This past decade has also seen the continuing development of digital technologies and the realization of AI and machine learning as practical tools for any number of applications, including law. And yet, the promise and premise of AI are still a bit of a work in progress in law and the legal profession. An August 2019 research paper by Jeffrey M. Lipshaw, a law professor at Boston’s Suffolk University Law School, titled “Lawyering Somewhere Between Computation and the Will to Act: A Digital Age Reflection,” touches on some of that promise and premise: It looks at the type of thinking and analysis that go into the work and practice of law and the capability of AI and machine learning to replicate those activities.

Lipshaw discusses two primary modes of analysis, identified by behavioral psychologists Amos Tversky and Daniel Kahneman, which are characterized as System 1 and System 2. He describes System 1 as “thinking fast”—a more intuitive form of thought that is often based on experience and recognition and sometimes unconscious recognition. While prone to more errors due to its haste and occasional bias, it remains a valued and necessary part of the work of law. System 2’s “thinking slow” is more deliberative and analytical and is based on the conscious (or occasionally unconscious) evaluation of information and analysis of data. The challenge is that while System 2 thinking may lend itself to AI and machine learning, System 1 thinking is more humanistic, relying on intuition as well as insight, which he suggests will be a “far tougher nut” for AI to crack.

Lipshaw offers a couple of examples to illustrate both types of thinking and the current role that AI is and is not playing. System 2 thinking is more along the lines of data-gathering and data recognition. Lawyers are already using AI-enhanced systems to review the documentation—often massive amounts—that might accompany litigation or a large transaction such as a corporate merger. By using “technology-assisted review,” an attorney can identify core and key datapoints that match with both relevant and irrelevant information. The AI process uses this core data to learn which documents are relevant and irrelevant, and Lipshaw reports that AI evaluation often beats humans at retrieving relevant information.

The line in the sand, however, may often be in the simple questions: What will happen? What should I do? The client who asks, “Will I win my case?” is asking not only for a data-driven assessment (analyzing the results of similar cases in the past), but also the insight and intuition that a lawyer brings to the table. Lipshaw uses an illustration involving a lawyer raising a hearsay objection in a trial. The data may suggest that the objection is relevant, but the intuition is to go against the data and not object so as not to frustrate the judge and jury, who are anxious to move on.

CURRENT TOOLS

There are certainly AI tools already in use in the practice of law. The document management tools mentioned previously are relatively commonplace in large law firms and corporate law offices. Legal research is another area in which AI is playing an increasingly common role. Westlaw Edge Quick Check and Casetext’s CARA A.I. both use AI to review documents and suggest additional relevant resources to consider. ROSS uses AI to supplement natural language searching through identifying and recommending additional cases, resources, and secondary research tools in addition to those identified by the search itself.

It’s still a bit early in the AI lifecycle, but it’s fair to ask whether there is more hype than hope. An American Bar Association (ABA) survey (subscription required) conducted in late 2018 found that only about 10% of the responding lawyers used AI-based tech tools in 2018. It also found that large law firms were the most likely users, at 26%, while small law firms ranged from a 3% to a 12% adoption rate. Uncertainty over the benefits of AI was cited as a primary reason for not pursuing AI-based tech tools.

STRENGTHS AND WEAKNESSES

I had the opportunity to use AI-based research tools in a test environment in 2018 and was a bit underwhelmed. The test had me upload a sample legal brief that had researched a fair use in copyright issue. The AI was to report back on additional cases and other resources that were not present in the original brief. The cases and resources were relevant, but not any more or less relevant than those in the original brief.

As I reflect on that experience, I can see the difference between Lipshaw’s System 1 and System 2 thinking. System 2 is the process of identifying or mining all of the available case law for relevant connected data, and the AI system does that well. However, System 1 is the expertise and intuition that I as the researcher and attorney bring to identify the cases that most effectively support the points I’m trying to make. The AI readily handled System 2, but maybe not so much System 1.

It’s still early. I am old enough to have been around at the beginnings of database research, having used Westlaw and LEXIS when I entered law school in 1982—on single-user terminals with 2,400-baud modems. It was more than a decade before they moved off dedicated terminals to browser-based platforms, and then another decade before natural language and keyword searching supplemented Boolean terms and connectors—but by then they had become universally accepted research tools, now even more powerful thanks to AI and machine learning.

The few years since the arrival of AI in the legal world have seen a number of efforts, a few successes, and some failures, but if the adoption of digital research tools is any indication, the development and adoption of AI legal tech will become more and more accepted.