This post is based on a paper written by Andy Clark, author of Surfing Uncertainty (See Paper Predictive Processing for a fuller treatment.), “A nice surprise? Predictive processing and the active pursuit of novelty,” that appeared in Phenomenology and the Cognitive Sciences, pp. 1-14. DOI: 10.1007/s11097-017-9525-z. For me this is a chance to learn how Andy Clark has polished up his arguments since his book. It also strikes me as connected to my recent posts on Curiosity and Creativity.
Clark and Friston (See post The Prediction Machine) depict human brains as devices that minimize prediction error signals: signals that encode the difference between actual and expected sensory simulations. But we know that we are attracted to the unexpected. We humans often seem to actively seek out surprising events, deliberately seeking novel and exciting streams of sensory stimulation. So how does that square with the idea of minimizing prediction error.
One way to explore this worry is via the so-called ‘Darkened Room’ scenario (See Post The Dark Room Problem–Minimizing Surprise). The idea is that we could best predict in an environment where not much is going on. Previously, Clark has dismissed this as not fitting the kind of creature that we are. More sophisticated dark room scenarios have evolved.
Clark notes that even in the dark room, there is plenty of prediction error. We tend to forget the stream of information specifying (via dense vascular feedback) the physiological state of the body. That includes the state of the gut and viscera, blood sugar levels, temperature, and a great deal more. What happens when a unified multi-level prediction engine crunches all that interoceptive information together with exteroceptive information specifying states of affairs in the world? As your bodily states alter, the importance of various worldly opportunities alters too. Such estimations of importance are written deep into the predictive processing model, where they appear as alterations to the weighting (the ‘precision’) of specific prediction error signals. As those estimations alter, you will act differently, harvesting different streams of exteroceptive and interoceptive information, that in turn determine subsequent actions, choices, and bodily states.
Clark suggests that the most genuinely challenging version of the Darkened Room worry is the ‘Merely Modest Exploration Trap’. This idea suggests that we should shift our attention from the minimization of prediction error to the maximization of mutual information (hence the maximization of prediction success).
Clark notes that compelling empirical evidence for such a profile includes work by a series of experiments with infants measuring attention to sequences of events of varying (and well-controlled) complexity. Infant attention was characterized by a ‘Goldilocks Effect’, focusing upon events presenting an intermediate degree of predictability—neither too easily predictable, nor too hard to predict. The probability of an infant looking away was thus greatest when complexity was either very high or very low. The functional upshot is that ‘infants implicitly seek to maintain intermediate rates of information absorption and avoid wasting cognitive resources on overly simple or overly complex events’. Such tendencies to seek out ‘just-novel-enough’ situations are a good candidate for some form of innate specification, since they would cause active agents to self-structure the flow of information in ways ideally suited to the incremental acquisition and tuning of a rich and informative generative model of their environment.
Clark rejects this by first noting that any purely information theoretically specifiable goal will be subvertable by the right set of environmental conditions. According to Clark such an agent could be ‘hijacked’ by a darkened room containing only a computer monitor and an endless supply of simple (but not too simple) puzzles, each of which allows a steady trajectory of improvement until the next one pops up on the screen.
Such subversion does not, however, typically occur. Clark believes that It is our cultural practices themselves that conspire to render us humans so deeply exploratory and that immunize us against information-theoretic subversion of even the ‘merely modest’ kind. But as a kind of spin-off from all this deep (and language-enriched ) model-building activity, we humans also began to construct complex social and physical environments. Our human minds are now deeply exposed to the unique statistical baths of a succession of such designer worlds –worlds characterized by the complex traditions and practices of art, science, recreation, and literature. Those practices reliably spawn an open-ended set of new local goals and projects (Clark suggests that we think Pokemon Go). The skilled pianist has learnt to reduce prediction error with respect to complex melodies and motor repertoires.
Predictive agents immersed in these kinds of designer environment learn to value (assign high precision to) states that make available wide ranges of new moves and outcomes, enabling scientists and designers to discover ideas that bring rewards that register in basic forms of neural circuitry. Our changing cultural practices thus piggyback upon ancient reward systems in ways that enable human populations to continuously seek out new ideas and perspectives.
Clark explains that most of the heavy lifting, when it comes to explaining the shape and nature of the modern mind, is thus be done by our peculiar social and intellectual histories – histories replete with chance and path-dependent unfolding.