Expecting the world: Perception, prediction, and the origins of human knowledge

Journal of Philosophy

Volumn 110, Issue 9, 2013, Pages 469-496

  Clark, Andy ^a  

^a UNIVERSITY OF EDINBURGH   (United Kingdom)

Author keywords

[No Author keywords available]

Indexed keywords

EID: 84890920983     PISSN: 0022362X     EISSN: 19398549     Source Type: Journal    
DOI: 10.5840/jphil2013110913     Document Type: Article

References (106)

1. Classic treatments include David Mumford, "On the Computational Architecture of the Neocortex II: The Role of Cortico-Cortical Loops," Biological Cybernetics, lxvi,3. (January 1992): 241-51
2. Rajesh Rao and Dana Ballard, "Predictive Coding in the Visual Cortex: A Functional Interpretation of Some Extra-Classical Receptive-Field Effects," Nature Neuroscience, ii, 1 (January 1999): 79-87
3. Tai Sing Lee and Mumford, "Hierarchical Bayesian Inference in the Visual Cortex," Journal of the Optical Society of America A, xx, 7 (Jul. 1, 2003): 1434-48
4. Karl Friston, "A Theory of Cortical Responses," Philosophical Transactions of the Royal Society B: Biological Sciences, ccclx, 1456 (Apr. 29, 2005): 815-36.
5. see Friston, "The Free-Energy Principle: A Unified Brain Theory?," Nature Reviews: Neuroscience, xi,2. (February 2010): 127-38.
6. Notable exceptions include work by Jakob Hohwy, including "Functional Integration and the Mind," Synthese, clix, 3 (December 2007): 315-28
7. "Attention and Conscious Perception in the Hypothesis Testing Brain," Frontiers in Psychology, iii,96. (Apr. 2, 2012): 1-14
8. see for example Grush, "The Emulation Theory of Representation: Motor Control, Imagery, and Perception," Behavioral and Brain Sciences, xxvii, 3 (April 2004): 377-96
9. Eliasmith, "How to Build a Brain: From Function to Implementation," Synthese, clix,3. (December 2007): 373-88).
10. David H. Hubel and Torsten N. Wiesel, "Receptive Fields and Functional Architecture in Two Nonstriate Visual Areas (18 and 19) of the Cat," Journal of Neurophysiology, xxviii, 2 (Mar. 1, 1965): 229-89
11. David Marr, Vision (San Francisco: Freeman, 1982)
12. Irving Biederman, "Recognition-by-Components: A Theory of Human Image Understanding," Psychological Review, xciv, 2 (April 1987): 115-47.
13. Thus consider expert observers of, say, sports or chess. Such observers benefit from much richer structures of knowledge supporting their top-down predictions. But their brains may, as a result, sometimes overweight acquired expectations ("priors") relative to the driving sensory signal. Daily life, where we are all expert observers to some degree, provides many examples of such overweighting, for example when we constantly seem to see our familiar but temporarily absent pet in the subtle play of light and shadow.
14. See Helmholtz, Handbuch der physiologischen optic (Leipzig, Germany: Voss, 1867)
15. English-language edition Treatise on Physiological Optics, Volume III: The Perceptions of Vision, ed. James P. C. Southall (New York: Dover, 1962).
16. Ulric Neisser, Cognitive Psychology (New York: Appleton-Century-Crofts, 1967
17. and, for a recent review, Daniel Kersten and Alan L. Yuille, "Vision as Bayesian Inference: Analysis by Synthesis?," Trends in Cognitive Sciences, x, 7 (July 2006): 301-08). In the last two decades these broad visions were given effective computational flesh.
18. Key publications include Geoffrey E. Hinton et al., "The 'Wake-Sleep' Algorithm for Unsupervised Neural Networks," Science, cclxviii, 5214 (May 26, 1995): 1158-61
19. Rao and Ballard, "Predictive Coding in the Visual Cortex: A Functional Interpretation of Some Extra-Classical Receptive-Field Effects."
20. For a useful review, see Hinton, "Learning Multiple Layers of Representation," Trends in Cognitive Sciences, xi,10. (October 2007): 428-34.
21. see David C. Knill and Alexandre Pouget, "The Bayesian Brain: The Role of Uncertainty in Neural Coding and Computation," Trends in Neuroscience, xxvii, 12 (December 2004): 712-19).
22. Kestutis Kveraga, Avniel S. Ghuman, and Moshe Bar, "Top-Down Predictions in the Cognitive Brain," Brain and Cognition, lxv, 2 (November 2007): 145-68.
23. Harriet Feldman and Friston, "Attention, Uncertainty, and Free-Energy," Frontiers in Human Neuroscience, iv, 215 (Dec. 2, 2010): 1-23.
24. This description of the standard model is from David Poeppel and Philip J. Monahan, "Feedforward and Feedback in Speech Perception: Revisiting Analysis by Synthesis," Language and Cognitive Processes, xxvi, 7 (2011): 935-51. The quoted passage is from p. 936. Poeppel and Monahan do not, however, endorse that traditional model, and instead argue for the alternative approach described here.
25. See Kenneth N. Stevens and Morris Halle, "Remarks on Analysis by Synthesis and Distinctive Features," in W. Wathen-Dunn, ed., Models for the Perception of Speech and Visual Form (Cambridge: MIT, 1967), pp. 88-102.
26. Poeppel and Monahan, "Feedforward and Feedback in Speech Perception: Revisiting Analysis by Synthesis," p. 939.
27. see Andy Clark, "Whatever Next? Predictive Brains, Situated Agents, and the Future of Cognitive Science," Behavioral and Brain Sciences, xxxvi, 3 (May 2013): 181-204.
28. Gareth Evans, The Varieties of Reference (Oxford: University Press, 1982).
29. "Invertebrate Concepts Confront the Generality Constraint (and Win)," in Robert W. Lurz, ed., The Philosophy of Animal Minds (New York: Cambridge, 2009), pp. 89-107
30. See, for example, George Edward Moore's "The Refutation of Idealism," reprinted in Philosophical Studies (London: Routledge and Kegan Paul, 1903/1922)
31. Gilbert Harman's 1990 paper, "The Intrinsic Quality of Experience," in James Tomberlin, ed., Philosophical Perspectives 4 (Atascadero, CA: Ridgeview, 1990).
32. The illustration is used with Dennett's kind permission.
33. In this context, "experience-driven" just means driven by the incoming streams of sensory information. Such incoming information need not (though it may) be consciously experienced.
34. Stefan J. Kiebel, Jean Daunizeau, and Friston, "Perception and Hierarchical Dynamics," Frontiers in Neuroinformatics, iii, 20 (2009): 1-9, at p. 7.
35. For a review, see Steven Pinker's How the Mind Works (New York: Norton, 1997).
36. See for example Jeffrey L. Elman, "Connectionist Models of Cognitive Development: Where Next?," Trends in Cognitive Sciences, ix, 3 (March 2005): 111-17.
37. See David E. Rumelhart, Hinton, and Robert J. Williams, "Learning Internal Representations by Error Propagation," in Rumelhart, James L. McClelland, and the PDP Research Group, eds., Parallel Distributed Processing: Explorations in the Microstructure of Cognition, Volume 1: Foundations (Cambridge: MIT, 1986), pp. 318-62.
38. see Hinton, "Learning Multiple Layers of Representation."
39. see Hinton, "To Recognize Shapes, First Learn to Generate Images," in P. Cisek, T. Drew, and J. Kalaska, eds., Computational Neuroscience: Theoretical Insights into Brain Function (Boston: Elsevier, 2007), pp. 535-48.
40. Peter Dayan et al., "The Helmholtz Machine," Neural Computation, vii, 5 (September 1995): 889-904
41. Dayan and Hinton, "Varieties of Helmholtz Machine," Neural Networks, ix, 8 (November 1996): 1385-403.
42. See also Hinton and Richard S. Zemel, "Autoencoders, Minimum Description Length and Helmholtz Free Energy," in Jack D. Cowan, Gerald Tesauro, and J. Alspector, eds., Advances in Neural Information Processing Systems 6 (San Mateo, CA: Morgan Kaufmann, 1994).
43. This was a computationally tractable approximation to "maximum-likelihood learning" as used in the expectation-maximization (EM) algorithm.
44. See Arthur P. Dempster, Nan M. Laird, and Donald B. Rubin, "Maximum Likelihood from Incomplete Data via the EM Algorithm," Journal of the Royal Statistical Society, Series B, xxxix, 1 (1977): 1-38.
45. Radford M. Neal and Hinton, "A View of the EM Algorithm that Justifies Incremental, Sparse, and Other Variants," in Michael I. Jordan, ed., Learning in Graphical Models (Dordrecht: Kluwer, 1998), pp. 355-68
46. Hinton et al., "The 'Wake-Sleep' Algorithm for Unsupervised Neural Networks."
47. see Yoshua Bengio, "Learning Deep Architectures for AI," Foundations and Trends in Machine Learning, ii, 1 (Nov. 15, 2009): 1-127.
48. see Friston and Klaas E. Stephan, "Free-Energy and the Brain," Synthese, clix, 3 (December 2007): 417-58.
49. See also Janneke F. M. Jehee and Ballard, "Predictive Feedback Can Account for Biphasic Responses in the Lateral Geniculate Nucleus," PLoS Computational Biology, v, 5 (May 2009): e1000373.
50. For a review, see Yanping Huang and Rao, "Predictive Coding," Wiley Interdisciplinary Reviews: Cognitive Science, ii, 5 (September/October 2011): 580-93.
51. Clark, "Whatever Next? Predictive Brains, Situated Agents, and the Future of Cognitive Science."
52. Rao and Ballard, "Predictive Coding in the Visual Cortex."
53. see also Rao and Terrence Sejnowksi, "Predictive Coding, Cortical Feedback, and Spike-Timing Dependent Cortical Plasticity," in Rao, Bruno A. Olshausen, and Michael S. Lewicki, eds., Probabilistic Models of the Brain: Perception and Neural Function (Cambridge: MIT, 2002), pp. 297-316.
54. "Predictive Coding Explains Binocular Rivalry: An Epistemological Review," Cognition, cviii, 3 (September 2008): 687-701
55. "Expectation and Surprise Determine Neural Population Responses in the Ventral Visual Stream," Journal of Neuroscience, xxx, 49 (Dec. 8, 2010): 16601-08.
56. see Hinton, "Learning to Represent Visual Input," Philosophical Transactions of the Royal Society B: Biological Sciences, ccclxv, 1537 (Jan. 12, 2010): 177-84.
57. Cathy Price and Joe Devlin, "The Interactive Account of Ventral Occipitotemporal Contributions to Reading," Trends in Cognitive Sciences, xv, 6 (June 2011): 246-53
58. Friston, Jérémie Mattout, and James M. Kilner, "Action Understanding and Active Inference," Biological Cybernetics, civ, 1-2 (February 2011): 137-60.
59. For simplicity, I shall not here pursue the important contribution made by the active embodied agent, but this contribution may be treated in just the same way. A creature's body and self-generated motions are additional hidden causes of sensory variation, and their nature and properties may be unearthed using the same learning routines.
60. See Friston et al., "Action and Behavior: A Free-Energy Formulation," Biological Cybernetics, cii, 3 (March 2010): 227-60.
61. Important computational differences separate the various bodies of work appealed to in section ii. These differences mostly concern the precise ways in which top-down expectations and bottom-up sensory signals are combined, both in learning and during online response. Although significant, these differences may safely be ignored for present purposes.
62. See the review by Huang and Rao, "Predictive Coding"; and the important body of work by Friston and collaborators, usefully summarized in Friston's "The Free-Energy Principle: A Rough Guide to the Brain?," Trends in Cognitive Sciences, xiii,7. (July 2009): 293-301.
63. See Joshua B. Tenenbaum et al., "How to Grow a Mind: Statistics, Structure, and Abstraction," Science, cccxxxi, 6022 (Mar. 11, 2011): 1279-85
64. Charles Kemp, Amy Perfors, and Tenenbaum, "Learning Overhypotheses with Hierarchical Bayesian Models," Developmental Science, x, 3 (May 2007): 307-21.
65. The process level here corresponds to what Marr, in Vision, described as the level of the algorithm.
66. see Feldman and Friston, "Attention, Uncertainty, and Free-Energy"
67. Hohwy, "Attention and Conscious Perception in the Hypothesis Testing Brain."
68. Priors are just prior probabilities, and they can take many forms. In the works cited, they mostly take the form of "probability density functions" or PDFs. Such a function assigns a distribution of probabilities across an uncountably large population, relative to which the observed data are treated as a random sample. In systems that learn hierarchical generative models to explain sensory inputs, probability density functions encode each level's knowledge about the level below. Considered in the most general terms, the role of such PDFs is to enable the system to compute the posterior density, where this names the likelihood of some candidate cause, given the stored knowledge and the current input.
69. see Michael Rescorla, "Bayesian Perceptual Psychology," to appear in Mohan Matthen, ed., The Oxford Handbook of the Philosophy of Perception (New York: Oxford, in press).
70. See Herbert E. Robbins, "An Empirical Bayes Approach to Statistics," in Jerzy Neyman, ed., Proceedings of the Third Berkeley Symposium on Mathematical Statistics and Probability, Volume 1: Contributions to the Theory of Statistics (Berkeley: California UP, 1956), pp. 157-63.
71. See Dempster, Laird, and Rubin, "Maximum Likelihood from Incomplete Data via the EM Algorithm"
72. Neal and Hinton, "A View of the EM Algorithm that Justifies Incremental, Sparse, and Other Variants."
73. The classic critique is that of Jerry A. Fodor and Zenon W. Pylyshyn ("Connectionism and Cognitive Architecture: A Critical Analysis," Cognition, xxviii, 1-2 (March 1988): 3-71)
74. but related points were made by more ecumenical theorists, such as Paul Smolensky ("On the Proper Treatment of Connectionism," Behavioral and Brain Sciences, xi, 1 (March 1988): 1-23)
75. Smolensky and Géraldine Legendre, The Harmonic Mind: From Neural Computation to Optimality-Theoretic Grammar, 2 vols. (Cambridge: MIT, 2006)
76. see Morten H. Christiansen and Nick Chater, "Constituency and Recursion in Language," in Michael A. Arbib, ed., The Handbook of Brain Theory and Neural Networks (Cambridge: MIT, 2003), pp. 267-71.
77. see Tenenbaum et al., "How to Grow a Mind: Statistics, Structure, and Abstraction." Caution is still required, however, since the mere fact that multiple forms of knowledge representation can coexist within such models does not show us, in any detail, how such various forms may effectively be combined in unified problem-solving episodes.
78. Kemp, Perfors, and Tenenbaum, "Learning Overhypotheses with Hierarchical Bayesian Models."
79. th Annual Conference of the Cognitive Science Society (Mahwah, NJ: Lawrence Erlbaum, 2006).
80. Sharon Goldwater, Thomas L. Griffiths, and Mark Johnson, "A Bayesian Framework for Word Segmentation: Exploring the Effects of Context," Cognition, cxii,1 (July 2009): 21-54.
81. Vikash Mansinghka et al., "Structured Priors for Structure Learning," in Proceedings of the 22nd Conference on Uncertainty in Artificial Intelligence (Arlington, VA: AUAI, 2006), pp. 324-31.
82. Griffiths et al., "Categorization as Nonparametric Bayesian Density Estimation," in Chater and Mike Oaksford, eds., The Probabilistic Mind: Prospects for Bayesian Cognitive Science (New York: Oxford, 2008), pp. 303-28.
83. Noah Goodman, Tomer D. Ullman, and Tenenbaum, "Learning a Theory of Causality," Psychological Review, cxviii, 1 (January 2011): 110-19.
84. Elizabeth S. Spelke, "Principles of Object Perception," Cognitive Science, xiv,1. (January-March 1990): 29-56.
85. see Brian J. Scholl, "Innateness and (Bayesian) Visual Perception: Reconciling Nativism and Development," in Carruthers, Stephen Laurence, and Stephen Stich, eds., The Innate Mind: Structure and Contents (New York: Oxford, 2005), pp. 34-52.
86. For example, a system might start with a set of so-called "perceptual input analyzers" (Susan Carey, The Origin of Concepts (New York: Oxford, 2009))
87. see Goodman, Ullman, and Tenenbaum, "Learning a Theory of Causality."
88. see my "Whatever Next? Predictive Brains, Situated Agents, and the Future of Cognitive Science." Finally, there are a variety of questions that arise from recent attempts to extend the predictive coding story so as to encompass action and the observation of action.
89. See for example Friston, Mattout, and Kilner, "Action Understanding and Active Inference."
90. Hinton and Vinod Nair, "Inferring Motor Programs from Images of Handwritten Digits," in Yair Weiss, Bernhard Schölkopf, and John Platt, eds., Advances in Neural Information Processing Systems 18 (Cambridge: MIT, 2006), pp. 515-22.
91. See also Hinton, "To Recognize Shapes, First Learn to Generate Images."
92. Tyler Burge, Origins of Objectivity (New York: Oxford, 2010).
93. see his "On Seeing Things," Artificial Intelligence, ii, 1 (Spring 1971): 79-116).
94. Frank Jackson, Perception: A Representative Theory (Cambridge, UK: University Press, 1977).
95. Helmholtz, Treatise on Physiological Optics, Volume III: The Perceptions of Vision.
96. See also Irvin Rock, Indirect Perception (Cambridge: MIT, 1997).
97. Hohwy, "Functional Integration and the Mind," p. 323.
98. Moore, "The Status of Sense-Data," Proceedings of the Aristotelian Society, xiv (1913-1914): 355-80, reprinted in Moore, Philosophical Studies.
99. Friston, "Beyond Phrenology: What Can Neuroimaging Tell Us about Distributed Circuitry?," Annual Review of Neuroscience, xxv, 1 (2002): 221-50. The quoted passage is from pp. 237-38.
100. see James J. Gibson, The Ecological Approach to Visual Perception (Boston: Houghton Mifflin, 1979) precisely insofar as those accounts seem to reject the explanatory need to appeal to complex internal representational cascades.
101. For some resources, see Tim Crane, "What Is the Problem of Perception?," Synthesis Philosophica, xx, 2 (December 2005): 237-64.
102. Rescorla, "Bayesian Perceptual Psychology."
103. Hinton, "What Kind of a Graphical Model Is the Brain?," in IJCAI-05: Proceedings of the Nineteenth International Joint Conference on Artificial Intelligence (San Francisco: Morgan Kaufmann, 2005), pp. 1765-75.
104. For a detailed account of how this might occur, see Paul C. Fletcher and Chris D. Frith, "Perceiving Is Believing: A Bayesian Approach to Explaining the Positive Symptoms of Schizophrenia," Nature Reviews: Neuroscience, x, 1 (January 2009): 48-58.
105. A. David Smith, The Problem of Perception (Cambridge: Harvard, 2002), p. 224.
106. For some critical discussion of this aspect of Smith's view, see Susanna Siegel, "Direct Realism and Perceptual Consciousness," Philosophy and Phenomenological Research, lxxiii, 2 (September 2006): 378-410.