AI in the Legal Sector

I first started researching the use of artificial intelligence (AI) in the legal space in mid-2017, when my consulting firm was commissioned by a client to conduct a study of its usage by law firms. I began as we often do—with a news search that produced about 150,000 results. I dug into several articles and industry pieces in an attempt to understand the challenges and opportunities that this emerging technology was creating and then began conducting interviews with law firms. After interviewing 22 leading firms (more about the definition of “leading” and the results of that research later) on their use of AI, I completed my research and submitted the results to the client late last year.

My partners and I gleaned so much information on the application of AI in the legal sector that we decided to make it the subject of our June session at the Somerville Forum. This intimate, one-of-a-kind forum is an invitation-only gathering of leaders within the legal industry who are actively involved with accelerating innovation.

About the Somerville Forum on AI

The Somerville Forum was named after Mary Somerville, a leading writer, astronomer, and mathematician of the 19th century. Somerville was known for her clear and adventurous thinking and for her remarkable skill at translating the arcane science writing of her age. Like Mary, we try to look across borders and disciplines to find insights into the 21st century client-centric firm, but also find practical ways to apply those insights into our world.

The June session was structured to include talks on robotics and the use of technology in law firms along with a summary of my report on the prevalence of client-facing AI within a certain segment of law firms. To allow our members (most of whom are neither engineers nor data scientists) to engage in a meaningful discussion on AI, we had to level-set its definition. When Rick McFarland, Chief Data Officer for LexisNexis®, and former Chief Data Scientist for the Hearst organization and Director of Data Engineering for Amazon.com, kicked off the session, he had our full attention.

Pulling Back the Curtain

Mr. McFarland defines AI as the perception of intelligence that we humans project on computer programs or technology. “This perception of intelligence,” he said, “is actually created by our developers with nothing more than a computer program following a master script, accessing various databases and specialized subroutines, or ‘cognitive engines’.”  To illustrate his point, Rick took us through the construction of a basic chatbot.

In practice, most chatbot processes start with the computer recognizing the language in which the user’s question is asked.  Once it confirms the language matches that of the program and databases, it then converts the human question into a “bag of words”: basically it parses the question and dumps “useless words” that computers don’t need.  It then passes this bag through a series of specialized subroutines, or “cognitive engines” (as he calls them), that extract key concepts and entities which it uses to determine how to best anwser the user’s question. If the program needs to tap into a database, it converts the bag into a database query, finds the answer, and presents this output to the user via the chatbot interface.  McFarland explained that most AI processes depend on a sequence of requests to these cognitive engines, but that the overall chatbot experience is “only as smart as its weakest engine.” This can present a problem in a profession like the Law which demands the responses to the questions be error-free. “There’s a super high bar in Law and Medicine,” he said. “You can never be wrong when large stakes are involved. That’s what makes creating professional-grade AI difficult and why we don’t see a lot of chatbots in doctors and lawyers offices…yet.”

Clearly, the cognitive engines in the process are central to the functionality of the latest AI technologies like visual recognition, chatbots, and voicebots (e.g. Amazon Alexa).  To construct the very best cognitive engines, Rick explained, you need a combination of two things: data and expertise.  Access to a richly annotated corpus of training data is essential and often difficult to access in the case of highly specialized subjects. Equally as important is assembling a team with the required expertise to construct these engines from the data.  For deeply specialized areas, the best teams include a combination of subject matter experts (e.g. lawyers, doctors, etc.) and AI developers known as data scientists who are expert in utilizing machine-learning methods like natural language processing, image recognition, computational linguistics, and deep learning.

Once we understood how AI programs work (and fail), Rick gave a demo with the Amazon Alexa device and we took turns asking “her” some questions. She responded correctly to our general knowledge questions like “Who is the head of the supreme court?” or “What is the Second Amendment?”  However, as we started to get into specific legal questions that required domain-specific or “post-graduate”-level knowledge as Rick described it, she was unable to answer correctly, for example: “What is the statute of repose for product liability in the state of Illinois?” or “Who is the chief justice of the United States Court of Appeals for the Ninth Circuit?”  Matters got even worse for her as we added in legalese or questions that mixed English and Latin words (e.g. “What is the definition of res judicata?”).

Somerville members felt that Rick’s talk and demo helped to demystify some of the concepts behind the construction of AI while broadening minds to the possible applications of such programs once the considerable challenges were overcome. Our members asked: If it were a matter of developing the right set of programs with the proper string of commands why weren’t we further along?  The answer, besides the high-bar set by lawyers, ultmately boils down to the data. As with any discipline, if one wants to predict outcomes accurately, we need a significant precedent set and a large amount of error-free and unbiased data upon which the cognitive engines can be developed on.

Khang Pham, Product Manager for InterAction®, demonstrated how they have taken large sets of CRM data and trained InterAction to aid in duplicate management. The final “a-ha” moment for the group was the acknowledgement that the intelligence we use every day relies on programs that have been decades in the making.

Where Are We Today?

That’s not to say firms haven’t made progress. After Rick concluded his presentation, I took the group through the public portion of my AI research and findings. I interviewed the CIOs, Directors of IT, Directors of KM, and “Innovation Partners” at 22 law firms.[1] At many firms, I spoke with two representatives as the firms had divided responsibilities between Innovation Partners, IT, and KM.

In reality, even the most innovative firms are in the early stages of developing AI as we envision it. The elite firms we surveyed are, for the most part, engaging in efforts to understand the technology, creating pilots of various software and developing a philosophy about how and when such technology should be incorporated and deployed within their organizations.

The difference between AI philosophy and deployment by the elite firms vs. the other firms is not as significant as one might be led to believe by the marketing efforts of those firms and the articles and reports landing on our desk on a daily basis.[2] Complicating the landscape further is the fact that many technology vendors and law firms are defining the term “artificial intelligence” broadly enough to include technology that has been available and in use for many years.[3] According to some, such work would not be considered AI; according to others, these kinds of tools and programs fit squarely within their firm’s definition of AI. Putting aside the disagreements on definition, the trend towards experimenting with and employing these technologies on client-facing work is increasing at a decent rate.

Currently, law firm use of AI is mainly limited to tasks where it would be beneficial to review large volumes of information for similar or anomalous attributes. Some areas of practice and tasks lend themselves more readily to this automation (e-discovery, for example). Adoption by firms is higher and occurred further back than has been the case for other areas (such as due diligence or contract review in the M&A context).

While the use cases and success stories across these firms tend to center around efficiencies gained by the use of AI, the philosophies and the marketing of AI capabilities were markedly different among those firms that view themselves on the AI cutting-edge from those that see it primarily as an efficiency tool. Ironically, although the former group may indeed be slightly ahead, it is likely the firm attitude and commitment that will help them advance more quickly and gain real marketing advantage from their efforts.

This is not to say that all firms are falling neatly into predictable groups of behavior. Some firms surveyed were experimenting with the use of programs that drafted documents; some were teaming up with universities to develop programs and technologies; some were teaming up with clients on projects; some were looking at programs that performed tasks and provided explanation for the choices it made.

Some of these same firms spent considerable time thinking about and experimenting with different pricing models that took into account the change in the way that portion of the legal work was being executed and delivered. (If the thousands of hours for due diligence are gone for good, then what is the fair price for the use of the technology as a result?)

And some firms were doing very little.

As we discussed these findings, the in-house counsel, in particular, were interested in better understanding which programs were employed and where firms were finding their biggest successes. There were nuanced differences among each firm interviewed. But the universal truth was that firms relied on lawyer champions. Projects moved forward when lawyers pushed. Priorities were more often dependent upon the lawyers raising their hands and less so on firm strategy. Narrowly-scoped projects gained more traction and paid benefits sooner than larger, firm-wide projects. Some firms reported intense client pressure to develop technological solutions.

The biggest successes came when individual lawyers felt compelled—through their own burning desire or client pressure—to solve a specific problem, and firms provided the expertise (either through a computer scientist or engineer) to help develop the technology that provided the solution. As Rick McFarland noted earlier in the day, you don’t generally find a lawyer who is also a computer scientist.

The Mythical Unicorn

We were treated to a demonstration by one such lawyer: Van Lindberg, an intellectual property partner at Dykema Cox Smith, who describes himself as an engineer who went to law school. Van got his Bachelor of Science in Computer Science and History at Brigham Young University (BYU), and his JD at BYU Law School. After law school, he was assigned to patent work with a lot of repetition. He wrote proprietary programs that helped eliminate the manual work and allowed him to focus on the portion of the work that required legal judgment.  As a result of his passion and expertise, he developed a method of analyzing a company’s patent portfolio at a level of detail and sophistication that amazed every in-house counsel in the room.

And that’s when you know you’ve got something real. Each of them recognized something that would be helpful to them today.

While Van may be the anomaly today, Nextlaw Labs or Allen & Overy’s Fuse are developing the complementary competencies that will allow them to create combinations of professionals that add up to Van’s equivalent and allow for technological solutions and breakthroughs.

Conclusion

Many of the firms I first interviewed last year, as well as others we’ve learned about, have made significant investments, including financial support, technical staff, and lawyer hours in the hopes of advancing the programs available for use on client-facing work. The implications of those investments, coupled with pressure from clients to reduce costs and gain efficiencies, signals a shift in how legal services and advice will be delivered, including how the leverage model, and the recruitment and training of young talent will evolve over the next five years.

Unlike the multitude of reports that fill our inboxes daily, I am not predicting a tectonic shift, but rather a gradual, rising sea level. However, firms that do not begin experimenting and investing now may find themselves behind within a year or two, mainly because thought leadership will have evolved more quickly at those firms making the investment. Even in the time between the date of the forum in late June and the release of this article, we are learning of more firms engaging in new ventures and experiments in the use of AI in client-facing work. Law firms would do well to pay particular attention to how AI can be part of the solution to help them bond even more closely with current clients as well as attract new clients. After all, every firm is looking for advantages in business development. Why turn your back on a rising tide?

[1] All of the 22 firms were AmLaw 50 or Global 100 firms. By status, there were seven National firms, four Global firms, six New York firms, three Magic Circle firms, and two California firms. By economic rank, ten of the firms were top tier, seven were one tier lower, and five were in lower strata, each with first tier practices.

[2] As I began to prepare this article, I was curious how much the coverage had increased, so I repeated the search that originally yielded 150,000 news results in July 2017. Today, I was greeted by 296,000 results. More impressive than the sheer increase in coverage is the weightiness of its contents: In-depth reports released by PwC, Deloitte, ALM; announcements of new or expanded start-up legal techs; launches of legal automation financings and investments. In the last week alone, we have witnessed articles predicting:

1.) significant decreases in the number of lawyer positions within law firms;
2.) significant increases in jobs in the legal sector;
3.) significant decreases in support roles;
4.) significant automation of in-house functions; and
5.) significant investments by venture capitalists and other industries

– all as a result of legal AI and related technology.

[3] Rather than attempting to define AI, I allowed the participants to define it for themselves.

“For more than 20 years, Yolanda Cartusciello has served in senior administrative leadership roles in major law firms, including Debevoise & Plimpton and Cleary Gottlieb. At both firms, she led the marketing teams, designed their business development and media strategies and took charge of their implementation. She was the chief architect of profile enhancement strategies, perception studies, branding exercises, comprehensive client interview programs, and practice and lateral partner rollouts. She co-developed marketing technology solutions and created media relations and digital strategies. She has also developed business development and communications training and coaching programs for lawyers at all levels. In addition, Yolanda has long experience handling organizational and personnel issues. She has advised on the development of the legal assistant, knowledge and practice management, and legal marketing staff roles at various firms. She hired, trained, and supervised more than 150 administrative employees for work in high-achieving cultures. Yolanda has a B.A. from the University of Iowa and an M.F.A. from Brooklyn College. She lives with her husband and daughter in Brooklyn.”

AI in the Legal Sector posted first on https://injuryhelpnowcom.blogspot.com