Could (Should) Legal Reasoning be Mechanized with the Help of AI Legal Expert Systems?

Could (Should) Legal Reasoning be Mechanized with the Help of AI Legal Expert Systems?

As I was applying to the Osgoode IP Intensive, I came across the CodeX placement at Stanford Law School, in partnership with the Department of Computer Science. CodeX has several projects and one of them is to advance legal technology with an emphasis on the research and development of computational law.

Computational law is the branch of legal informatics involved with the automation and mechanization of legal analysis. As advancements in technology and increases in computing power continue, legal technological systems may be used for the purposes of legal reasoning, and even in a judicial capacity, which can drastically change the nature of the legal profession, improve the legal system as a whole and even enhance access to justice. In fact, in the United States, the Internal Revenue Service (IRS) has for several years been using legal technology to render administrative decisions concerning areas of taxation, which were previously handled by the human personnel. In Canada, automated decision-making is being used to recommend prison classification and conditions for inmates.

Nevertheless, even with the positive impact these new legal systems may generate, I was still left with questions pertaining to whether we should use legal expert systems that mechanize the legal reasoning process; and if we could - or most importantly should - mechanize legal reasoning, particularly in an extremely complex regulatory and jurisdictional environment where several laws and jurisdictions are always in play. What is the effect of automating legal reasoning on the field of legal jurisprudence? Do AI legal systems have the capability to apply appropriate legal reasoning and, most importantly, make decisions adapted to deal with important policy changes over time?

Firstly, current expert systems based on computational logic, which represents regulations as sentences in formal logic and mechanical reasoning techniques, are then applied to the facts at hand to derive consequences from the application of logic to the facts. However, in order to conduct a proper legal analysis, an expansive multi-factorial analysis that includes several variables must be conducted. Expansive legal analysis may not be properly met with the current foundations of the legal informatics and computational law.

Philosophically, computational law exists within the legal formalism school of jurisprudence.  Legal formalism is the idea that all questions of policy should be made by the legislature alone and we should enforce what the law actually says, rather than what it should say. Basically, legal formalism is all about applying a set of rules and principles, independent of the social trends and politics. One legal formalist scholar, H.L.A. Hart, even stated that law cannot be divorced from the concept of a rule. However, legal formalism may promote objective and consistent decision-making.

Legal formalism, which most legal AI systems use, contrasts with legal realism, which promotes discretion in legal decision making. This approach balances the interests of the affected parties on a case-by-case basis. Legal formalism is easier and more relevant to apply in Civil Law systems, as the legal rules are applied more literally, with less room for discretion. However, it may be not suitable to apply it in Common Law systems, where decision-makers have discretion to apply laws with consideration of complex social, economic, technological, and global trends, in order to meet the changing needs of society. Strict application of legal formalism may result in neglecting societal values. Justice Cardozo once stated that “The final course of the law is the welfare of society. The rule that misses its aim cannot permanently justify its existence”. Perhaps, with further advancements in artificial intelligence, these expert systems will be also capable of applying complex policy rationales to generate fairer outcomes. It will be very interesting to see how computational logic could be used to code complex non-binary social policies.

Written by Elif Babaoglu, Contributing IPilogue Editor and JD Candidate at Osgoode Hall Law School with a strong interest in AI and law. Elif is also the co-director of events at the Osgoode Privacy Law Society.