Robocop Finds His Man, But Man Gets Robocop’s Instruction Manual

By: James Trusty

Many years ago, a prosecutor I worked with at the time was in a fascinating murder trial, where the defense included a forensic psychiatrist opining that the defendant suffered from multiple personalities. According to this expert, one of the “inhabitants” of the defendant’s mind was a creature named Tofu the Demon Dog. On cross examination, the doctor volunteered to the jury that these distinct personalities within the defendant were “as individualized as a fingerprint.” The prosecutor then corrected him with, “don’t you mean a paw print, doctor?” A single snarky question destroyed the unique and pseudo-scientific effort at establishing reasonable doubt.

The admissibility of scientific—and certainly psychiatric—expert testimony and methodology in criminal cases has been a recurring battlefield ever since courts across the country accepted their “gatekeeping” function under Daubert v. Merrell Dow Pharmaceuticals, Inc., 509 U.S. 579 (1993). Courts have had pretrial hearings regarding the reliability of all sorts of forensic testing—including evolving methods of DNA analysis, gunshot residue, bullet-lead “batch match” comparison, and motor vehicle accident reconstruction, to name a few. A Daubert hearing is designed to allow the judge to assess the validity of the particular forensic test within the scientific community as well as the reliability of the testing (and tester) as applied to the particular trial. Two related principles come together in this context. First, that as the particular science becomes generally accepted, there is no longer a need to hold Daubert hearings (ex. fingerprint identification, and radar/speed guns). Second, when a hearing is necessary, it logically follows that the opposing side should be able to obtain methodological discovery, or the hearing is essentially a one-sided dress rehearsal for trial.

One of the latest tech battles implicates serious concerns as we enter the world of Artificial Intelligence. In New Jersey v. Francisco Arteaga,¹ the defendant was accused of armed robbery, and the two witnesses from the store picked a photograph of the defendant from a photographic array. The court took little issue with the process of presenting the array to the witnesses, finding the defendant’s challenge failed to meet the high standard for suppressing an extrajudicial identification, that there was a “substantial likelihood for irreparable misidentification.” But the photo of Mr. Arteaga was a different story, and one that takes us to the cutting edge of investigative technology, facial recognition technology (“FRT”). Faced with a relatively grainy and murky surveillance video, the detectives generated a still image and sent it to the New Jersey Regional Operations Intelligence Center for facial recognition analysis. The New Jersey investigator informed the detectives that he could not provide a match for the still photo, but that if he could receive a better image he would try again. The detectives instead sent all of their footage to the New York Police Department Real Time Crime Center (“RTCC”), where a detective there said their analysis showed a “possible match” to the defendant. The RTCC database photo of the defendant was then provided to the investigators, who used it in the array presented to the witnesses.

The defense provided the prosecution with a lengthy list of FRT-related discovery demands, of which only a couple of responsive items were provided to them before trial. A good number of the requests seemed perfectly tailored to the reliability issues that require pretrial resolution under Daubert. Error rates for the algorithms used in searching for matching photos, for example, are standard fare for new technological procedures. The court noted that “[p]oorer quality images can be edited for lighting or color correction, to enhance detail, or even to change facial expression.”² So while the computerized search for matches relies heavily on precise measurements of facial structure, the investigator conducting the search can alter the initial photo’s contents. This kind of procedural subjectivity invites scrutiny and could even lead to exculpatory disclosure obligations under Brady.

The New Jersey appellate court felt strongly that the defendant “must have the tools to impeach the State’s case and sow reasonable doubt”³ and ordered full compliance with the original discovery request regarding FRT. It also strongly hinted that the trial court should entertain a Daubert hearing because of the novel nature of this technology. Mr. Arteaga, then, gets another bite at the apple.

In other areas of expert testimony, a robust challenge to reliability may not lead to exclusion, but it can curtail the ultimate conclusion of the testifying expert. For example, the Maryland Supreme Court recently addressed the long-established concept of firearms comparison, by which an expert tries to match (or exclude) a spent bullet with a recovered firearm.⁴ The Supreme Court painstakingly reviewed the trial court’s 4-day Daubert hearing and granted a new trial not on the overall state of the science, but because the expert’s opinion was “unqualified.” Specifically, the expert testified that the recovered bullet had been fired at some point by the recovered gun. Uncomfortable with the powerful impact an expert opinion can have on a jury—particularly when unaccompanied by any hedging language like “to a reasonable degree of scientific certainty” or “consistent with…”, the Court handed Mr. Abruquah at least a temporary reprieve where at the re-trial the expert will have to hedge his stated certainty with some form of qualifying language.

Ultimately, a defense challenge as to how a photograph came to be inserted in an array is a fairly weak hand to play. Even if the FRT process is fraught with error or manipulation, there is an inherently attenuated nature to a witness identification of the photo as well as of the defendant in court, so do not expect a “fruit of the poisonous tree” argument to effectively ban the use of this latest investigative technology from criminal trials. And unlike the situation in Maryland, the issue in New Jersey’s retrial does not lend itself to judicial modification of the non- expert testimony of eyewitnesses.

The defense in Arteaga will be armed with more discovery, but at the end of the day they are mostly arguing reasonable doubt, hoping for the “paw print” moment that irreverently mocks the government’s novel tactic while not preventing its admissibility.

¹ Superior Court of New Jersey, Appellate Division, Docket No. A-3078-21, approved for publication on June 7, 2023.

² Arteaga, at 29, citing Clare Garie, Garbage In, Garbage Out: Face Recognition on Flawed Data, Geo. L. Ctr. On Priv. & Tech. (May 16, 2019) https://www.flawedfacedata.com/

³ Arteaga, at 36.

⁴ Kobina Ebo Abruquah v. State of Maryland, No. 10, September Term, 2022.

Related: White Collar Defense Lawyer