Predictive coding is machine-learning technology that lets computers automatically predict how documents should be classified with limited human input. Some e-discovery commentators (and vendors) have spent the past few years touting the technology as the cure for all of e-discovery’s ills.

But the technology’s actual arrival to the world of litigation and discovery motions reveals that it is not without side effects. Two recent cases involved protracted predictive coding disputes. Digging into those disputes highlights some of the technology’s potential future problems and clarifies those instances and situations in which it may be put to appropriate use.

Da Silva Moore v. Publicis Groupe: Proponent or prejudice?

Da Silva Moore v. Publicis Groupe is the first, and most high-profile, judicial opinion endorsing the use of computer-assisted coding—and a founding saga of the technology’s potential areas of future dispute. The case is, essentially, a Title VII gender discrimination class action. Discovery in the matter, which is venued in the Southern District of New York, is being overseen by Magistrate Judge Andrew Peck, who has been a leading proponent of predictive coding for some time. In addition to appearing at industry conferences as a speaker, Judge Peck wrote an article called “Search, Forward” in which he advocated use of computer-assisted review technologies and said the article itself could serve as a “sign of judicial approval of predictive coding” in appropriate cases. 

The parties in Da Silva Moore initially agreed to use predictive coding; however, negotiations over a mutually agreeable protocol for programming the coding technology eventually broke down. Plaintiffs argued that the court erred in accepting the defendant’s recommended protocol, which they contend is flawed. As a result, discovery was stayed and plaintiffs moved for Judge Peck’s recusal, arguing his support of the technology and relationships with its vendors have made him biased. Judge Peck denied plaintiff’s requests to recuse himself. Judge Andrew Carter, the presiding district court judge in the case, is now considering the plaintiffs’ objections to Judge Peck’s orders denying recusal and concerning the parameters of the predictive coding protocol.

Kleen Products v. Packaging Corp. of America: You can’t make me

In Kleen Products v. Packaging Corp. of America, an antitrust matter set in the Northern District of Illinois, the plaintiffs asked that the defendants redo their document production and all future productions using predictive coding technology. The plaintiffs argued that use of the method would realize more thorough results than those provided by defendants’ keyword search technology. This request, however, came after defendants had already spent thousands of hours reviewing documents and had produced more than one million documents. 

After hearing expert witness testimony on the sufficiency of the initial production, Magistrate Judge Nan Nolan asked the parties to reach a compromise on keyword searches. In comments on the record, Judge Nolan confirmed the parties cannot dictate what technology their opponent may use without indicating how document production results are insufficient or inaccurate. Subsequently, in a recent stipulation and order, the plaintiffs withdrew their demand.

Conclusion: Predictive coding predictions

Predictive coding is here to stay, and disputes about its uses and limitations are sure to add to the e-discovery bill for at least some of its early adopters. Potential areas of dispute include not only use of the technology in general but also the specifics of operating and programming protocols, even if everyone agrees to use the technology.  

For the time being, the availability of predictive coding has not yet displaced other commonly accepted e-discovery methods such as keyword searches, nor will it anytime soon. Despite the fact that predictive coding is fast becoming an accepted technology for e-discovery, its availability is not the basis for challenging appropriate uses of other e-discovery methods, including standard keyword search technology. As a result, the selection of the technology (or technologies) to be used for e-discovery remains an important strategic choice with the potential to impact both the cost and duration of litigation.