Abstract
I offer an account of perceptual warrant according to which one’s basic perceptual beliefs are immediately and defeasibly warranted if they are formed on the basis of experiences produced by a competent perceptual system. I claim that sub-personal features of one’s perceptual systems can render one competent to perceptually represent a particular environment. When these conditions are met, one is warranted in forming beliefs on the basis of one’s perceptual experiences. I develop my account of perceptual warrant in the context of a Bayesian theory of perceptual processing.
This is a preview of subscription content, log in via an institution to check access.
Similar content being viewed by others
Notes
Plantinga (1993), in contrast, defines epistemic warrant as “whatever must be added to true belief in order to have knowledge.” I am skeptical of the theoretic utility of this notion and so do not use the term in this sense.
The retinal state is not the only thing relevant to visual processing. For instance, the visual system takes into account the motion of the eye, and the tension of the muscles that control it in order to produce representations of the environment. I simplify here for the sake of exposition.
I will be provisionally adopting the assumption that human visual systems render reliably veridical verdicts about a range of distal features of their local environments such as: relative object distance, object size, the edges of a surfaces, lighting conditions, the distributions of shadows, surface colors of objects, etc. The assumption of reliability is in place to aid in the presentation of the Bayesian account of perceptual processing that underlies my epistemological account. The view I present here, once fully stated, is ultimately neutral on the degree to which our visual systems are actually reliable. It is, of course, an empirical matter precisely how reliable our visual systems are and under what conditions this reliability is preserved. It is entirely consistent with the present approach that our perceptual systems fail to be competent for some range of distal environmental properties.
One’s judgments about the perceptual competence of a particular visual system might turn on one’s metaphysical commitments. Suppose that one believes, for instance, that there is no such thing as surface color, that our experiences of color do not reliably track any distal properties of object surfaces, that there is no appropriately available error theory that vindicates our perceptual experiences and judgements of surface color, and so on. I (for reasons well beyond the scope of the present discussion) do not find the conjunction of these claims to be plausible. However, if one holds this package of views, then, I think, one ought to conclude that our visual systems are not competent with regards to determinations of color. The degree to which this would impact a judgment of the general competence of a perceptual system would depend crucially on the degree to which the system could reliably extract other information about the distal environment despite this systematic failure. Thanks to an anonymous referee for asking a helpful question on this point.
For more classical instances of a Helmholtzian approach see Gregory (1997) and Rock (1983). Marr (1982) is a famous version of computational theory of vision in the Helmholtzian tradition. Finally, as we will see in the next section, Bayesian models of perception such as those advocated by Knill et al. (1996) Maloney and Mamassian (2009), Mamassian et al. (2002), and Rescorla (2015) are contemporary successors to the general framework that Helmholtz developed.
The following provide useful summaries of the central hypotheses, assumptions, and methodology of Bayesian theories of perception: Feldman (2015), Knill et al. (1996) Maloney and Mamassian (2009), Mamassian et al. (2002), and Rescorla (2015). Geisler (2008) is also useful for an overview about the study of properties of natural scenes and their relevance to the Bayesian approach to the study of vision.
I adopt a realist approach to Bayesian models of perception in what follows. I will be assuming that the priors relied on in these models are actually “encoded by” or “realized in” the perceptual systems that Bayesian theorists are modeling and that Bayesian updating on the basis of these priors is similarly encoded by or neurologically realized in perceptual processing. There are unsettled questions regarding the degree to which one should maintain that perceptual systems like ours realize such models and very good reasons to think that at least some features of these models (e.g. a distribution of prior probabilities over an infinite number of propositions) are not implemented by actual perceptual systems. I adopt a realist interpretation here for ease of presentation. My primary interests in this paper are epistemological. For my purposes it suffices for a perceptual system to be normatively evaluable in Bayesian terms and for this it suffices that such systems can be usefully modeled in these terms. To give one example of how this might work in practice: Orlandi (2014) argues that the best explanation of the behavior of our visual systems involves no genuinely representational sub-personal states. According to Orlandi our visual systems encode no contentful priors or likelihoods about distal states. However, Orlandi does not reject the empirical Bayesian approach to visual processing described here. To oversimplify the book’s excellent discussion, Orlandi claims that these Bayesian accounts can be re-interpreted in purely functional terms with priors and likelihoods amounting to nothing more than mere biases or functional tendencies. While I have reservations about this view, I take it that Orlandi could consistently endorse the epistemological framework I present here. So long as the Bayesian models are explanatorily successful in Orlandi’s terms, one could argue that the appropriate epistemic evaluation of the competence of such a system ought to be carried out in Bayesian terms such as those developed above. That is, since the perceptual system in question is best modeled as a Bayesian system, even if it does not literally encode priors and likelihoods etc. one could carry out an evaluation of its epistemic competence in the Bayesian terms I propose here.
I will bracket discussion of the predictive processing or predictive coding Bayesian models championed by Hohwy (2013) and Clark (2016). Though these approaches utilize Bayesian updating in their models of perceptual processing, they possess fundamental structural differences to the approach described here. I will characterize my view in standard constructivist, bottom-up terms. Clark and Hohwy, by contrast, hypothesize a hierarchical sequence of models, each of which attempts to predict the inputs of the model immediately preceding it in the hierarchy. While I believe that the general approach to explaining perceptual competence I employ here could be adapted to these accounts, their structure complicates matters to a sufficient degree that these models warrant their own discussion.
This is a variation on a case presented by Burge (1986).
This terminology is due to Burge (2003).
Burge (2003) provides an extensive argument to this effect. His discussion of these matters was highly influential on the development of the view I present here.
The contemporary science associated with an examination of the properties of the visually relevant regularities of the distal environment in which we find ourselves is known as Natural Scene Statistics. Researchers in this field use advanced algorithms to survey massive numbers of high-quality images of natural scenes searching for regularities. These regularities in natural scenes are what the visual system exploits in order to solve the underdetermination problem (Geisler 2008, 172).
One could hold a view according to which the perceptual experience itself is not epistemically relevant, or at least not necessarily so. An anonymous referee suggests, for instance, that the posterior probability distribution could be understood as a ranked set of warranted perceptual beliefs or credences that guide action and further inference. Additionally, it seems clear that one could at least imagine a cognitive system that simply move directly from the posterior probability distribution to a perceptual belief without any intervening experience. Finally, a more classical reliabilist position (e.g. Goldman 1979, or perhaps Lyons 2009) could maintain that the perceptual experience is not necessary for a warranted perceptual belief so long as the resulting belief is produced by a reliable process. I accept that an organism with a sufficiently different cognitive structure to our own that does not have anything analogous to our perceptual experiences would operate according to a distinct set of epistemic norms than those that govern us. However, I find it difficult to deny, from both an intuitive and theoretical perspective that perceptual experiences are epistemically relevant for normal adult humans. If a contentful perceptual experience plays any functional or causal role in the formation of the resulting perceptual beliefs, then the manner in which the experience is generated from a probability distribution will be epistemically relevant (for the reasons discussed above). However, a reader who questions the epistemic relevance of the perceptual experience itself (and/or denies the need to include the of the utility function in our normative evaluations) could adapt the view I present here to their purposes by omitting consideration of the utility function in their preferred formulation and applying the general structure of my view directly to the warrant-conditions for perceptual beliefs. My thanks to the aforementioned referee for a very astute series of questions regarding this and some closely related issues that significantly improved this paper.
This example of a probability distribution is due to a very helpful conversation with Michael Rescorla. I learned much on both Bayesian probability theory and its role in contemporary perceptual psychology from two of his excellent graduate seminars in 2012 and 2014 at U.C. Santa Barbara.
Mamassian et al. (2002) present an overview of some of the issues involved in determining the relative merits of utility functions, and Brainard and Freeman (1997) provide a highly technical discussion of the statistical merits of several competitors. The exact details of these accounts, while interesting in their own right, are somewhat beyond the bounds of my present discussion.
It will be noted that the preceding proposal has been developed in terms of a particular sort of model of perceptual processing. One objection to the present approach could be stated as follows: if this Bayesian picture turns out to be inadequate or is supplanted by future developments in perceptual psychology, the account I present here will thereby be shown to be unsound. To a certain extent, I am happy to grant that this is so. One goal of the project that I am engaged in here is to take stock of the manner in which perceptual systems like ours function in order to explain how and under what conditions our perceptual experiences warrant our perceptual beliefs. If it were to turn out that human perceptual systems are in no way Bayesian in the sense described here, then this would suffice to show that the preceding account does not explain our perceptual warrant. Therefore, should the basic Bayesian theory described in the preceding pages turn out to be explanatorily inadequate, many of the details of the present account would have to change. Still, I would caution the reader against a too-pessimistic interpretation of this concession. The general framework that Helmholtz pioneered of explaining visual processing in terms of unconscious inferences and implicit assumptions has proven to be remarkably resilient and explanatorily fertile. Conditions for perceptual competence can be developed in the terminology of any broadly Helmholtzian theory of the perceptual psychology. So long as this central idea is preserved, then much of the preceding approach could be adapted to the structure and terminology of competing perceptual models. Thanks to an anonymous referee for asking for clarification on these issues.
Though the specific elements of my proposal are different, these arguments in favor of privileging an organism’s normal environment in evaluating its perceptual competence closely parallel those given by Burge (2003).
Gibson (1979) argues that an environment in this context should be distinguished from a set of physical states. An environment, in his terminology, is something that an organism perceives and behaves in. I am happy to take this constraint on board, but this will not help in the present context. Our evolutionary ancestors and the environments they inhabited were causally relevant to certain features of the human perceptual system.
This diagram is of similar form to several presented in Woodward (2003).
Consider again the results of Weiss et al. (2002).
For instance, Sanborn and Chater (2016) hypothesize that the brain is a Bayesian sampler, that does not represent an entire probability distribution over infinite hypothesis spaces, but samples a much smaller “local” probability distribution on which to carry out approximated Bayesian calculations (Sanborn and Chater 2016, 884). Though this opens such a system up to probabilistic fallacies (e.g. base-rate neglect, the conjunction fallacy, and the unpacking fallacy), it involves much more tractable calculations than those involved in the idealized Bayesian inference employed in standard Bayesian models.
Thanks to an anonymous referee for asking a helpful question on this point.
References
Brainard, D. H., & Freeman, W. T. (1997). Bayesian color constancy. Journal of the Optical Society of America, 14(7), 1393–1411.
Burge, T. (1986). Individualism and psychology. Philosophical Review, 95, 3–45.
Burge, T. (1993). Content Preservation. Philosophical Review, 102(4), 457–488.
Burge, T. (2003). Perceptual entitlement. Philosophy and Phenomenological Research, 67(3), 503–548.
Burge. T. (2010). Origins of Objectivity. Oxford University Press.
Clark, A. (2016). Surfing uncertainty: Prediction, action, and the embodied mind. Oxford: Oxford University Press.
Feldman, J. (2015). Bayesian models of perception: A tutorial introduction. Oxford: Oxford Handbook of Perceptual Organization.
Geisler, W. S. (2008). Visual perception and the statistical properties of natural scenes. Annual Review of Psychology, 59, 167–192.
Geisler, W. S., & Kersten, D. (2002). Illusions, perception and Bayes. Nature Neuroscience, 5(6), 508–510.
Gettier, E. L. (1963). Is justified true belief knowledge? Analysis, 23(6), 121–123.
Gibson, J. J. (1979). The ecological approach to visual perception: Classic Edition (1st ed.). New York: Hove.
Goldman, A. I. (1979). “What is justified belief?” In Justification and knowledge (pp. 1–23). Springer.
Goldman, A. I. (1986). Epistemology and cognition. Cambridge: Harvard University Press.
Graham, P. J. (2011). Perceptual entitlement and basic beliefs. Philosophical Studies, 153(3), 467–475.
Gregory, R. L. (1997). Eye and brain (5th ed.). Princeton, N.J: Princeton University Press.
Hohwy, J. (2013). The Predictive Mind. Oxford: Oxford University Press.
Huemer, M. (2001). Skepticism and the veil of perception. Lanham: Rowman & Littlefield.
Huemer, M. (2007). Compassionate phenomenal conservatism. Philosophy and Phenomenological Research, 74(1), 30–55.
Kersten, D., Mamassian, P., & Yuille, A. (2004). Object perception as Bayesian inference. Annual Review of Psychology, 55, 271–304.
Knill, D. C., Kersten, D., & Yuille, A. (1996). Introduction: A Bayesian formulation of visual perception. In Perception as Bayesian Inference (pp. 1–21). Cambridge University Press.
Knill, D. C., & Richards, W. (1996). Perception as Bayesian inference. Cambridge: Cambridge University Press.
Lyons, J. (2009). Perception and basic beliefs: Zombies, modules, and the problem of the external world. Oxford: Oxford University Press.
Maloney, L. T., & Mamassian, P. (2009). Bayesian decision theory as a model of human visual perception: Testing Bayesian transfer. Visual Neuroscience, 26(1), 147–155.
Mamassian, P, Landy, M. S., & Maloney, L. T. (2002). Bayesian modelling of visual perception. In Probabilistic Models of the Brain: Perception and Neural Function (pp. 13–36). MIT Press.
Marr, D. (1982). Vision. New York: Freeman.
Matthen, M. (Ed.). (2015). The Oxford handbook of the philosophy of perception (1st ed.). Oxford: Oxford University Press.
Orlandi, N. (2014). The innocent eye: Why vision is not a cognitive process. Oxford: Oxford University Press.
Palmer, S. E. (1999). Vision Science: Photons to Phenomenology (1st ed.). Cambridge, Mass: The MIT Press.
Peacocke, C. (2004). The realm of reason. Oxford: Oxford University Press.
Plantinga, A. (1993). Warrant and proper function. Oxford: Oxford University Press.
Pryor, J. (2000). The skeptic and the dogmatist. Noûs, 34(4), 517–549.
Rao, R. P. N., Olshausen, B. A., & Lewicki, M. S. (2002). Probabilistic models of the brain: Perception and neural function. Cambridge: MIT press.
Rescorla, M. (2015). Bayesian perceptual psychology. Oxford: The Oxford Handbook of the Philosophy of Perception.
Rock, I. (1983). The logic of perception. Cambridge: MIT Press.
Sanborn, A. N., & Chater, N. (2016). Bayesian brains without probabilities. Trends in Cognitive Sciences, 20(12), 883–893.
Sosa, E. (2007). A virtue epistemology: Apt belief and reflective knowledge (Vol. I). Oxford: Oxford University Press.
Sosa, E. (2009). Reflective knowledge: Apt belief and reflective knowledge (Vol. II). Oxford: Oxford University Press.
Von Helmholtz, H. (1867). Handbuch der physiologischen Optik. III Voss.
Wagemans, J. (2015). The Oxford handbook of perceptual organization. Oxford: Oxford University Press.
Weiss, Y., Simoncelli, E. P., & Adelson, E. H. (2002). Motion illusions as optimal percepts. Nature Neuroscience, 5(6), 598–604.
Woodward, J. (2003). Making things happen: A theory of causal explanation (1st ed.). Oxford: Oxford University Press.
Acknowledgments
I am grateful to Kevin Falvey, Michael Rescorla, and Aaron Zimmerman for reviewing these ideas in their most rudimentary form and providing me with excellent suggestions, criticisms and notes. Previous versions of this paper have been presented at: the U.C. Santa Barbara Philosophy Graduate Colloquium Series, the Alabama Philosophical Society, the Southern Society for Philosophy and Psychology, the University of Alabama, and the University of Alabama, Birmingham. My thanks to the commenters and audiences at all of these events. Finally, thanks to two anonymous reviewers at Erkenntnis who provided excellent feedback instrumental in improving the final version of this paper.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
About this article
Cite this article
Butzer, T. Towards a Bayesian Account of Perceptual Competence. Erkenn 87, 1043–1061 (2022). https://doi.org/10.1007/s10670-020-00229-0
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10670-020-00229-0