On the Impact of Explanations on Understanding of Algorithmic Decision-Making

Explanations alone don't guarantee understanding: The study found that simply providing explanations of high-risk algorithmic decisions doesn't automatically lead to meaningful comprehension. Understanding is complex and multifaceted, requiring consideration of how people actually form and evaluate knowledge rather than just presenting information.

Different explanation formats produce different results: The research tested three explanation modalities (textual, dialogue, and interactive) and found that dialogue-based explanations increased participant engagement and interaction, allowing researchers to better assess what people actually understood about the algorithmic system.

Understanding involves both rational and emotional dimensions: Using the "six facets of understanding" framework, the study revealed that people's comprehension of algorithmic decisions includes not just logical reasoning but also emotional and intuitive responses—suggesting that effective explanations must address both dimensions.

Individual differences matter significantly: The research indicates that people's personal characteristics and backgrounds affect how they understand and perceive algorithmic fairness, meaning one-size-fits-all explanations may be insufficient for diverse stakeholders affected by these systems.

Current XAI research may miss important factors: The paper suggests that explainable AI (XAI) research often overlooks established theories about how humans actually learn and understand, highlighting a gap between technical explanation design and educational psychology principles.