The Hierarchies and Systems that Underlie Routine Behavior: Evidence from an Experiment in Virtual Gardening

Previous behavioral research into the processing mechanisms that support action selection in complex sequential tasks has focused on errors or ‘slips of action’ (Norman, 1981; Reason, 1979, 1984). This work has been taken to suggest that sequential behavior is controlled by the dynamic interaction of two systems – a resource hungry “supervisory” system and an automatic “routine” system that employs hierarchically structured task representations (Norman & Shallice, 1986). To counter some of the shortcomings of purely error-based natural history studies we developed an alternative methodology which enables more controlled, direct and finegrained investigation of the underlying processes, namely the use of action selection latencies within novel, computerbased, tasks. While the methodology is relatively general, the primary aim of the specific experiment reported here was to test the hypotheses that routine behavior is governed by (a) hierarchically structured task representations, and (b) the dynamic interaction of two systems. The experiment was additionally designed to investigate the impact of several factors on the processing difficulty of specific action selection steps, including experience, number of choices, availability of external disambiguation cues, and minor variations within subsequences. Our results provide some direct support for the Dual Systems framework, but also pose challenges with regard to further refinements and computational models.