Action Selection and Action Execution in Human Learning

Abstract

Intelligent behavior requires knowing both what to do in a given situation, and how to do it. Knowledge of the requisite whats and hows in life is not acquired through a single process — often a variety of distinct neural systems guide an organism through the learning process. For illustration, consider a child first learning to tie her shoes: This relatively simple behavior requires a confluence of factors that each draw on different neural mechanisms, including, but not limited to, a declarative memory system that recalls the crucial instructions her mother gives her, a reinforcement system that rewards successful attempts and thus reinforces the corresponding maneuvers, and a calibration system that coordinates, maintains, and updates the necessary muscle commands to account for an ever-changing environment and body (e.g. a new pair of laces, cold fingers, etc.). The cooperation of these various systems is vital for the acquisition of skills.

The first stage of any non-reflexive motor behavior is defined by volition: What do I want to do? I’ll refer to this stage as action selection. Investigations of selection processes will be presented in the context of simple, discrete value-based decisions (Chapter 2), and more complex sensorimotor decisions (Chapters 3 and 4). After selection of an action, the motor system must then attempt to successfully carry out that action. I’ll refer to this process as action execution. Execution is characterized by the accuracy of the selected action, and represented in the neural activity that ultimately leads to repeatable spatio-temporal patterns of coordinated muscle activation. Execution will be discussed in the context of reaching movements. The overarching theme of this thesis is the interaction of selection and execution systems during human reinforcement and sensorimotor learning.