Your face has likely been in a virtual photo lineup

The face of nearly one in two American adults are in law enforcement face recognition networks. So whether you like it or not, and without your permission, chances are your face has been used in some kind of virtual photo lineup used in a criminal investigation.

                   

That's because those numbers are not only made up of mugshots taken of people arrested by law enforcement. That's especially true if you live in Florida.

A report was recently released by Georgetown University Law School following a year-long investigation that included over 100 public records requests to law enforcement agencies and police departments across the country. According to the report, the faces of over 117 million American adults are stored in law enforcement face recognition databases, and close to 64 million Americans have no say in the matter; Florida being one of the states.

What are we really talking about here, though, when we say face recognition? Many criminal defense attorneys, and even the public at large watching television crime dramas, will know about a photo lineup or even a physical lineup. Typically a police officer will compile six photographs for a lineup to show a witness, with one being a known suspect and the other five looking generally similar. Many times that is not the case, and I’ve successfully defended clients with highly suggestive and impermissible lineups where the lineup was all but singling out my client. Nevertheless, the witness will look at those six photos and pick out the one that he or she is confident committed the act.

As the report defines it, face recognition is “the automated process of comparing two images of faces to determine whether they represent the same individual.” Basically an algorithm is used to “find” a person’s face in a picture (characteristics and features that are numerically quantified), and then using the algorithm that “face” is compared with others in a database of other faces to find a set of matches. The results don't specifically identify a single match, but instead identify a list of candidates or top matches that are then provided to law enforcement. 

The report focused on face identification (as opposed to verification), which seeks to identify an unknown face. Law enforcement, for example, could stop an individual and snap a photo using their cellphone or tablet and run the photo on their squad car computer to determine a match. Other examples include identifying individuals after arrest (with other cases); identifying suspects during investigations with images from security cameras or smartphones or social media; and real-time video surveillance where an image of an individual law enforcement is looking for is uploaded and compared with faces that are extracted real-time from live security video feeds (a match would alert a law enforcement officer nearby that camera). Ultimately, a human determines who the final match is based on the list of candidates.

Image by George C. Palaidis

The report uncovered some startling discoveries, as I’m sure you could expect. The first problem is that no state has passed a law comprehensively regulating police face recognition and there are no known agencies which require warrants to search these enormous databases of photographs. This means law enforcement agencies are free to use and search the systems without any guidelines, boundaries, or parameters restricting their use.

Many major police departments are also now exploring real-time face recognition on live surveillance camera videos (like the giant cameras you’ll see on street corners in New York City). This same technology could also be applied to body cameras and dashboard cameras as both are used by more and more law enforcement agencies.

Since there are no laws or rules regarding the use of the technology, most agencies have no rules prohibiting officers from using face recognition to track individuals engaging in political, religious, or other protected free speech.

Further, little is done to ensure that the systems are even accurate, and the human (police officer) factor utilized to determine if a candidate photo is in fact a match is largely wrong (typically wrong about half of the time). These systems also tend to disproportionately affect African Americans.

Here in Florida the Pinellas County Sheriff’s Office has a system called FACES (Face Analysis Comparison & Examination System) that searches over 33 million faces, including 22 million Florida driver’s license and ID photos and over 11 million law enforcement photos. Florida law enforcement can also search the FBI’s database of 24.9 million mug shots. Using the system, the FBI’s Florida field office and 243 other local, state, and federal agencies run close to 8,000 monthly searches without requiring even a reasonable suspicion before running a search!

Regardless of whether you've committed a crime or not, Florida law enforcement agencies are using your driver’s license photo (along with other photographs) to create a virtual line-up when looking for the identity of a suspect accused of a crime. And you have no say in the matter.

You may ask ‘what’s the big deal if I’m not guilty of anything?’ Well, an experiment was conducted by a professor at Michigan State University which ran police security camera photos of the Tsarnaev brothers, who committed the Boston Marathon bombings, against a database of a million driver’s licenses. The system did find the younger brother in a match of 10 possible candidates, but it also identified nine other people who were clearly not guilty. What would have happened if law enforcement had used that system in searching for the bombers and showed up at the homes of the nine other people?

We already run into the same problem with eye witness identification and photo lineups. I once successfully represented a client accused of armed robbery who was “identified” using a drawing from a police sketch artist based on the description from the victims. Another officer thought the sketch resembled my client, and he was arrested solely on that identification. Needless to say, the sketch looked nothing like my client. But he was stuck in jail for months as I had to gather the evidence to prove to the prosecutor that my client was innocent. Even after passing a polygraph, the prosecutor was unwilling to dismiss the case. On the eve of trial the proper decision was made and the case was dropped, but during that time period my innocent client had to sit in jail where he couldn’t work, pay bills, be with his family, and had suffer the stigma of this serious criminal charge.

These same problems can and will arise as law enforcement agencies move headstrong into face recognition technology and systems without any guidance, laws, or regulations keeping it in check. The report makes several recommendations, including: passing comprehensive laws regulating law enforcement’s use of face recognition; that police should not run face recognition searches of license photos without clear legislative approval; that accuracy tests should be created and expanded to ensure the highest accuracy with the systems; and that community leaders should press for policies and legislation that protect privacy, civil liberties, and civil rights.  

We’ve seen it many times in the movies and in real life: technology can be a good and useful thing, but it can also be manipulated and abused, whether intentionally or not. This new and evolving technology should be held in check to prevent any misuse and abuse by its users. Otherwise it could be you identified in a virtual lineup based on your driver's license photo and accused of a crime you just didn't commit.

George Palaidis is a criminal defense and civil business attorney practicing in Miami, Fort Lauderdale and the South Florida region.

Fighting "Junk" Sciences Resulting in Wrongful Convictions

You have watched CSI or NCIS, right? Maybe the Forensic Files? You’ve seen how the “bad guy” was found using bite-marks, hair analysis, fingerprints, firearm identification, shoe-prints, or even DNA. Thousands of convictions solely from those forensic sciences. What if I told you these “junk” sciences were essentially debunked by a top government report released last week?

The report, titled “Forensic Science in Criminal Courts: Ensuring Scientific Validity of Feature-Comparison Methods,” comes from the President’s Council of Advisors on Science and Technology (PCAST), which was tasked by President Obama back in 2015 to consider whether additional steps could be taken to strengthen the forensic-science disciplines and ensure validity of forensic evidence. PCAST compiled and reviewed more than 2,000 papers, submissions and other reports in preparing its report.

This report took to task seven forensic feature-comparison methods, even finding some methods “far from meeting the scientific standards for foundational validity.”

So what does this really mean? Well the report looked at “feature-comparison methods” which are methods that compare two samples to see if there is a match based on identified features. Basically, feature-comparison methods work like this: the Crime Scene Tech shows up at a crime scene and takes some sample of evidence (a photograph, a sample of blood from the ground, a partial fingerprint) and brings it back to the lab in an evidence bag. Then at the crime lab, an examiner analyzes that sample from the scene and attempts to compare it with samples of from known sources, such as an already identified suspect. A “match” means the police have ‘found their man.’

Five years at the Miami-Public Defender’s Office as an assistant public defender, and another three years in private practice as a criminal defense attorney, has shown me that the “match” is not always as it seems. Now the government and scientific community is catching up and providing valuable research and evidence challenging for the first time what has been held as infallible science.

The PCAST report looked at first at foundational validity of the method and then validity as applied. Foundational validity of a method means the method is based on studies and evidence and that it can be repeatable, reproducible, and accurate at levels that have been measured and are appropriate – in essence, the method is reliable.  

Validity as applied focuses more on how the method is applied in practice. For example, did the tech reliably apply the method in the case?

In essence, is the method reliable and did the examiner reliably apply the method in the case.

What the report found should be shocking, at least to those who don’t practice daily in criminal law.

As to DNA, what many consider the infallible forensic science, the PCAST report found that depending on the type of analysis (single-source/simple-mixture sample vs. complex-mixture samples) there were several factors that can and do cause errors, whether based on the foundational validity or, more frequently, as applied when human error comes into play. Errors can come into play where the tech inappropriately collects the sample from the scene, or in sample mix-ups, contamination, incorrect interpretation, and errors in reporting.

The PCAST report also found serious issues with latent fingerprint comparisons (comparing a partial fingerprint from a crime scene with known prints taken in a controlled setting). This field is highly subjective because it requires examiners to analyze that partial print to find certain characteristics and then compare those with other prints. Many of these examiners are also members of law enforcement and the prosecution team, thereby carrying an inherent bias that shouldn’t exist in a scientific setting.

The report noted that some studies have found false positive rates at 1 error in 306, and even 1 in 18! Yet these fingerprint identifications are sometimes the sole evidence used to convict defendants in criminal courts. Remember that the FBI announced a print from the 2004 Madrid train bombing was a perfect match to an American lawyer, but that was quickly debunked by Spanish law enforcement whom identified a different individual as the source of the print.

Bite-mark analysis was another “science” that PCAST reviewed. As the name suggests, the tech compares mark left on a victim or object with the dental impressions taken from a suspect. The PCAST report found bite-mark analysis is “far from meeting the scientific standards for foundational validity.” That examiners couldn’t agree on whether an injury is a human bite-mark surely assisted in that conclusion.

The report also found that firearm, footwear, and hair analysis were not foundationally valid as there were no real studies or empirical evidence backing up the methods as repeatable, reproducible, or accurate. A 2015 Department of Justice (DOJ) and FBI report that reviewed 3,000 criminal cases involving hair analysis discovered that in more than 95% of those cases the FBI examiners provided scientifically invalid testimony that was used to inculpate or implicate he defendant. Convictions based off of “junk” science.

All of this is important because judges and jurors in this country are often presented “scientific” evidence that is “100% certain” or that have “zero,” “essentially zero,” or “negligible,” error rates. The reality is that much of the so-called forensic sciences aren’t based off of any science at all, but are resulting in the convictions of thousands of innocent criminal defendants. And criminal defendants, when presented pre-trial with this “scientific” evidence, often are compelled to plead guilty for a negotiated sentence rather than go to trial because of the misplaced trust so many have in this “scientific” evidence.

That’s why it’s important for not only criminal defense lawyers like myself to be familiar and proficient in attacking this evidence, but also for the community at large to challenge these “scientific” methods that have long been held to be infallible, and for the scientific community to develop scientifically valid methods to ensure the innocent aren’t wrongly convicted.  

It is important that we start now in challenging this evidence, because there is no guarantee that past convictions will get the review they should receive. As Federal Judge Alex Kozinski noted in a recent commentary in the Wall Street Journal, even though a 1997 DOJ report impugned 13 FBI lab examiners involved in more than 7,600 cases (64 capital murder cases), by 2014 only 312 had been reviewed to determine whether convictions were obtained through discredited methods.

I would certainly suggest that criminal defense attorneys review the full report, but I’d recommend everyone take a few minutes to read the 20 page Executive Summary. The PCAST report is not only eye opening, but offers quite a few of important recommendations to fix the problems presented today and to work towards eliminating wrongful convictions. 


George Palaidis is a criminal defense and civil business attorney practicing in Miami, Fort Lauderdale and the South Florida region.