Content area
Full text
1. Introduction: Prediction machines
1.1. From Helmholtz to action-oriented predictive processing
“The whole function of the brain is summed up in: error correction.” So wrote W. Ross Ashby, the British psychiatrist and cyberneticist, some half a century ago.1 Computational neuroscience has come a very long way since then. There is now increasing reason to believe that Ashby's (admittedly somewhat vague) statement is correct, and that it captures something crucial about the way that spending metabolic money to build complex brains pays dividends in the search for adaptive success. In particular, one of the brain's key tricks, it now seems, is to implement dumb processes that correct a certain kind of error: error in the multi-layered prediction of input. In mammalian brains, such errors look to be corrected within a cascade of cortical processing events in which higher-level systems attempt to predict the inputs to lower-level ones on the basis of their own emerging models of the causal structure of the world (i.e., the signal source). Errors in predicting lower level inputs cause the higher-level models to adapt so as to reduce the discrepancy. Such a process, operating over multiple linked higher-level models, yields a brain that encodes a rich body of information about the source of the signals that regularly perturb it.
Such models follow Helmholtz (1860) in depicting perception as a process of probabilistic, knowledge-driven inference. From Helmholz comes the key idea that sensory systems are in the tricky business of inferring sensory causes from their bodily effects. This in turn involves computing multiple probability distributions, since a single such effect will be consistent with many different sets of causes distinguished only by their relative (and context dependent) probability of occurrence.
Helmholz's insight informed influential work by MacKay (1956), Neisser (1967), and Gregory (1980), as part of the cognitive psychological tradition that became known as “analysis-by-synthesis” (for a review, see Yuille & Kersten 2006). In this paradigm, the brain does not build its current model of distal causes (its model of how the world is) simply by accumulating, from the bottom-up, a mass of low-level cues such as edge-maps and so forth. Instead (see Hohwy 2007), the brain tries to predict the current suite of cues from its best models of the possible...