Artificial intelligence (AI) is transforming the legal profession in many positive ways, from automating tasks to enhancing research and analysis. However, AI also poses significant challenges and risks for lawyers and the legal system as reported recently in “Artificial Intelligence: A Legal Minefield for Lawyers.” In this article, we will explore one of the main legal pitfalls of AI’s impact on the legal system and lawyers: fabricated voices and faces.
Deepfakes and voice cloning technologies can create audio/ video evidence that looks and sounds authentic, making it much more difficult for the legal system to authenticate and verify evidence. Most of us have heard of the term “deepfake.” The Merriam-Webster online dictionary defines deepfake as “an image or recording that has been convincingly altered and manipulated to misrepresent someone as doing or saying something that was not actually done or said.” Merriam-Webster online dictionary. Voice cloning is sometimes referred to as “soundalikes,” or “one that sounds like another.” Id.
Lawyers know that photos, videos, and audio recordings can be crucial pieces of evidence in a legal case, as they can provide visual or audio information that may not be available from other sources. Photos and videos can capture the scene of a crime, an accident, a violation, or a dispute, and show details such as the location, the time, the people involved, the damages, the injuries, or the actions. Recordings can capture a conversation and any audio that may be relevant to a case. Photos, videos and recordings can also corroborate or contradict the testimonies of witnesses, experts, or parties to the case, and help establish the credibility, reliability, or accuracy of their statements. Good litigators know that photos, videos, and recordings can elicit emotional responses from the judge, jury, or public, and influence their perception of the case. Considering the significant impact on the outcome of a legal case, this evidence should be carefully collected, preserved, analyzed, and presented by the lawyers. Of course, photos, videos and recordings have been susceptible to manipulation, e.g. Photoshop, video and audio editors.
Enter AI, where now deepfakes can be used to threaten individuals and organizations and can alter or create fictitious photos and videos. It is important for lawyers to have a better understanding of deepfakes, what they are and how they are created. Facial recognition algorithms and a computer variational auto encoder (VAE) are used to create deepfakes. The VAE’s are programmed to recognize and encode a photo into low dimensional images and then decode someone else’s image back into a photo. For example, applying the VAE to the face of a young boy and a decoder on the face of Tom Cruise, would result in the face of Tom Cruise on the body of a young boy. This has already been done. A deepfake of Scarlett Johansson’s image and voice recently appeared on X (formerly Twitter) in an ad. The real Scarlett Johansson took immediate legal action as did fellow thespian Tom Hanks when a deepfake image of his likeness and voice promoted a dental plan. Last year, 20 photos of girls had been stolen from their Instagram accounts. Someone altered their photos using AI, created naked images of them, then shared the AI manipulated naked photos in Whatsapp groups.[1] And now, the most famous person on the planet, Taylor Swift, has been a victim of AI manipulated images producing fake pornographic images of her. Then earlier this week another deepfake image of Taylor was touting Le Creuset cookware.
Laws Desperately Needed to Govern AI Manipulation of Photos, Videos and Audio Recordings
A number of states are rushing to regulate deepfakes as AI continues to go mainstream. But unless you’re a digital or audio forensic expert, it is hard to determine if something is fake or not. At least one Congressman has acknowledged the problem and is hoping something positive can come from the fake images of Taylor Swift. Rep. Joe Morelle has presented the “Preventing Deepfakes of Intimate Images Act,” making non-consensual sharing of digitally-altered explicit images a federal crime.
Voice cloning has also made the news. Recently, AI created a song mimicking the work of Drake and The Weeknd. The song, “Heart on My Sleeve,” featured on TikTok and Spotify and was very popular. The fake song was eventually removed from music platforms after protests by Drake, The Weeknd, and their record label. A bi-partisan proposal called the No AI Fraud Act and its Senate counterpart, the NO FAKES Act, hopes to stop the threat of deepfakes and voice cloning AI software that can manipulate voices and images without a person’s consent.
Evidentiary Problems
Artificial intelligence has the ability to create an event that never occurred, or showing a person saying something that they never uttered. Lawyers need to educate themselves on the science behind creating deepfakes and voice cloning. If a deepfake video shows a witness lying or contradicting themselves in a court case it can have serious implications for opposing counsel and for lawyers who inadvertently use the video as evidence. Lawyers should have a heightened awareness that AI manipulation is a possibility and harder to detect.
Deepfakes and voice cloning can and will shake our evidentiary foundations. Digital and/or audio forensic experts may be needed in every case to authenticate the photo, video and/or recording to ensure it is original and has not beenmanipulated and/or created by AI. This issue was manifested in a riveting TV series called, “The Capture.” Its premise was the danger of artificial intelligence in the context of surveillance and criminal justice. The main character was wrongly convicted of murder based upon a deepfake video created by AI using security footage. The show revealed how artificial intelligence can be used to create convincing deepfakes, alter evidence, and influence the judge and jury.
Lawyers in Maryland may also want to consider Maryland Attorney’s Rules of Professional Conduct 19-303.3, Candor Toward the Tribunal. The rule provides a prohibition against offering evidence that an attorney knows to be false, or failing to correct a false statement of material fact previously made to the tribunal. Remedial measures are also required by the rule, in the event the attorney becomes aware of the falsity of any evidence presented. Section (c) of the rule, allows an attorney to “refuse to offer evidence that the attorney reasonably believes is false.” Id.
Summary
Artificial intelligence in the context of deepfakes and voice cloning highlights the disruptive nature of these technologies and the challenges they pose for evidence and truth-finding in the law. The legal system relies on audio and photographic and video evidence to establish facts. But deepfakes and voice cloning can be difficult to distinguish from the truth. These technologies have the potential to undermine the integrity of the evidentiary process, and the veracity, authentication and admissibility of evidence. Evidentiary standards may need to be adapted and new protocols established to adapt to these new technologies. Regardless, lawyers may want to think about the ethical challenges posed by these new technologies and make adjustments accordingly.
___
1 Ai-generated naked child images shock Spanish town of Almendralejo, By Guy Hedgecoe, BBC news, September 23, 2023. bbc.com/news/world-europe-66877718.
This article appeared in the Spring 2024 edition of the Maryland Bar Journal. To read this article and more, click here.