Incremental XAI: Memorable Understanding of AI with Incremental Explanations

Incremental explanations improve how people remember AI decisions: The research found that by breaking down explanations into a consistent "base" set of factors plus additional "incremental" factors for unusual cases, users better retained and recalled how the AI system worked compared to traditional explanation methods.

Current XAI methods force users to choose between accuracy and simplicity: The paper identifies a key problem—existing explainable AI techniques either show oversimplified global explanations that aren't fully accurate, or show highly variable local explanations that are hard to follow and remember across different decisions.

Reusing common factors helps people build mental models: By showing the same foundational factors repeatedly and only adding new factors when needed for atypical cases, Incremental XAI reduces cognitive load and helps users develop a more coherent understanding of how the AI system makes decisions.

The approach was validated through multiple types of user testing: The researchers conducted formative studies (to design the method), user studies (to test memorability and understanding), and summative evaluations to confirm that Incremental XAI improved both how well people understood AI explanations and how accurately they could remember them later.

This work addresses a practical usability gap in AI transparency: Rather than just making explanations more technically interpretable, this research focuses on how to present explanations in ways that align with how humans actually learn and remember information.