Context Matters: The Challenges of Aligning Legal Justification with Explainable AI

ABSTRACT In view of the global interest in deploying artificial intelligence technology to enhance efficiency in public administration, one issue which seems to have received less attention is the importance of context sensitivity. A large part of public administration is regulated by law and concerns the legal rights and obligations of individual citizens. Legal decision-making in public administration takes place under the rule of law requirement, that such decisions must be accompanied by reasons that explain the decision. In this paper we show that justificatory explainability should not be confused with technological explainability (often referred to as XAI). Instead, justificatory explainability is a requirement that decisions in law are based on reasons that explain how the facts of a case relate to the legal rule that is applied in the case. The synthesis of fact and law in relation to the individual decision being made requires situational awareness of the context in which the rule operates. AI systems that are being introduced to interact with legal decision-making processes in public administration must be designed in a way which ensures that such context sensitivity is not erased from the decision-making process.

KEYWORDS AI in public administration - Legal explainability - Justificatory reasoning in law - Context-sensitivity - Rule of law - Automated decision-making

TABLE OF CONTENTS 1. Introduction: Justificatory Explainability and Situational Awareness. – 2. A Starting Point: The Need for Situational Awareness in Legal Decision-Making. – 3. The Black Box Problem and the Distinction between Causal Explanation and Justificatory Explanation. – 4. Legal Machines: Three Different Approaches. – 4.1. Fixed Rule Systems. – 4.2. Expert Systems. – 4.3. Large Language Models and Retrieval Augmented Generation. – 5. The Legal Requirement for Public Administration to Give Reasons. – 6. Elements in Justificatory Explanations for Legal Decisions. – 6.1. Identifying and Interpreting Legal Rules. – 6.2. The Role of Factual Information in Justifying Legal Decisions. – 7. (X)AI for Justificatory Explainability. – 7.1. Fixed Rule Systems. – 7.2. Expert Systems. – 7.2.1. Case- Based Reasoning Systems. – 7.2.2. Rule-Based Reasoning Systems. – 7.3. Large Language Models and Retrieval Augmented Generation. – 8. Conclusion: The Need for Further Research on How AI Can Support Justificatory Explainability.