Explainable artificial intelligence: A survey of needs, techniques, applications, and future direction

2024-06-01

Neurocomputing

AI Summary

The deployment of AI models in safety-critical domains—such as healthcare, finance, and autonomous driving—is hindered by their lack of explainability. Method: This paper presents a systematic survey of eXplainable Artificial Intelligence (XAI), uniquely integrating *user-centric* (demand-side) and *model-centric* (supply-side) perspectives to establish a four-dimensional evaluation framework—encompassing transparency, faithfulness, practicality, and fairness—and to identify critical gaps between methodological development and real-world deployment. It rigorously analyzes core XAI techniques—including surrogate modeling, attention visualization, counterfactual explanation, causal reasoning, and formal verification—characterizing their applicability boundaries and limitations. Contribution/Results: The study produces an XAI practice atlas spanning 12 application domains and defines explanation priority criteria for high-risk scenarios. These findings provide foundational theoretical and methodological support for XAI standardization, regulatory policy formulation, and cross-domain adoption.

Technology Category

• Humans and AI: Explainable AI (XAI) for Human Understanding
• Philosophy and Ethics of AI: Accountability, Interpretability & Explainability
• Machine Learning: Transparent, Interpretable, Explainable ML

Application Category

• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Problem

• Explainable AI
• Transparency
• Decision Clarity

Innovation

• Interpretable AI
• XAI Methodology
• Comprehensive Review