Bridging the Gap Between Explainability and Large Language Models
Year: 2025
Type: preprint
Abstract: The rapid advancement of Large Language Models (LLMs) has revolutionized artificial intelligence applications, enabling unprecedented capabilities in natural language understanding, generation, and decision-making. However, the increasing complexity and opacity of these models pose significant challenges in interpretability, trust, and accountability. Explainable AI (XAI) has emerged as a critical research area to address these concerns by providing insights into model predictions, enhancing user trust, and ensuring compliance with ethical and regulatory standards. This paper explores the intersection of explainability and LLMs, presenting an overview of state-of-the-art XAI techniques such as SHAP, LIME, attention-based interpretability, counterfactual explanations, and causal inference. We analyze real-world applications of XAI in healthcare, finance, legal AI, and education, demonstrating how explainability fosters transparency and fairness in high-stakes domains. Furthermore, we discuss key challenges in XAI for LLMs, including scalability, causal reasoning, multimodal interpretability, and human-centered explanations. Finally, we outline future research directions, emphasizing the need for hybrid explainability frameworks, interactive user interfaces, and fairness-aware models. By bridging the gap between black-box LLMs and interpretable AI, this study contributes to the development of trustworthy, ethical, and responsible AI systems.
Source: HAL
Authors: Shui, Xia, Ru, Zhihao
Institution: Chinese University of Hong Kong
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W4415144282