Modeling Reliance on XAI Indicating Its Purpose and Attention

Transparency about AI purpose has mixed effects: Simply telling people why an AI system exists doesn't automatically build trust—the impact depends on whether users are meant to rely on the AI for decisions or just comply with its recommendations, and on how difficult the task is.

Visual explanations can backfire on complex tasks: When heat maps showing where an AI focuses its attention were displayed for difficult tasks, they actually decreased trust rather than increasing it, suggesting that unclear visual explanations may confuse users more than help them.

Explanation quality matters more than explanation presence: Heat maps only improved trust when users could actually understand and interpret them clearly, particularly on harder tasks—simply adding visual explanations without ensuring they're comprehensible may be counterproductive.

Task difficulty changes how explanations work: The same explanation (purpose statement or heat map) had different effects depending on whether the task was easy or difficult, indicating that one-size-fits-all approaches to AI transparency won't work across different use cases.

Different AI uses need different transparency strategies: The study distinguishes between "reliance" (depending on AI for decisions) and "compliance" (accepting AI recommendations), and shows these require different approaches to building appropriate trust through explanations.