Your litigation analytical tool says your win rate for summary judgement motions in class action employment discrimination cases is ranked the best in your local jurisdiction according to the database used. Forget the problem with using PACER data for litigation analytics, possible modeling error or possible bias embedded in the tool. Can you communicate this applied AI output to a client or potential client? Are you creating an “unjustified expectation” that your client or potential client will achieve the same result for your next client matter?
According to the ABA’s Model Rules of Professional Conduct Rule 7.1, you are probably creating an “unjustified expectation.” However you may be required to use that information under Model Rule 1.1 because that rule creates a duty of technological competence. This tension between Model Rule 7.1 and Model Rule 1.1 is just begining to be played out.
For more, see Roy Strom’s The Algorithm Says You’ll Win the Case. What Do You Say? US Law Week’s Big Law Business column for August 5, 2019. See also Melissa Heelan Stanzione, Courts, Lawyers Must Address AI Ethics, ABA Proposal Says, Bloomberg Law, August 6, 2019.
From the abstract for Charlotte Alexander and Mohammed Javad Feizollahi On Dragons, Caves, Teeth, and Claws: Legal Analytics and the Problem of Court Data Access, (Computational Legal Studies: The Promise and Challenge of Data-Driven Legal Research (Ryan Whalen, ed., Edward Elgar, 2019, Forthcoming).
This chapter provides a case study of data access challenges in a legal analytics project that attempted to study all U.S. district court judges’ decisions in employee misclassification disputes over a ten-year period. The chapter details the data assembly process, problems, and workarounds, and considers the implications for legal analytics and computational law.
Early results from 25% of the AmLaw 200 participating so far in a Feit Consulting survey indicate that the adoption rate of Westlaw Edge and Context by LN is roughly the same, trending at 15%. “Context seems to be getting much more consideration, however, because of its much lower cost. At this point 40% of firms with Lexis are actively considering Context,” according to Feit Consulting’s blog post.
My primary concern is that comparing Westlaw Edge and Context because both offer litigation analytics may only be part of the story. Westlaw Edge offers much more than just the litigation analytics offered by Context; Westlaw Edge includes WestSearch Plus, KeyCite Overruling Risk, Statutes Compare and Regulations Compare. And Westlaw Edge will eventually replace Westlaw whereas Context will not replace Lexis Advance.
From the blurb for Kevin D. Ashley, Artificial Intelligence and Legal Analytics: New Tools for Law Practice in the Digital Age (Campridge UP, 2017):
The field of artificial intelligence (AI) and the law is on the cusp of a revolution that began with text analytic programs like IBM’s Watson and Debater and the open-source information management architectures on which they are based. Today, new legal applications are beginning to appear and this book – designed to explain computational processes to non-programmers – describes how they will change the practice of law, specifically by connecting computational models of legal reasoning directly with legal text, generating arguments for and against particular outcomes, predicting outcomes and explaining these predictions with reasons that legal professionals will be able to evaluate for themselves. These legal applications will support conceptual legal information retrieval and allow cognitive computing, enabling a collaboration between humans and computers in which each does what it can do best. Anyone interested in how AI is changing the practice of law should read this illuminating work.
Trust is a state of readiness to take a risk in a relationship. Once upon a time most law librarians were predisposed to trust legal information vendors and their products and services. Think Shepard’s in print when Shepard’s was the only available citator with signals that were by default the industry standard. Think late 1970s-early 1980s for computer-assisted legal research where the degree of risk taken by a searcher was partially controlled by properly using Boolean operators when Lexis was the only full-text legal search vendor.
Today, output from legal information platforms does not always result in building confidence around the use of the information provided be it legal search or legal citator outputs as comparative studies of each by Mart and Hellyer have demonstrated. What about the output we are now being offered by way of the implementation of artificial intelligence for legal analytics and predictive technology? As legal information professionals are we willing to be vulnerable to the actions of our vendors based on some sort of expectation that vendors will provide actionable intelligence important to our user population, irrespective of our ability to monitor or control vendors’ use of artificial intelligence for legal analytics and predictive technology?
Hopefully we are not so naive as to trust our vendors applied AI output at face value. But we won’t be given the opportunity to shine a light into the “black box” because of understandable proprietary concerns. What’s needed is a way to identify the impact of model error and bias. One way is to compare similar legal analytic outputs that identify trends and patterns using data points from past case law, win/loss rates and even a judge’s history or similar predictive technology outputs that forecast litigation outcome like Mart did for legal search and Hellyer did for citators. At the present time, however, our legal information providers do not offer similar enough AI tools for comparative studies and who knows if they will. Early days… .
Until such time as there is a legitimate certification process to validate each individual AI product to the end user when the end user calls up specific applied AI output for legal analytics and predictive technology, is there any reason to assume the risk of using them? No, not really, but use them our end users will. Trust but (try to) validate otherwise the output remains opaque to the end user and that can lead to illusions of understanding.