AI seminar: Providing users with explanations in a mixed-initiative system for GUI customization

Speaker: Andrea Bunt

Mixed-initiative approaches to GUI personalization combine aspects of purely adaptive approaches, which rely on AI techniques to automatically personalize the GUI, and adaptable approaches, which place the personalization onus solely on the user. Ideally, in combining aspects of adaptive and adaptable approaches, a mixed-initiative system can leverage each of their respective advantages while minimizing their disadvantages. In this talk I will present our mixed-initiative approach to GUI personalization, focusing on the system's ability to explain its underlying reasoning. In providing users with explanations, the goal is to increase interaction transparency and predictability -- the lack of which are often cited as disadvantages of purely adaptive solutions. An evaluation comparing versions of the system with and without the explanation facility shows a number of individual differences in users' desire to have the explanations available in this context and their impact on users' impressions of the system.