Predictive Processing
The MIT Press eBooks · 2025
Predictive Pro cessing Minimal Predictive Pro cessingThe introduction of predictive pro cessing (PP) in the philosophy of mind and cognitive science can be traced to the work of Jakob Hohwy (2013) and Andy Clark (2013, 2016), both of whom pick up Karl Friston's (2003, 2005, 2010) proposal for how predictive coding might be used to provide a generalized theory of brain functioning (2005), said proposal, in turn, having been influenced by Rao and Ballard's (1999) account of hierarchical predictive coding in the visual cortex.While predictive coding is the most basic component of PP, it is not distinct to PP-indeed its origins are not in neuroscience but data compression, as a strategy developed as a means for the storage and transmission of image and video files during the 1950s (Clark, 2016; for an overview, see Shi & Sun, 1999; Musmann, 1979).The basic idea is that there are typically regular patterns in the data that we wish to store or send, and, thus, rather than encoding the value of each pixel individually, we can encode an image more efficiently by only encoding this pattern and its occasional violations.In a video, for instance, large areas of background often remain unchanged over some duration, so, rather than retransmitting the entire scene anew for each frame, we can simply transmit the pattern once, then only encode the subsequent "errors" induced by local movements of objects and agents in front of this background.The proposal that the brain uses such a coding strategy, as Sprevak (2021) describes, dates back at least to Attneave (1954) and Barlow (1961), who argued that bottlenecks in the early visual system-for instance, the number of neurons, their dynamic range, limitations on firing rate, and the metabolic costs of firing-require the brain to use such "redundancy-reducing"