Design and evaluation of AI-based implicit interactions

The rapid advancements in Artificial Intelligence (AI) have transformed how we design and interact with intelligent systems. Among these advancements, implicit interaction systems, capable of seamlessly adapting to user behaviour and context without explicit commands, have emerged as a promising frontier in Human-Computer Interaction (HCI) research. This thesis examines the potential and challenges of AI-based implicit interactions, considering both user and designer perspectives, as well as the evaluation of such interfaces. First, we introduce the concept of implicit interaction, outlining our interpretation and its relevance within the broader HCI landscape. Then, the first part of this thesis explores the driving and parking context as a testbed for implicit interaction systems, investigating how smartphones can be leveraged as sensing devices to infer user behaviour without requiring dedicated external infrastructure. Specifically, we present and discuss how we designed and implemented implicit interactions for driving-related tasks, such as detecting when a driver cruises for parking and predicting parking availability using aggregated contextual data. These studies provide insights into how implicit interaction can enhance mobility-related user experiences by minimizing explicit input while seamlessly adapting to behavioural and environmental conditions. The second part focuses on the challenges associated with designing and evaluating AI-based implicit interactions. A key issue is the lack of established guidelines. To address this, we propose a design and evaluation framework with a particular emphasis on error handling, examining how implicit systems convey uncertainty and enable users to recover from incorrect inferences. Finally, we discuss how AI techniques, specifically based on Large Language Models (LLMs), can be integrated into interface evaluation to improve the design process. Our findings highlight the trade-offs between automation and user control, providing guidelines for developing usable, user-centred systems. By addressing these challenges, this research contributes to the broader field of HCI proposing methodologies for developing intelligent, context-aware interfaces. While the driving domain is a concrete application, the principles and techniques outlined in this thesis have broader implications, and they can be adapted to other domains where implicit interaction can enhance usability and user experience.