An optimal control approach to testing theories of human information processing constraints

This thesis is concerned with explaining human control and decision making behaviours in an integrated framework. The framework provides a means of explaining human behaviour in terms of adaptation to the constraints imposed by both the task environment and the information processing mechanisms of the mind. Some previous approaches tended to have been polarised between those that have focused on rational analyses of the task environment, on the one hand, and those that have focused on the mechanisms that give rise to cognition on the other hand. The former usually is based on the assumption that rational human beings adapt to the external environment by achieving ‘goals’ defined only by the task environment and with minimal consideration of the mechanisms of the human mind; while the latter focuses on information processing mechanisms that are hypothesised to generate behaviour, e.g., heuristics, or rules. In contrast, in the approach explored in this thesis, mechanism and rationality are tightly integrated. One interpretation of the framework decomposes the modelling problem into two parts: the state estimation problem and the optimal control problem. This thesis investigates a state estimation and optimal control approach, or optimal control approach for brevity, in which human behavioural strategies and heuristics, rather than being programmed into the model, emerge as a consequence of rational adaptation given a theory of the information processing constraints. The thesis takes problems, each at different level, from visual perception (Chapter 4: the Distractor Ratio task) through immediate behaviour (Chapter 5: a Menu Search task), to probabilistic inference (Chapter 6: a information gathering and decision making task) and uses the optimal control approach to obtain the optimal policy; the behaviour of which is then compared to human behaviour. The results show that (1) constraints imposed by human peripheral vision give rise to colour and letter search phenomena; (2) constraints imposed by peripheral vision, the statistics of the environment, and knowledge give rise to menu search phenomena; (3) constraints imposed by the information validities ranking discovery and the information cost give rise to certain decision making phenomena. In each case, previous models, requiring the modeller to provide control rules or heuristics, are replaced with an optimal control model in which behaviour emerges from constraints. Furthermore, the results show that the optimal control approach meets three challenges: (1) the challenge of finding emergent control strategies, rather than programming them in the model (2) the challenge of breadth of application and, (3) the challenge of testing different information processing theories by deriving optimal strategies. Subsequently, the discussion considers the broader role of the optimal control approach in cognitive science. In particular, it focuses on the issue of whether or not people are optimal and the different roles of optimal state estimation and optimal control. It also, considers the possibility of integrated models of visual search and decision making.