Legal reasoning through factor-based reasoning and argumentation in the context of explainability

This thesis explores methods for explaining AI decisions, with a particular focus on ensuring users’ rights to appeal such decisions. The central research question guiding this work is: \textbf{How can we ensure the right to appeal an AI-generated decision?} This question is critical for safeguarding fairness and transparency in AI-driven systems, especially in the legal domain, where the outcome can have serious implications for individuals. While the AI Act establishes a 'right to request clear and meaningful explanations' from AI systems involved in decision-making, it stops short of mandating fully transparent, white-box models. This thesis introduces methods to meet transparency requirements by combining symbolic legal models and machine learning techniques to enhance the explainability of AI-driven legal decisions. A series of experiments provide guidance on representing legally relevant factors and establishing logical connections to outcomes, starting with symbolic expert systems for EU and national law compatibility. The thesis extends to machine learning models for classifying legal judgments, balancing transparency with usability. Expert systems are shown to excel in transparency by offering step-by-step reasoning that enhances user understanding, while machine learning models improve accessibility by streamlining interaction. In conclusion, ensuring user rights in AI-driven legal contexts requires clear, comprehensible explanations of legal factors, grounded in both statutory and case law. This thesis emphasizes that AI in legal domains must support rational decision-making, aligning with legal standards and user expectations.