A critical engagement with Clark's account of the role of motivation in a Bayesian information processing model of the brain.

Abstract

The action selection problem can be approached with two goals in mind: to account for why actions are selected, i.e. what factors play a role in making the decision, and how actions are selected, i.e. what mechanisms are involved in action selection. These two goals form the main focus of this dissertation and can be referred to as the efficiency problem and the architecture problem. This dissertation examines the Environmental Complexity Hypothesis as an approach to the former goal (why actions are selected). The Environmental Complexity Hypothesis adequately explains why actions are selected but lacks consideration of the architecture and mechanisms involved in complex cognitive systems. In response to this shortcoming, the predictive processing account is offered as a solution to the second problem of action selection, the architecture problem. The predictive processing account will be critically examined and one objection will be raised against its claim about the function of cognition. According to the predictive processing account, the function of cognition is to minimize free energy. One objection to this claim takes the form of a reductio ad absurdum argument and suggests that if the predictive processing account is correct about the function of cognition, then biological agents will find themselves in the state of seeking environments with no sensory stimuli. This is biologically nonsensical because environments with no sensory stimuli will result in no surprise and no prediction error which, the account claims, needs to be reduced for successful adaptation. However, reducing prediction error alone does not result in adaptive fitness and is not the sole function of the cognitive system; agents explore and participate in the environment to reproduce and to satisfy desires and preferences as well. This view is reflected in the field of neuroeconomics which provides a fairly novel solution to the problem of action selection and decision making. Neuroeconomic models of action selection can provide insight about why agents select certain actions when alternatives are available and it also gives an account of the architecture involved in action selection. It is proposed that prediction error is coded in relation to rewards. By recommending neuroeconomics as an alternative account of action selection, nothing is lost in either the Environmental Complexity Hypothesis or the predictive processing account. Instead, neuroeconomics represents the best features of the two theories in a unified account.