ExplainabilityTransparencyTrustworthinesscs.HC
Towards a Comprehensive Human-Centred Evaluation Framework for Explainable AI
Ivania Donoso-Guzmán, Jeroen Ooge, Denis Parra, Katrien Verbert
July 31, 2023
xAI
14 citations
1 influential
Key Takeaways
-
Current XAI evaluation lacks standardization: The paper identifies that while explainable AI techniques are rapidly advancing, there are no standardized procedures for evaluating how well these explanations actually work for real people—a critical gap for responsible AI deployment.
-
Explanations affect user experience holistically: The research emphasizes that XAI evaluation should not treat explanations in isolation, but rather assess their complete impact on how users understand and interact with AI systems, similar to how user experience is evaluated in recommender systems.
-
A comprehensive framework is needed: The authors propose adapting evaluation methods from recommender systems to create a unified framework that integrates multiple explanation aspects, clarifies relationships between different explanation properties, and organizes metrics for measuring them—providing a roadmap for more consistent XAI assessment.
-
Implications for responsible AI: By establishing standardized, human-centered evaluation procedures, this framework helps ensure that explainable AI systems are genuinely useful and trustworthy for end-users, rather than merely appearing transparent without meaningfully improving human understanding or decision-making.
Abstract
While research on explainable AI (XAI) is booming and explanation techniques have proven promising in many application domains, standardised human-centred evaluation procedures are still missing. In addition, current evaluation procedures do not assess XAI methods holistically in the sense that they do not treat explanations' effects on humans as a complex user experience. To tackle this challenge, we propose to adapt the User-Centric Evaluation Framework used in recommender systems: we integrate explanation aspects, summarise explanation properties, indicate relations between them, and categorise metrics that measure these properties. With this comprehensive evaluation framework, we hope to contribute to the human-centred standardisation of XAI evaluation.