About five years ago, machine learning reached a point where software could, with guidance from senior lawyers, effectively take over the time-intensive task of legal discovery, in which one party in a lawsuit combs through its documents to determine what it must show to the other side before trial.
This is a job that junior lawyers, paralegals, or—increasingly—less expensive contract lawyers had traditionally done, and some fretted that the change might be just the first step in the computerization of the law. But while machine learning does well with structured tasks like searching for relevant words, handling documents similar to others already identified, and even reconstructing simple summaries of a baseball game, it is far less adept at constructing something like a legal memo, where persuasiveness can rely on developing novel arguments, explains economist Frank Levy, an MIT professor emeritus who, with Dana Remus, a professor at the University of North Carolina School of Law, is researching computers’ impact on the practice of law.