The idea that the brain is basically an engine of prediction is one that will, I believe, turn out to be very valuable not just within its current home (computational cognitive neuroscience) but across the board: for the arts, for the humanities, and for our own personal understanding of what it is to be a human being in contact with the world.
The term 'predictive coding' is currently used in many ways, across a variety of disciplines. The usage I recommend for the Everyday Cognitive Toolkit is, however, more restricted in scope. It concerns the way the brain exploits prediction and anticipation in making sense of incoming signals and using them to guide perception, thought, and action. Used in this way 'predictive coding' names a technically rich body of computational and neuroscientific research (key theorists include Dana Ballard, Tobias Egner, Paul Fletcher, Karl Friston, David Mumford, and Rajesh Rao) . This corpus of research uses mathematical principles and models that explore in detail the ways that this form of coding might underlie perception, and inform belief, choice, and reasoning.
The basic idea is simple. It is that to perceive the world is to successfully predict our own sensory states. The brain uses stored knowledge about the structure of the world and the probabilities of one state or event following another to generate a prediction of what the current state is likely to be, given the previous one and this body of knowledge. Mismatches between the prediction and the received signal generate error signals that nuance the prediction or (in more extreme cases) drive learning and plasticity.
We may contrast this with older models in which perception is a 'bottom-up' process, in which incoming information is progressively built (via some kind of evidence accumulation process, starting with simple features and working up) into a high-level model of the world. According to the predictive coding alternative, the reverse is the case. For the most part, we determine the low-level features by applying a cascade of predictions that begin at the very top; with our most general expectations about the nature and state of the world providing constraints on our successively more detailed (fine grain) predictions.
This inversion has some quite profound implications.
First, the notion of good ('veridical') sensory contact with the world becomes a matter of applying the right expectations to the incoming signal. Subtract such expectations and the best we can hope for are prediction errors that elicit plasticity and learning. This means, in effect, that all perception is some form of 'expert perception', and that the idea of accessing some kind of unvarnished sensory truth is untenable (unless that merely names another kind of trained, expert perception!).
Second, the time course of perception becomes critical. Predictive coding models suggest that what emerges first is the general gist (including the general affective feel) of the scene, with the details becoming progressively filled in as the brain uses that larger context — time and task allowing — to generate finer and finer predictions of detail. There is a very real sense in which we properly perceive the forest before the trees.
Third, the line between perception and cognition becomes blurred. What we perceive (or think we perceive) is heavily determined by what we know, and what we know (or think we know) is constantly conditioned on what we perceive (or think we perceive). This turns out to offer a powerful window on various pathologies of thought and action, explaining the way hallucinations and false beliefs go hand-in-hand in schizophrenia, as well as other more familiar states such as 'confirmation bias' (our tendency to 'spot' confirming evidence more readily than disconfirming evidence).
Fourth, if we now consider that prediction errors can be suppressed not just by changing predictions but by changing the things predicted, we have a simple and powerful explanation for behavior and the way we manipulate and sample our environment. In this view, action is there to make predictions come true and provides a nice account of phenomena that range from homeostasis to the maintenance of our emotional and interpersonal status quo.
Understanding perception as prediction thus offers, it seems to me, a powerful tool for appreciating both the power and the potential pitfalls of our primary way of being in contact with the world. Our primary contact with the world, all this suggests, is via our expectations about what we are about to see or experience. The notion of predictive coding, by offering a concise and technically rich way of gesturing at this fact, provides a cognitive tool that will more than earn its keep in science, law, ethics, and the understanding of our own daily experience.