See Also: Book Notes, (me), Notes on Consciousness, Ravenous Brain, Dynamic Global Workspace, Brain Region Network, Consciousness and the Brain, Touching a Nerve, Righteous Mind, Consciousness: Confessions, The Quest for Consciousness, Happiness Hypothesis, Blank Slate, Info Viz & Perception, On Intelligence, The Stuff of Thought, Neuroscience of Human Relationships, Human: Makes Us Unique, Thinking, Fast and Slow

Surfing Uncertainty:

Prediction, Action, and the Embodied Mind

Andy Clark

(2015) Oxford University Press, ISBN 9780190217013

The main thrust of this book is that the brain is a prediction machine (PP == Prediction Processor) and that when the sensory input does not match the prediction, the difference is the interesting thing.

The hierarchical Prediction machines run at difference timescales. Your limb level prediction machines parse interoception and proprioception sensory streams and run at a high rate to minimize energy in dealing with the physical world. Above those might be mid level action planners that run a process to achieve simple goals, like walk to the refrigerator.

Jeff Hawkins describes the cortical algorithm which might be a building block for the PP. See On Intelligence

presence is possible by suppressing interoception.
The language takes some time to unpack.

Direct quotes from the book


Our neural economy exists to serve the needs of embodied action. It does so by initiating and sustaining complex circular causal flows in which actions and perceptions are co-determined and co-determining. These flows enact structural couplings that serve our needs while keeping the organism within its own specialized window of viability. All this is orchestrated, or so our story suggests, by a multilevel generative model tuned to predict task-salient aspects of the current sensory signal.


Words, we might say, are (for us language users) a metabolically cheap and flexible source of 'artificial contexts' (Lupyan & Clark, in press). Viewed from the PP perspective, the impact of strings of words upon neural processing is thus flexibly to modify both what top-down information is brought to bear, and how much influence it has at every level of processing (see Figure 9.1).

PP == Prediciton Processing

[jch: p.277 . . . Human culture provides]

. . . at multiple spatial and temporal scales, to the statistical structure of the action-ready, organism-salient world, as reflected in the training signals. But those training signals are now delivered as part of a complex developmental web that gradually comes to include all the complex regularities embodied in the web of statistical relations among the symbols and other forms of sociocultural scaffolding in which we are immersed. We thus self-construct a kind of rolling 'cognitive niche' able to induce the acquisition of generative models whose reach and depth far exceeds their apparent base in simple forms of sensory contact with the world.

To see how this might work, recall that the way to construct a new idea or concept (assuming the resources of PP) is to encounter a new sensory pattern that results in highly weighted (organism-salient) prediction error. Highly weighted errors, if the system is unable to explain them away by recruiting some model that it already commands, result in increased plasticity and (if all goes well) the acquisition of new knowledge about the shape and nature of the distal causes responsible for the surprising sensory inputs. But we humans are also expert at deliberately manipulating our physical and social worlds so that they provide new and ever-more-challenging patterns that will drive new learning.

A very simple example is the way that learning to perform mental arithmetic has been scaffolded, in some cultures, by the deliberate use of an abacus. Experience with the sensory patterns thus made available helps to install appreciation of many complex arithmetical operations and relations (for discussion, see Stigler et al., 1986). The specific example does not matter very much, but the general strategy does. We structure (and repeatedly restructure) our physical and social environments in ways that make available new knowledge and skills (for some lovely explorations, see Goldstone, Landy, & Brunel, 2011; Landy & Goldstone, 2005; and, for an information-theoretic twist, Salge, Glackin, & Polani, 2014).

Prediction-hungry brains, exposed in the course of embodied action to novel patterns of sensory stimulation, may thus acquire forms of knowledge that were genuinely out-of-reach prior to such physical-manipulation-based retuning of the generative model.

Such retuning and enhancement is now served by a huge variety of symbol-mediated loops into material and social culture: loops that involve notebooks(1), sketchpads, smartphones, and also written and spoken conversations with other agents(2). *1 (see Clark, 2003, 2008) *2 (see Pickering & Garrod, 2007)

Some other great tidbits, and some gripes to be added 2016.05.15   YON   Book Notes