Well-Being AI Encouragement in Sustained Interaction: Attitude-Aware Design Principles

Abstract

Encouraging utterances is widely used in supportive dialogue systems; however, how this should be designed for Well-Being AI remains underexplored. The Technology Acceptance Model (TAM) is sometimes interpreted as suggesting a simple, monotonic relationship in which more positive attitudes toward AI lead to higher perceived effectiveness of AI-mediated support. In our preliminary study with older Japanese adults examining AI-delivered encouragement, we measured participants’ baseline attitudes toward AI and collected perceived effectiveness ratings of encouragement strategies across multiple strategy types and worry contexts. This study provides preliminary indications that perceived effectiveness may not be well explained well by a simple linear assumption regarding users’ attitudes toward AI. Based on these observations, this paper proposes five design principles for encouragement in Well-Being AI, with a focus on adapting encouragement type selection based on users’ attitudes toward AI. Our key ideas include reframing “user attitudes toward AI” from a pre- and post-interaction evaluation measure to a dynamic signal that should be continuously recognized and responded to during interaction. Encouragement strategy selection and realization should be guided by users’ cur-rent attitudes and interaction feedback. In this process, users’ evolving attitudes reshape the system’s behavior, while the system’s responsive support in turn influences users’ subsequent attitudes. This bidirectional adaptation informs the design of emotionally supportive inter-actions and may contribute to the co-evolution of human and machine intelligence in Well-Being AI.