Explainable Artificial Intelligence (XAI): Concepts, Taxonomies, Opportunities and Challenges toward Responsible AI

Explainability is a critical barrier to AI deployment: The paper identifies explainability as an "inherent problem" of modern AI techniques like deep neural networks and ensemble methods, and argues it is "acknowledged as a crucial feature for the practical deployment of AI models" in real-world applications.

Different audiences need different explanations: The paper emphasizes that explainability must be tailored to its intended audience, establishing a novel definition of explainability that prioritizes "the audience for which explainability is sought" rather than using a one-size-fits-all approach.

Deep learning requires specialized explanation methods: The authors found that explaining deep learning models presents unique challenges distinct from other machine learning approaches, warranting a separate taxonomy of explanation techniques specifically designed for these complex neural network systems.

Explainability is foundational to responsible AI: The paper connects XAI directly to the broader concept of "Responsible Artificial Intelligence," suggesting that transparency and explainability are essential components for implementing AI ethically and fairly at scale in organizations.

Data fusion and explainability create unresolved tensions: The paper identifies a significant challenge at "the crossroads between data fusion and explainability," indicating that combining multiple data sources while maintaining clear, understandable AI explanations remains an open problem in the field.