A Survey of Explainable AI and Proposal for a Discipline of Explanation Engineering

Explanations are not one-size-fits-all: The paper emphasizes that "what an explanation really is" requires careful definition, suggesting that different stakeholders (regulators, patients, engineers) may need different types of explanations from the same AI system.

Multiple XAI methods exist but require domain-specific selection: The survey identifies a taxonomy of popular XAI approaches and demonstrates their application across finance, autonomous driving, healthcare, and manufacturing, indicating that the most appropriate explainability technique depends on the specific industry and use case.

Explainability should be engineered into systems from the start: Rather than treating explanation as an afterthought, the paper proposes "Explanation Engineering" as a systematic discipline for designing explainability into AI systems from inception, suggesting this is more effective than retrofitting explanations later.

High-stakes domains demand tailored approaches: The focus on critical applications like healthcare (where patients need to understand treatment recommendations) and autonomous driving (where safety decisions must be justified) highlights that explainability is particularly crucial when AI systems affect human welfare and safety.

The field is maturing toward practical implementation: By moving beyond theoretical XAI methods to propose a structured engineering discipline, the paper signals that explainable AI is transitioning from academic research toward practical, systematic deployment in real-world systems.