Liability in motion: A retrieval augmented generation application in road traffic liability apportionment

The rapid advancement of transformer-based large language models (LLMs) has opened promising avenues for their application in increasingly complex legal domain tasks. This study introduces a retrieval-augmented generation (RAG) framework specifically tailored to predict liability allocations in traffic accident cases, utilizing an extensive dataset derived from a prominent German legal commentary. The proposed model is distinct from conventional classification methods in that it addresses the unique challenge of accurately predicting nuanced legal outcomes, such as proportional liability shares. The research systematically evaluates and compares several state-of-the-art language modeling techniques, providing a comprehensive overview of recent developments in this field. The findings underscore the considerable potential of incorporating advanced language modeling approaches into judicial decision-making processes, supporting a complementary role for AI through a close integration of legal precedent, conceptual frameworks, and reasoning processes. This work offers critical insights into both the strengths and limitations of RAG-based LLM applications in judicial contexts, laying a robust foundation for future research on AI and Law.