Exploration of Temporal Aspects of AI Explanations

The increasing integration of AI into high-stakes domains such as healthcare, finance, and transportation highlights the need for XAI tofoster appropriate trust and collaboration between humans and AI systems. While previous research has explored the interpretabilityand effectiveness of different XAI methods, the temporal aspects of AI-generated explanations remain underexplored. This paperinvestigates how temporal constraints influence human understanding and decision-making when interacting with AI-generatedexplanations. Specifically, we investigate the impact of explanation timing on user trust, agreement with AI decisions, and responsetimes. We use a controlled user experiment setup with Where’s Waldo puzzles to simulate decision-making under time pressure.Participants are presented with AI-generated statements regarding Waldo’s presence in a picture, with some receiving additionalexplanations via the Local Interpretable Model-Agnostic Explanations (LIME) framework. We investigate how AI confidence scoresinfluence user confidence and decision latency. We hypothesize that time constraints may lead to cognitive biases that affect theeffectiveness of AI explanations. Our findings will provide insights into the design of AI explanations that support users in bothtime-sensitive and deliberative decision-making contexts