UNC, MIT Study Probes AI Threats to Big Law

UNC, MIT Study Probes AI Threats to Big Law Vladgrin

Recent advances in artificial intelligence have led many law firm leaders to conclude that computers will replace more and more junior lawyers over the coming decades, with employment gradually hollowing out from the bottom up.

Such predictions may be off base, if a new academic paper, “Can Robots Be Lawyers?,” is correct. The draft paper, authored by Dana Remus, a professor currently on leave from the University of North Carolina School of Law, and Frank Levy, an urban economics professor emeritus at Massachusetts Institute of Technology, examines specific ways that automation might or might not be applied to a range of legal tasks. The study finds that while many tasks may be automated, most legal work is too complex—and too important—for even the most advanced machines to learn and replicate.

A few years ago, Levy said, he started growing skeptical about claims by advocates of artificial intelligence that AI would inevitably replace even highly skilled white-collar workers. “There was a general feeling among economists that AI writing was full of spin and hyperbole,” said Levy, who has long been interested in the effects of computers in the labor market. He reached out to Remus, a legal ethics expert who had studied the adoption of predictive coding technology by law firms.

“We wanted to demystify the artificial intelligence stuff and to get into the weeds so that interested lawyers could understand how all the new software works,” Levy said.

To estimate how automation is likely to affect law firm employment, the two analyzed $2.3 billion in detailed billing records submitted by large firms to clients with the help of Sky Analytics, a billing analytics firm, to determine how much time firms currently spend on a variety of legal tasks. If the firms were to widely adopt new automation technologies, they found, overall employment could drop by 13 percent in five years at large firms. (The analysis assumes no increase in the amount or nature of the legal work performed.)

“Automation acts as one more drag on a legal market that had already become relatively saturated by the middle of the last decade,” the authors write.

But the authors also write that if the legal market increases, or if law firms don’t adopt all the available legal technologies—both likely scenarios, the authors say—firms could see little or no reduction in lawyer staffing. In previous technological leaps, from industrialized agriculture to new medical imaging technologies, the adoption of labor-saving technologies did not result in widespread layoffs, and in fact often led to an increase in the size of the labor market, they note.

The study finds that automation’s effects on legal work flow should vary widely according to the type of work that lawyers perform. Some tasks, such as document review in litigation, are best suited to the application of new search technologies. The authors estimate that automation could replace about 85 percent of the lawyers currently assigned to document review—but in the end, that won’t affect law firm employment much, since document review represents only about 4 percent of invoiced hours.

Most document review, the study finds, has either already been automated or is in the hands of contract attorneys, who are indeed likely to see dramatic declines in employment.

Many other tasks can now be partly automated, including document drafting, due diligence on deals, legal research and legal analysis, the paper says. Lawyers spend roughly 39 percent of their billed time on such tasks. The authors report that it would be reasonable to expect a 19 percent decline in the number of employees required to handle these tasks.

But many of a lawyer’s most critical tasks—representing 56 percent of billed time—are and probably will remain ill-suited to automation, the study concludes. Those tasks include fact investigation, legal writing, advising and communicating with clients, court preparation and appearances, and negotiation. Overall, the authors expect a drop-off of 5 percent in lawyer-hours in those areas.

Most client advising, the authors write, “remains outside of the current domain of automation.”

The study acknowledges that legal automation is making inroads in surprising areas. One company, Modria, markets an online dispute resolution technology that summarizes areas of agreement and disagreement and makes suggestions for resolving issues. Currently, the approach is used primarily for small e-commerce disputes, but Modria is expanding into larger and more complicated disputes, the authors say, suggesting that “computers may eventually play a larger role in aiding, if not replacing, lawyers’ negotiating work.”

But again, such technologies have important limits. Many interpersonal aspects of lawyering that remain beyond the scope of automation are irreplaceable, the authors stress. Computers can’t replicate the emotional intelligence that allows a lawyer to read facial expression and body language to determine how to proceed with a deposition or a sensitive deal negotiation. Specific tasks such as fact investigation, they note, require both analysis of nonverbal information and flexibility to adjust questions when new information surfaces.

Predictive technology, which can help determine the likelihood of various outcomes, is helpful but only as part of a lawyer’s toolkit. “Effective advising encompasses more than prediction,” the authors write. “It requires a lawyer to understand a client’s situation, goals and interests; to think creatively about how best to serve those interests pursuant to law; and sometimes, to push back against a client’s proposed course of action and counsel compliance.”

Assessing the growth of predictive technology and other tools, the authors also outline a series of potential dangers to the profession that go beyond employment.

The study cautions that overadoption of cutting-edge Big Data-based technology could hamper lawyers’ ability to thoroughly advise their clients. Lawyers will increasingly be relying on complex search engines or predictive technologies that they don’t fully understand, posing problems when the predictions fail or discovery fails to turn up a “hot document” that falls outside certain algorithmic parameters.

“Much of the prediction process is hidden from view,” the authors write—and out of a lawyer’s knowledge or control.

There are also ethical considerations.

“Reducing advice to prediction would eliminate a core function of lawyerin—counseling compliance with the law,” the authors write. “If a client’s only legal advice comes from a computer’s prediction of how a court will likely respond, advising will become calculating what a client can get away with.”

In the longer term, the study says, increasing automation of legal analysis may be a drag on the development of the law. “Creativity and novelty in legal argument generally comes from importing legal concepts from one area of law into another, and by combining existing arguments in new and persuasive ways,” the authors assert. But computer programs, while highly effective in making predictions given the legal system as it currently exists, are “far less so in making suggestions for how the legal system could or should evolve.”

The authors hope that readers will download the paper and comment on it. “We wanted to start a discussion on how it would impact the profession. It’s by no means the last word on the subject,” Levy said.

LOAD MORE
Industry: