Juries are funny microcosms of human perception. And human perception has an odd way of changing in sometimes unexpected ways. Decades ago, a hit television show changed the way lawyers try cases, all because human perception of trial evidence shifted. Trial lawyers struggled to adapt to juries’ shifted thinking, and since then we have been on the lookout for the next big shift.
Some think that the next big shift is here in the form of AI’s ability to create increasingly convincing altered media, commonly known as “deepfakes.” This article explores how juries might be on the lookout for deepfakes and how trial lawyers can adapt to the potential for shifted jury perception on the topic of deepfakes.
The year 2000 was a big year for the history books. Flip phones were at their peak. The United States Supreme Court had to help resolve a Presidential election. AltaVista was clinging to its search engine throne, but an upstart company called Google was gaining serious traction.
But also in 2000, a television show debuted that changed the way we try cases. That show was CSI: Crime Scene Investigation. CSI changed the way we try cases because it changed the way that jurors thought of evidence. Over the last couple of decades, the exact change that occurred because of CSI has been hotly debated in forums that range from the pages of the Yale Law Journal to the microphones of the post-trial press conferences from disappointed prosecutors. But it’s clear that CSI changed trials because CSI changed juror expectations.
Jurors armed with CSI knowledge began criticizing criminal cases in new ways. Studies seeking to quantify this big shift struggled due to the fact that prior to CSI, we did not think to ask jurors about their expectations when it comes to evidence, but one study in the years following CSI’s rise to television greatness saw some shocking new expectations. That study found that nearly half of jurors expected to see some kind of scientific evidence in every type of criminal case, and nearly a quarter of jurors expected to see DNA evidence in every type of criminal case. And those numbers increased in certain types of cases. For example, nearly three quarters of jurors expected to see DNA evidence in rape cases, and nearly three quarters of jurors expected to see fingerprint evidence in breaking and entering cases.
Those of us who learned how to try criminal cases since CSI benefitted from the early struggles prosecutors had in the years CSI was popular. For example, in my first trial as a federal prosecutor, the defendant was a man charged with a one-count violation of being a felon in possession of a firearm. Our evidence included body-cam footage of the gun being removed from the defendant’s pocket and video of a post-Miranda interview where the defendant admitted that the gun was his. My rookie inclination was to merely play those videos and rest, but my more experienced trial partner insisted that we put on a full case in order to rebut the CSI effect, which we did through forensics experts who in detail explained which scientific tests they ran on the firearm and the defendant’s fingerprints, why they did those tests, and why they did not do other tests. It was overkill, and yet the reason federal prosecutors are trained for overkill is because of myriad stories of juries hanging or even acquitting after jurors start speculating about whether various other exotic types of evidence would have exonerated the defendant.
And so, the CSI effect changed the way cases are tried; not because the science behind evidence changed, but because of what jurors think about science behind evidence changed. It took prosecutors several years and extensive growing pains to become accustomed to the new normal, and defense attorneys who caught on to the growing juror skepticism had great results in the early years of the phenomenon.
Some legal scholars have speculated as to whether we may be on the precipice of the next paradigm shift in trials: altered videos and images known as “deepfakes.” The technology behind deepfakes has advanced in leaps and bounds in recent years, but surveys show that the American public is already skeptical of videos and images because of the proliferation of deepfakes. For example, one Pew study in 2019 found that nearly two thirds of Americans say deepfakes cause a great deal of confusion about the facts of current issues and events. That same study found that two thirds of Americans reported coming across deepfakes themselves, and six in ten Americans said it was unreasonable to expect average Americans to be able to recognize the differences between real media and deepfakes. It is reasonable to think that numbers have changed since 2019 as deepfakes become better and more prevalent on the internet.
These same AI skeptics are our jurors. Jurors know deepfakes exist and are used in certain contexts. But it remains to be seen whether those sentiments could result in increased skepticism during jury deliberations and what kind of impact that skepticism could have. Attention should be paid to figuring out that impact, though, since our jurors are comprised of some of the people who drove the value of Dogecoin—a cryptocurrency literally started as a joke—north of $10 billion. Our society is through the looking glass on many things right now; expecting jurors to be skeptical of deepfake evidence in an age when so many are exposed to deepfakes is actually a rationale consideration, relatively speaking.
As an initial matter, to date, there have not been any actual reported occurrences of prosecutors introducing or attempting to introduce deepfakes as evidence at criminal trials. There will always be outlier bad actors within law enforcement, such as the recent report of a Virginia police department that on at least five occasions faked DNA reports to trick suspects into confessing to crimes. Deepfakes could absolutely be used in that vein, and the potential for misuse is serious. But the potential for using actual deepfakes as evidence has yet to materialize, at least in terms of reported conduct.
Yet juror perception does not always match reality. And so, the potential for jurors to suspect that prosecutors may knowingly or unknowingly attempt to convict a defendant using deepfake evidence is a very real prospect that practitioners should be conscious of.
The CSI effect was met by prosecutors with what ended up being a fairly simple—though time- consuming—adjustment. Prosecutors tested more evidence, but they also solicited testimony from those conducting tests to educate jurors on why some tests were run and why some tests were not run, and why inconclusive results are common in so many tests. This type of testimony satisfied most juries’ collective curiosity, and the CSI effect was mitigated.
The most likely jury deepfake collective curiosity is likely to come in the form of questions as to how evidence came into existence and whether the evidence could have been altered at some point. And so many of the foundational inquiries leading up to admission of evidence—which has long puzzled juries—may actually become scrutinized by jurors who are prone to deepfake skepticism.
A starting point should be for a witness to address metadata in some way. For those admitting digital evidence, the authenticating witness could attest that he or she reviewed the metadata for the digital evidence, and that review confirmed that the file was created around the time it should have been created and was not modified since creation. For those opposing admission of digital evidence, crossing the witness on that front—e.g., “you didn’t even bother looking at the metadata before coming to testify today?”—could also be effective.
A more comprehensive approach could involve expert witnesses, as fingerprint and DNA evidence commonly requires. For the prosecution seeking to introduce evidence and wanting to assure juries that the evidence is not a deepfake, one strategy could be to have some law enforcement officers receive deepfake training and then educate the jury on some of the things to look for to detect deepfakes. For example, one MIT project dedicated to helping the general public spot deepfakes recommends examining a person’s face within a video to spot errant shadows from lighting, and to look for a particular glare if the person is wearing glasses. It is possible that this type of testimony could be enough to assuage deepfake skeptics that the prosecution has done their due diligence on the issue, much like testimony from a fingerprint expert who explains why it is rare for usable fingerprints to be pulled from many items. For the defense seeking to introduce doubt into the credibility of evidence, witnesses with demonstrable experience in deepfakes could be effective in the right situations, particularly in the absence of a prosecution witness on the topic and play on the deepfake-distrusting sentiments of jurors.
However, one tactic that is ill-advised for parties seeking to oppose admission of evidence is to categorically argue that because of the prevalence of deepfakes, anything could be a deepfake. This is a tactic recently tried by attorneys for Tesla. These lawyers argued in a court filing seeking to discredit video of Tesla founder Elon Musk making a statement that was probative of the allegations that Musk, “like many public figures, is the subject of many 'deepfake' videos and audio recordings that purport to show him saying and doing things he never actually said or did.” Because of the prevalence of deepfakes, according to Tesla’s lawyers, the court shouldn’t rely on statements made by Musk in the contested video.
The court soundly rejected Tesla’s argument, which California Superior Court Judge Evette Pennypacker called “deeply troubling to the Court.” Judge Pennypacker, in an order requiring Musk to testify in the proceedings, wrote:
[Tesla’s] position is that because Mr. Musk is famous and might be more of a target for deep fakes, his public statements are immune In other words, Mr.
Musk, and others in his position, can simply say whatever they like in the public domain, then hide behind the potential for their recorded statements being a deep fake to avoid taking ownership of what they did actually say and do. The Court is unwilling to set such a precedent by condoning Tesla’s approach here.
A similar effort with a similar result came in the trial of January 6 insurrectionist Guy Reffitt. In that trial, Reffitt’s attorney adopted the tactic of cross-examining an FBI agent on the theoretical possibility that evidence showing Reffitt inside of the Capitol building could have been a deepfake. According to reports, the FBI agent’s response to that line of questioning was that there were no indicators that the video was a deepfake, especially because the video was found on a hard drive in the defendant’s own possession. The defense offered no witnesses or evidence. Yet based on the failed cross-examination of the FBI agent alone, the defense argued at closing that the jury should acquit Reffitt because the video could have been a deepfake. The jury convicted Reffitt.
In other words, categorical denunciation of evidence because deepfakes exist is not a wise strategy. Defense attorneys should be selective in bringing this type of defense and should be brought only when there is something the defense can point to that would suggest that the evidence is fake. But thus far, judges and juries are not responding to categorical denunciation absent particular reasons for deepfake concerns.
The CSI effect was drastic in that juries at times acquitted defendants or hung based on their own evidentiary curiosity, at times not even stoked by defense questioning of prosecution witnesses. What’s clear is that the AI/deepfake effect on juries is not nearly as drastic as that. Judges and juries have rejected categorical challenges to evidence on mere speculation that the evidence could be a deepfake.
But the jury sentiment towards the prevalence of deepfakes is still there, ready to be tapped if the defense has genuine suspicions that some government actor or some custodian of evidence may have tampered with evidence. The way to attack such a genuine suspicion is not some general commentary, but through learned witnesses sharing specific concerns. And given the steady growth in deepfakes in society, prosecutors should be on the lookout for signs that victims or witnesses could be providing them with falsified evidence, because if the CSI effect has taught us anything, the power of story has an amazing impact on the human mind. It takes one compelling story of a person wrongfully convicted because a jury was convinced of a deepfake, and then the deepfake effect could be very, very real to jurors.
