![Sample our Humanities journals, sign in here to start your FREE access for 14 days]({"type":"image" , "src" : "/sda/5939/TOC-Subject-Sample-2021-Humanities.jpg"})
Abstract
This paper starts by considering an argument for thinking that predictive processing (PP) is representational. This argument suggests that the Kullback–Leibler (KL)-divergence provides an accessible measure of misrepresentation, and therefore, a measure of representational content in hierarchical Bayesian inference. The paper then argues that while the KL-divergence is a measure of information, it does not establish a sufficient measure of representational content. We argue that this follows from the fact that the KL-divergence is a measure of relative entropy, which can be shown to be the same as covariance (through a set of additional steps). It is well known that facts about covariance do not entail facts about representational content. So there is no reason to think that the KL-divergence is a measure of (mis-)representational content. This paper thus provides an enactive, non-representational account of Bayesian belief optimisation in hierarchical PP.
Acknowledgements
We would like to thank the special issue editors Tobias Schlicht, Luke Roelofs, and Krzysztof Dołęga for the invitation to contribute to the special issue. We would also like to thank Julian Kiverstein, Daniel D. Hutto, Shaun Gallagher, Maxwell Ramstead, Karl Friston and select audience members at the Naturally Evolving Minds conference at the University of Wollongong for discussion of several core issues related to this paper. Thanks also to Anco Peeters and Miguel Segundo Ortin for assistance with formatting. Finally, thanks to three anonymous referees for their helpful and constructive comments.
Disclosure statement
No potential conflict of interest was reported by the authors.
Notes on contributors
Michael D. Kirchhoff is an early career researcher and senior lecturer in Philosophy at the University of Wollongong. He is a Chief Investigator on an Australian Research Council funded Discovery Project, Minds in Skilful Performance (DP170102987). He has recently edited a special issue in Synthese on Predictive Brains and Embodied, Enactive Cognition. His works cover different topics across philosophy of cognitive science and theoretical biology. He has published articles in philosophical journals such as the Australasian Journal of Philosophy, Philosophy and Phenomenological Research, Philosophical Studies, and Synthese. He has also published work in interdisciplinary journals like The Journal of the Royal Society Interface, Entropy and Frontiers in Human Neuroscience. He has just submitted a book manuscript, co-written with Dr Julian Kiverstein, to Routledge. The book is entitled Extended Consciousness and Predictive Processing: A Third-Wave View.
Ian Robertson is a PhD candidate at the University of Wollongong, Australia. His PhD is part of an Australian Research Council Discovery Project, “Minds in Skilled Performance” [DP170102987]. He is supervised by Prof. Daniel D. Hutto and Dr Michael D. Kirchhoff. His research concerns novelty and unpredictability in skilful performance, and he focuses particularly on the explanatory capacity of predictive processing accounts of cognition in our understanding the mentality involved in skilful action.
Notes
1 One can find additional support for cognitivist PP from the following authors: Hohwy says that PP provides a view “of the brain as a secluded inference-machine populated with massive, hierarchical representations of the world external to the brain and its sensory organs” (Citation2016, 277). Williams says: “predictive processing is a robustly representational theory of mind” (Citation2017, 13). Williams and Colling think that PP “manifests how a regulative and thoroughly predictive perspective on brain function harmonises with an iconic approach to cognitive representation” (2017, 20). Finally, Wiese assumes content, and asks: “What are the contents of representation in PP?” (Citation2017, 2).
2 We would like to thank an anonymous reviewer for pressing us to make these points clearer.
3 This is not the way the action-guiding idea is presented by cognitivists. But, we find it the best way of making sense of the idea that generative models are action-guiding: the generative model is dependent on the generative process.
4 See Dołęga (Citation2017) for similar observations and a further discussion about the problem of content determination in PP.
5 One reviewer observes that the KL-divergence is sometimes used as a measure of how much information is lost when using a second distribution q to approximate the true distribution p and that this approximation can be used to signal how much information is lost when using an approximation about p instead of p itself. The reviewer speculates that this might indicate a notion of misrepresentation. This is an interesting observation, but one we do not think guarantees the further claim that brains that engage in PP engage in processing that is necessarily representational. We have two reasons for thinking this. The first is that the internal states of the brain do not have “access” to the true distribution. So thinking that the internal states could “know” the difference between the approximation and the true distribution turns on what appears to be an unjustified assumption. Of course, a scientist knowing the true distribution yet using an approximation would know the information loss. But this observation does not warrant the additional claim that internal brain states would be in the same position as the knowing scientist (cf. Bruineberg et al. Citation2016). The second and final point is: if the information measure of the KL-divergence is relative Shannon entropy, then as we have argued this does not license the claim that one variable carries false information about another variable – thus there could be no issue about misrepresentation to the extent that the information in question is relative Shannon information.