AI and Legal Reasoning

This paper presents a summary of the responses of a panel to issues on AI and legal reasoning. The panel consisted Among the issues addressed by the panel were: 1. What are the characteristics of the legal domain that make it interesting or amenable to AI approaches-what is special about it; 2. The open-textured nature of legal concepts and the implications this has for using Al-techniques, especially knowledge representation; 3. The complementarity of rule-based and case-based reasoning-how cases are used, especially when the rules ''run out"; 4. The pervasive role of analogy in legal reasoning; 5. The special role played by hypothetical! in the legal domain and how hypos help with argumentation and strategic case planning; 6. The interleaving of justification, explanation, and argumentation; 7. How common law systems can be seen to be systems which learn from cases; 8. The appropriateness and feasibility of intelligent aids for practicing litigators and other legal experts; 9. Implications for other domains-like medicine-that use case-based reasoning; 10. Methodological and other issues. For each issue considered, the comments of the panelists are summarized. 1. The Challenge and Special Characteristics of the Legal Domain The legal domain presents some very interesting challenges to the AI researcher. While it is a domain which has established standards for deriving new truths (e.g., stare decisis or the doctrine of precedent), it is more of a "scruffy" domain than a "neat" one, despite its orderly, rule-like surface veneer. It is very much an experience-based example-driven field. Legal reasoning is also heavily intertwined with natural language processing and common sense reasoning and therefore inherits all the hard problems that these imply. Several panelists emphasize that legal reasoning and argumentation take special skills and that learning to think like a lawyer requires considerably more than rote memorization of a large number of cases, a daunting task in itself. For instance, Dyer says: Modeling what a lawyer does is more complex than modeling experts in technical/scientific domains. First, all of these complex conceptualizations are expressed in natural language, so modeling the comprehension ability of the lawyer requires solving the natural language problem. For example, giving legal advice often starts with hearing a "story" where the client was one of the actors. So legal advice often presupposes a story understanding capability. In other expert systems, the natural language problem can be largely finessed since the task (e.g., disease diagnosis, reconfiguring hardware, analyzing dipmeters) …