The Machine Learning Curve

Research is the lifeblood that fuels legal practice. But as litigators, we balance on the knife's edge between our duty to provide excellent service to our clients and the financial realities of case management. On the one hand, we owe our clients and the court our best work product supported by airtight legal precedent. But, on the other hand, we cannot simply bill our clients into oblivion each time we draft a motion by following every thread across every research database we can access. The costs of searching across these platforms can be staggering, as are the attorney fees to pay for unfortunate junior associates tasked with finding an elusive silver bullet case with the perfect quotable holding.

This is why we, the frugal and tech-oriented IP litigators that we are, did a double-take at a recent post in the Law Librarian Blog: "A judge capped the costs award ... writing that the use of artificial intelligence should have 'significantly reduced' counsel's preparation time. ... citing both research fees as well as other aspects of the lawyers' bill." The case, Cass v. 1410088 Ontario Inc., 2018 ONSC 6959 (Nov. 22, 2018), appeared at first blush to be an ordinary Canadian occupier's liability personal injury costs judgment. But Justice A.C.R. Whitten of the Ontario Superior Court of Justice felt that counsel overstepped the bounds of appropriate research for the case in calculating their costs, and posited a novel justification for reducing the award: "All in all, whatever this 'research' was would be well within the preparation for the motion. There was no need for outsider or third-party research. If artificial intelligence sources were employed, no doubt counsel's preparation time would have been significantly reduced. ... Therefore, as a starting point the disbursements claim should be reduced by $11,404.08." (Emphasis added.)

Justice Whitten did not elaborate on this reasoning, but it was the first instance we had come across of a judge reducing a costs award for not using artificial intelligence to streamline legal research. And, to the best of our own research (which includes the use of artificial intelligence), we did not find any similar holdings in the United States.

But could our juridical cousins north of the border be signaling the future of American legal practice? Over past decade, "artificial intelligence" has grown from a mere buzzword into an array of real tools applicable to nearly every aspect of contemporary legal practice. Generally accepted as a subfield of computer science, AI was defined in one scholarly article as "a system's ability to correctly interpret external data, to learn from such data, and to use those learnings to achieve specific goals and tasks through flexible adaptation." Most lawyers use some form of AI in their everyday practice, from reducing document review through de-duplication, to keyword searches on their favorite research platform. We humans can make "basic" AI word searches more effective -- that is, more likely to reveal results closer to what we want to see -- by applying our own brainpower to craft Boolean or natural-language search strings. But search results across various platforms using these human-generated searches can differ greatly even with identical search input. See Susan Nevelow Mart, "The Algorithm as a Human Artifact: Implications for Legal [Re]Search," 109 LAW LIBR. J. 387 (2017); "Results May Vary," A.B.A. J., Mar. 2018, at 48. In addition, even despite these attorneys' best efforts, judges report that attorneys miss relevant cases in their briefing. Developers at emerging AI legal research companies such as ROSS Intelligence and Casetext are attempting to prevent such "misses." For example, the newest tools allow attorneys to upload litigation documents such as briefs and complaints in toto, and AI uses the entire document to springboard its search. AI reads the document and generates a list of cases, briefs and statements of law that are relevant to the facts and issues contained therein. Users can even layer keyword searches on top of their document uploads to focus the AI to pull cases from specific areas of law.

We had the opportunity to try out one of these emerging AI research tools, Casetext, in writing this article. We uploaded a sample legal brief into Casetext's CARA AI search tool and watched as it spat out cases with similar facts, laws and policies to those underlying the brief and the cases cited therein. Our research librarians explained excitedly that CARA had denoted one of the results to be a case with controlling negative precedent that had not been included in the brief, which indicates the brief's drafters either had not found the case or intentionally omitted it. This had us envisioning the many potential uses for such technology. Could we hamstring both an opponent's legal bases and credibility with the court in a matter of seconds using such search technology? Could we upload patents? License agreements? Technical documents? Are there different use cases on ROSS Intelligence or the other emerging AI research platforms? As with all computer systems, these platforms are only as good as the data they have been provided. But as more information is uploaded into the systems' knowledge bases, the use cases for these technologies will only continue to expand. Casetext, for example, reports that its performance has improved by 67 percent since 2015, updating a previous study using 14 diverse queries. Heller, "What a difference a few years makes: The rapid change of legal search technology," Casetext (Mar. 7, 2018). In short, the technology, its accuracy, and the possibilities for its use are on the rise.

As is to be expected, the firms who explore these technologies as they evolve likely will be those with the budget for it. However, even with a potentially steep initial investment, it is predicted that practitioners will see reductions to their overall expenses as AI takes the heavy lifting out of costly research and other time-intensive aspects of legal practice. William J. Connell, "Artificial Intelligence in the Legal Profession -- What You Might Want to Know," 66-JUN R.I. B.J. 5 (2018). We do not foresee Cass v. 1410088 Ontario Inc. being the harbinger of judicially enforced cost cutting through AI in the United States -- though as with dinosaurs, drive-ins and the dodo, the era of analog research is coming to a close. The real-world advantages of AI-assisted research may render traditional research avenues prohibitively slow, and future crops of law school graduates may scoff at our complex Boolean strings the same way we scoffed at our predecessors' predilection for book research. Let this article be your canary in the coal mine. Get to know these and other emerging AI technologies and their applications to your practice sooner rather than later. Our profession is set on an exciting course, and the future is already ripe for plucking today.