The Measure of Madness: Philosophy of Mind, Cognitive Neuroscience, and Delusional Thought

3 The Processing Hierarchy and the Salience System

3 The Processing Hierarchy and the Salience System 3.1 The Processing Hierarchy The mind is organized as a hierarchical system that uses representations of the world and its own states to control behavior . According to recently influential Bayesian theories of the mind, all levels of the cognitive hierarchy exploit the same principle : error correction (Clark 2012; Jones and Love 2011; Friston 2003). Each cognitive system uses models of its domain to predict its future informational states, given actions performed by the organism. When those predictions are satisfied, the model is reinforced; when they are not, the model is revised or updated, and new predictions are generated to govern the process of error correction. Discrepancy between actual and predicted information state is called surprisal and represented in the form of an error signal . That signal is referred to a higher-level supervisory system, which has access to a larger database of potential solutions, to generate an instruction whose execution will cancel the error and minimize surprisal. (If my serve goes into the net, I must aim higher.) 44 Chapter 3 Surprisal arises when informational states are not predicted by the model. This can range from losing one’s balance to having a marriage proposal rejected or failing to find weapons of mass destruction. In each case, a prediction is not satisfied, and the model that generates the prediction must be revised. (According to prescriptive theory, that is. In practice, as we know, evidence is often ignored, reinterpreted, or discounted.) In this sense, a scientist explaining some discrepant evidence is doing the same thing as the oculomotor system controlling the trajectory of a limb: using and revising a model according to the degree of predictive accuracy it produces. The essence of Bayesian conceptions of cognitive processes, understood as models, can actually be separated from the question of whether human cognitive systems implement Bayes’ theorem —that is, whether Bayesianism is a true theory of the way in which the mind detects and corrects errors. Predictive coding theories of human cognition (Friston 2003; Hohwy, Roepstorff, and Friston 2008) treat Bayesianism as a neurocognitive theory. Much of human cognition is quite automatic. The detection and correction of error occurs at low levels in the processing hierarchy at temporal thresholds and uses coding formats that are opaque to introspection. Keeping one’s balance and phoneme restoration are examples. We have no introspective access to the cognitive operations involved and are aware only of the outputs. This is the sense in which our mental life is tacit: automatic, hard to verbalize, and experienced as fleeting sensations that vanish quickly in the flux of experience. This is the “Unbearable Automaticity of Being” (Bargh and Chartrand 1999). We are not, however, complete automata. At the higher levels of cognitive control, surprisal is signaled in experience or explicit thought: formats available to metacognitive systems The Processing Hierarchy and the Salience System 45 that evolved to enable humans to reflect and deliberate to control their behavior. These metacognitive mechanisms released us from cognitive automatism: control by rigid routines automatically initiated by encounters with the environment (Lieberman et al. 2002; Proust 2006, 2007; Miller and Cohen 2001). Delusions arise at the highest levels in the hierarchy when agents reflect on salient information referred by lower-level systems . Thus, to explain them we need to explain the basic computational architecture of referral and supervision. In particular, we need to explain how information becomes salient; that is to say, becomes the object of cognition at the imprecise borders between controlled and automatic cognition. The mind has evolved a salience system that manages the flow of information throughout the hierarchy, ensuring that cognitive resources are allocated to minimization of the most important categories of surprisal. At the highest levels in the hierarchy, we are aware of this resource allocation as cognitively depleting directed attention and concentration. Resource allocation systems at these high levels of executive control recapitulate very basic mechanisms of salience that evolved to bias automatically controlled behavior (Berridge and Robinson 1998, 2003; McClure, Daw, and Montague 2003). In order to explain how these mechanisms work, we need a cognitive theory that displays the relevant properties of the salience system at all levels, from automatic and reflexive to deliberate and reflective. That theory is provided in the form of a distinction between weight- and activation-based processing in neural networks, which allows us to show how the neural mechanisms of salience...