orbit around the earth. This scientific model is decidedly false; the sun is not in orbit around the earth, however things appear to us. We shouldn't be so preoccupied with the language itself that we forget there is an independent scientific question to be asked. In a similar way, questions pertaining to the use of language about perception should not lead us to ignore the scientific questions at hand.
#2. The Scientific Question About Perception
Speaking about the scientific question, for Gipps there isn’t even a meaningful scientific question to raise here. He writes in one passage:
“Now, one way to mobilise the 'how do we see?' question, one way to give it at least the appearance of intelligibility, is to imagine first that we really are somehow stuck inside our own skulls, and therefore forced to perceptually reconstruct a now external world using images that appear on the retina. And now the question 'well how do we do that?!' will - to say the least - appear pressing. But undo, avoid, this alienated conception of our perceptual encounter with the world, and it's none too obvious that there's a question remaining which requires the provision of an alternative answer. If 'how do we see?' is to be understood as inviting an answer in neurological terms, then all well and good.  But a psychological or an epistemological answer? What, exactly, is the psychological or epistemological problem that the question is addressing?! I can't myself see one, and so don't see what it is that a psychology of vision is here supposed to be doing. But perhaps there is a good question hereabouts? Well, I'm all ears: do tell!”
This strikes me as a product of a confusion between “I am stuck inside my skull” with “the brain is stuck inside the skull.” am not struck inside the skull because I am not my brain. have direct access to the world around me. I can look up at the sky and see the clouds. I can reach out and grab the coffee cup on the table in the front. can raise my hands and say “Here is one hand” and “Here is another”, and experience no worries about the existence of the “external word.” But this also exists in the realm of ordinary language, of appearance, of how things as they seem. What is the scientific reality of this is not a question that can be answered simply by an examination of the language. I am not stuck inside the skull, but the brain is. This is simply an anatomical fact. The brain is confined to the skull. The only access it has to the world outside the skull is via sensory nerves. There is no mystery is how I see thing. I just do. I look around and see the world in all its beauty and ugliness. But in order for the brain to make this possible, an explanation is needed. So there is a scientific question here: how do we go from brain being confined to the skull, with access only to the “signals” (fluctuations in the membrane potential) in the sensory nerves, to the I that just looks around and sees things. What is the explanation here? How is the latter made possible? If we are dependent for our perception on the activity of the brain, as Gipps agrees, then what are the processes by which perception takes place?
Gipps says that it is “all well and good” if this is understood as “inviting an answer in neurological terms.” But an answer in purely neurological terms will not offer us the explanation we need; what will be missing will be explanation that connects the neurological activity to the perceiving I. There is, therefore, a cognitive and psychological question here as well. What cognitive and psychological processes are involved in our ordinary experience of perception? It is worth remembering that an explanation of how perception works in ordinary language is not a scientific explanation of the psychology of perception. As Gipps himself writes: “what you're conscious of, when you actually perceive something, is something that's on the desk in front of you! (To voice this is not to engage in either naive or sophisticated or philosophical or psychological theorising about perception: it's merely to remember how to use the word 'perceive'.)” To remember how to correctly use the word perceive doesn’t by itself tell us what the correct scientific psychological account of perception is, and any scientific psychological account of perception has to take into account the fact that the brain is confined inside the skull and only has access to signals in the sensory nerves. The relationship between the voltage changes in the nerve membranes and the world outside the sensory organs is not a question that has a straightforward obvious answer, and I refuse to accept that this is a meaningless question that arises only because we are confused about how we use the word “perceive”!
Going back to the notion that scientific models can be extrapolated from what we observe, what we see, what it appears, what it seems to us, it is tempting to hypothesize that perception primarily involves a bottom-up neurological process of “information” flowing from sensory organs to the sensory cortices of the brain via sensory nerves. That in fact has been the traditional framework. This sort of bottom-up information flow has a lot of problems that both Seth and Gipps go into, and I will bypass here. But there is a different scientific story to be told as well, a difference hypothesis to be considered, one that would result in me being able to perceive and reach out to grab the coffee cup in front of me. This alternative hypothesis is described, for instance, by Walsh et al as “perception arises from a purely inferential process supported by two distinct classes of neurons: those that transmit predictions about sensory states and those that signal sensory information that deviates from those predictions.”
#3. Prediction and Inference
Gipps is of the view that brains cannot meaningfully be said to predict or infer anything, since this is something that requires intention and agency, among other things, and not something that “behaviourally inert, non-vocal, non-verbal internal organs” such as brain can be said to do. Gipps allows us that we may use the word predict in some different sense when applied to the brain and validly complains that Seth doesn’t offer any definition in the book.
I have several reactions to this. The first is that it strikes me as quite valid to hypothesize or talk about prediction in an analogous way to ordinary language but in a manner that doesn’t require intention or agency, etc. This is especially because we can meaningfully talk about mathematical models making predictions, and a variety of non-intentional, non-agential systems can enact said mathematical models. In mathematics and computer science, therefore, we already have scientific precedents of the use of the term “predict” in a manner that deviates from prediction as something that humans intentionally and rationally engage in. Helmholtz, when he described perception as an inference, used inference in an analogous way: “[the “psychical activities” leading to perception] are in general not conscious, but rather unconscious. In their outcomes they are like inferences insofar as we from the observed effect on our senses arrive at an idea of the cause of this effect.” (Helmholtz 1867) [my emphasis, notice use of “like inference” suggesting an analogy].
The second problem I have is with the insistence that unless an explicit definition is offered, the use must be considered muddled or nonsensical. Just because Helmholtz does not further specify what “like inference” is supposed to be, does that make it muddled and nonsensical? I don’t think so. Scientific ideas often begin as a sort of analogy, and are further refined and made more precise over time. Didn’t Wittgenstein have something to say about the meaning of a word being its use? If a word is being used by an entire community of scientists, can we not recognize that use as legitimate, even if a formal definition is lacking?
Third, we can actually describe prediction and inference in a more formal and precise way. One way to do so (but I suspect not the only way) is to invoke Bayes’s rule. Jokob Hohwy has a wonderful chapter “The Predictive Processing Hypothesis” in the Oxford Handbook of 4E Cognition in which he discusses the compatibility between predictive processing and 4E cognition. (Hohwy concludes: “The initial impression may be that predictive processing is too representational and inferential to fit well to 4E cognition. But, in fact, predictive processing encompasses many phenomena prevalent in 4E approaches, while remaining both inferential and representational.” Gipps and others may disagree with the compatibility between PP and 4E, but that's a separate issue. My concern here is primarily about the intelligibility of PP.)
Hohwy discusses how “… subject to a number of assumptions about the shape of the probability distributions and the context in which they are considered, a system that minimizes prediction error in the long run will approximate Bayesian inference.”
“The heart of PEM [Prediction Error Minimization] is then the idea that a system need not explicitly know or calculate Bayes’s rule to approximate Bayesian inference. All the system needs is the ability to
minimize prediction error in the long run. This is the sense in which unconscious perceptual inference is inference: internal models are refined through prediction error minimization such that Bayesian inference is approximated. The notion of inference is therefore nothing to do with propositional logic or deduction, nor with overly intellectual application of theorems of probability theory.”
“perceivers harbor internal models that give rise to precision-weighted predictions of what the sensory input should be, and that these predictions can be compared to the actual sensory input. The ensuing prediction error guides the updates of the internal model such that prediction error in the long run is minimized and Bayesian inference approximated”
References to internal models are not subject to the “fallacy of double transduction” here, since it is not being asserted that these internal models are then being mysteriously perceived in the form of inner images. Rather, the process of perception itself involves the generation and updating of models. Again, one may disagree with that, but the claim is not itself meaningless or nonsensical.
So here we have an articulation of predictive processing and inference as a process that approximates Bayes’s rule. Does the brain approximate Bayes’s rule in the process of perception? This is an empirical question, to be settled by scientific inquiry, but it certainly cannot be settled or eliminated by an analysis of ordinary language.