In large lawsuits, both sides must hand over relevant documents during a phase called discovery. By the 2010s a single case could involve millions of emails and files, and reviewing them all by hand had become slow and expensive. “Predictive coding” - also called technology-assisted review (TAR) - was an answer: lawyers label a sample of documents as relevant or not, a machine-learning model learns from those labels, and the model then ranks the entire collection so the most relevant documents surface first. It is supervised learning applied to legal document review.
On February 24, 2012, in Da Silva Moore v. Publicis Groupe (11 Civ. 1279, S.D.N.Y.), Magistrate Judge Andrew J. Peck issued an Opinion and Order that he himself described as a first. Quoting his own earlier article, he wrote: “This judicial opinion now recognizes that computer-assisted review is an acceptable way to search for relevant ESI in appropriate cases.” He noted that before his ruling, “no reported case (federal or state) has ruled on the use of computer-assisted coding,” and that lawyers had been waiting for a judge to bless the method. The case itself was a gender-discrimination suit; the predictive-coding question arose over how the defendant would produce its electronically stored information.
Peck was careful about the limits of the holding. The parties had already largely agreed to use predictive coding, so the court was resolving disputes over scope and implementation rather than forcing the technology on anyone. He grounded the decision in Federal Rule of Civil Procedure 1’s command to secure a “just, speedy, and inexpensive” determination and Rule 26’s proportionality limits on discovery burden. The District Judge later overruled the plaintiffs’ objections and left the approach in place.
Why business readers should care: this was an early, influential example of a court accepting that a statistical model can do work previously reserved for trained professionals - provided the process is transparent and defensible. The same questions Peck weighed (how to validate the model, how to sample, who is accountable for errors) recur whenever a business swaps human review for an algorithm.