Editable XAI: Toward Bidirectional Human-AI Alignment with Co-Editable Explanations of Interpretable Attributes

Researchers have introduced 'Editable XAI,' a framework that allows users to directly edit an AI's explanations and rules to better align with their own knowledge. Their system, CoExplain, uses a neural network and symbolic rules to create a collaborative, editable decision tree. In a user study with 43 participants, this editable approach significantly improved user understanding and model alignment compared to traditional read-only explanations, requiring fewer edits and less time to use effectively.