Transferable XAI: Relating Understanding Across Domains with Explanation Transfer
This new framework could finally make AI explanations reusable and practical.
Researchers have proposed 'Transferable XAI,' a new framework that allows users to transfer their understanding of one AI model's explanations to another related model. It uses a general affine transformation to map explanations across different domains, tasks, or attributes. In user studies, it outperformed single-domain explanations, leading to better decision faithfulness and factor recall. This addresses the key problem of users overgeneralizing or independently memorizing explanations for every new AI application they encounter.
Why It Matters
It could dramatically reduce the learning curve and increase trust when deploying multiple, related AI systems in real-world settings.