A theory describing when picures look distorted.

Here we see an artichoke spinning while being videotaped at 24 frames-per-second with a very fast shutter speed (1/4000 sec). The rotation speed is chosen to cause…

Abstract: The popular Predictive Processing (PP) framework posits prediction error minimization as the sole mechanism implemented in the brain that can account for all mental phenomena, including consciousness (Hohwy 2013, Clark 2016). In this talk, I emphasize three ambitions put forward by its proponents, relate them to each other, and conclude that none of them can be satisfied by PP. These ambitions can be labelled (1) “Comprehensiveness”, (2) “Realism about Bayes”, and (3) "Naturalism”. The starting point is the central claim that experiential content is identified as the brain’s currently best hypothesis about the world, which “leaves phenomenology at one remove from the world” (Hohwy 2013, 48) and turns perception into “controlled hallucination” (Seth 2021, 186), such that “what we see is never simply how things are” (Clark 2023, xiv/24), i.e., “we never experience the world as it is” (Seth 2021, 92). This claim results from scientific realism about Bayesian modelling, manifest in statements like that the brain “is a Bayesian mechanism” (Hohwy 2013, 25). I connect this result to Kant’s Transcendental Idealism and argue that this impedes PP’s ambition of naturalism as a metaphysical claim about the world as it is, going beyond the contents of experience. A further tension arises with the ambition of comprehensiveness, since when the framework is applied to our investigations of the brain, it is clear that not only cannot consciousness be explained in this way, but neither can we make sense of neuroscience (Zahavi 2018), Frank, Gleiser, Thompson 2024). This leaves proponents of PP with two options. The first is to endorse Transcendental Idealism and give up ambitions (1) and (3). Another option is to retreat from the scientific realism about the Bayesian modelling in favor of either a more moderate instrumentalism (Colombo, Elkin, Hartmann 2018) or a more moderate “haptic realism”, neither of which are committed to the assumption that a Bayesian model of the brain is to be taken literally, as a “truth-apt representation” (Chirimuuta 2024, 38). But this way of conceptualizing the status of Predictive Processing makes it completely independent of naturalism and does not support any claims in this regard.