A Comparative review of Symbolism and Connectionism in AI&Law: Tracing evolution and exploring integration

Abstract

AI&Law explores computational methods for automating legal reasoning and prediction, evolving in parallel with AI research and developing along two primary paths: symbolic and connectionist approaches. Symbolic AI&Law centers on the formal representation of legal concepts and performing reasoning based on statutes and case law. These methods have led to the development of rule-based and case-based reasoning systems, successfully implemented in legal expert systems. The primary advantage of symbolic approaches is their inherent explainability, although they face limitations due to the knowledge acquisition bottleneck. Connectionist AI&Law encourages legal professionals to adopt inductive inference and use “bottom-up” learning models to extract hidden features from large datasets. This paradigm incorporates machine learning and natural language processing (NLP) techniques to address legal information extraction, retrieval, text classification, summarization, and legal prediction tasks. The advent of large language models (LLMs) has further expanded the capabilities of connectionist models, enabling more sophisticated legal text analysis and predictive accuracy, though issues of model transparency and hallucination remain active areas of research. The interaction between symbolic and connectionist approaches can complement each other. Symbolic models can enhance the transparency and explainability of connectionist systems, while connectionist techniques can optimize the scalability and efficiency of symbolic reasoning processes. These two paradigms exhibit strong potential for collaboration, particularly in the domains of explainable dialogue systems, neuro-symbolic systems, legal knowledge embedding and legal argumentation mining, etc.