Law, Artificial Intelligence, and Natural Language Processing: A Funny Thing Happened on the Way to My Search Results

Renowned legal educator Roscoe Pound stated, “Law must be stable and yet it cannot stand still.” Yet, as Susan Nevelow Mart has demonstrated in a seminal article that the different online research services (Westlaw, Lexis Advance, Fastcase, Google Scholar, Ravel and Casetext) produce significantly different results when researching case law. Furthermore, a recent study of 325 federal courts of appeals decisions, revealed that only 16% of the cases cited in appellate briefs make it into the courts’ opinions. This does not exactly inspire confidence in legal research or its tools to maintain stability of the law. As Robert Berring foresaw, “The world of established sources and sets of law book that has been so stable at to seem inevitable suddenly has vanished. The familiar set of printed case reporters, citators, and second sources that were the core of legal research are being minimized before our eyes.” In this article I focus on Artificial Intelligence (AI) and natural language processing with respect to searching. My article will proceeds as follows. To understand how effective natural language processing is in current legal research, I go about building a model of a legal information retrieval system that incorporates natural language processing. I have had to build my own model because we do not know very much about how the proprietary systems of Westlaw, Lexis, Bloomberg, Fastcase and Casetext work. However, there are descriptions in information science literature and on the Internet of how systems with advanced programing techniques actually work or could work. Next, I compare such systems with the features and search results produced by the major vendors to illustrate the probable use of natural language processing, similar to the models. In addition, the use of word prediction or type ahead techniques in the major research services are studied--particularly, how such techniques can be used to bring secondary resources to the forefront of a search. Finally, I explore how the knowledge gained may help us to better instruct law students and attorneys in the use of the major legal information retrieval systems. My conclusion is that the adeptness of natural language processing is uneven among the various vendors and that what we receive in search results from such systems varies widely depending on a host of unknown variables. Natural language processing has introduced uncertainty to the law. We are a long way from AI systems that understand, let alone search, legal texts in a stable and consistent way.