Towards Balancing Preference and Performance through Adaptive Personalized Explainability

One-size-fits-all explanations don't work: The research found significant differences in how well people understood AI decisions depending on the explanation format (language, visual maps, or decision trees), suggesting that deploying a single explanation method across all users is ineffective for building genuine understanding.

What people prefer isn't always what helps them understand: A key finding is that user preferences for how AI explains itself don't always match their actual performance—some people preferred explanations that didn't help them comprehend the AI's reasoning as well as alternative formats, highlighting a gap between user satisfaction and actual comprehension.

Personalized explanations improve real performance: The researchers developed an adaptive strategy that tailors explanations to individual users by balancing their stated preferences with their demonstrated understanding, resulting in statistically significant performance improvements (p < 0.05) in their autonomous vehicle study.

Explanation style matters for trust and collaboration: The study emphasizes that how robots and AI systems communicate their decisions directly impacts whether humans can trust them and work effectively with them—particularly important as these systems move from labs into real-world deployment in safety-critical domains like autonomous vehicles.

Diverse user needs require flexible AI systems: Rather than assuming all users benefit from the same explanation approach, responsible AI deployment should recognize and accommodate different communication preferences and cognitive styles to serve diverse populations effectively.