This research relates to the usability of mixed-initiative interaction systems, in which actions can be initiated either through a choice by the user or through intelligent decisions taken by the system. The key issue addressed here is how to preserve the user's perceived control ("sense of agency'') when the control of the interaction is being transferred between the system and the user in a back-and-forth manner.

Previous research in social psychology and cognitive neuroscience suggests timing is a factor that can influence perceived control in such back-and-forth interactions. This dissertation explores the hypothesis that in mixed-initiative interaction, a predictable interaction rhythm can preserve the user's sense of control and enhance their experience during a task (e.g. higher confidence in task performance, stronger temporal alignment, lower perceived levels of stress and effort), whereas irregular interaction timing can have the opposite effect. Three controlled experiments compare alternative rhythmic strategies when users interact with simple visual stimuli, simple auditory stimuli, and a more realistic assisted text labelling task. The results of all three experiments support the hypothesis that a predictable interaction rhythm is beneficial in a range of interaction modalities and applications.

This research contributes to the field of human-computer interaction (HCI) in four ways. Firstly, it builds novel connections between existing theories in cognitive neuroscience, social psychology and HCI, highlighting how rhythmic temporal structures can be beneficial to the user's experience: particularly, their sense of control. Secondly, it establishes timing as a crucial design resource for mixed-initiative interaction, and provides empirical evidence of how the user's perceived control and other task experiences (such as reported levels of confidence, stress and effort) can be influenced by the manipulation of timing. Thirdly, it provides quantitative measures for the user's entrainment behaviours that are applicable to a wide range of interaction timescales. Lastly, it contextualises the design of timing in a realistic application scenario and offers insights to the design of general end-user automation and decision support tools.