A Washington State Superior Court judge recently rejected the introduction of “artificial intelligence enhanced” video evidence for use in a jury trial, believed to be the first time this has happened. State of Washington v. ProcaWashington state has charged Joshua Proca with three counts of murder stemming from a shooting in 2021. The shooting was filmed on a passerby's smartphone, and 10 seconds of uncensored source video of the shooting was entered into evidence.
However, the defense also asked to submit an AI-enhanced version of the video. The defense's experts argued that the source video was low resolution, had significant motion blur, and contained fuzzy images with “blocky” edge patterns. To address these issues, the defense's experts said they used the Topaz Labs AI program's AI video editing tools to enhance the clarity of the source video, and then further processed the video using an Adobe program. The defense's experts said the Topaz Labs AI program used techniques to “intelligently enlarge and increase the resolution” of the video, adding clarity, definition, and smooth edges to objects in the video.
The state said the submitted AI-enhanced video was Frye v. United States – Evidence that uses new scientific theories or principles must be generally accepted in the relevant scientific community. According to the state's expert (a certified forensic video analyst), the AI tools used by the defense made forensic analysis of the video impossible. The state's expert listed the problems with AI-enhanced video:
- The video was enriched with 16 times the number of pixels than in the original video using algorithms and enhancement methods that are unknown and have not been verified by forensic video experts.
- Add information that was not in the original file,
- Remove artifacts from individual images,
- Shapes and colors have changed in the video.
The state's experts also testified that the Scientific Working Group on Digital Evidence, whose members represent state, local and federal law enforcement agencies involved in forensic video examination, had issued warnings about the use of AI-enhanced tools in courtrooms.
After hearing oral arguments on both sides, the court rejected the defendant's submission of the AI-enhanced video, finding that the proposed evidence did not meet the requirements. fly Standards. The Court first noted that because using AI tools to enhance videos presented in a criminal trial is a novel technique, the defendant bears the burden of showing that the method is accepted by the relevant community. Finding that the relevant scientific community is the “forensic video analysis community,” the Court found that the defense did not meet that burden.
Specifically, the court found that the Topaz Video AI-enhanced tool, which uses machine learning algorithms, has not been peer reviewed by, is not replicable in, and is not generally accepted in the forensic video analysis community. The court further noted that the defense has not presented any state or federal appellate court decisions that have examined or approved AI-enhanced video in criminal or civil cases. It also noted that the defense has not presented any articles, publications, or secondary legal sources that approve the introduction of AI-enhanced video evidence in criminal or civil cases. The court also noted that the defense's experts themselves acknowledged that they did not know what videos the AI models were trained on, did not know whether such models employed generative AI in their algorithms, and agreed that such algorithms were “opaque and proprietary.”
The court further found that the AI-enhanced video did not satisfy Washington Rule of Evidence (ER) 702, which states that evidence is only admissible if it is reliable and helpful to the trier of fact, and ER 403, which states that evidence is inadmissible if its probative value substantially outweighs the risk of unfair prejudice. The court found that the AI-enhanced video did not honestly show what actually happened, but instead represented in an opaque manner what the AI model thought it should show, and that there was a significant risk of a lengthy trial in court over the unpeer-reviewed process used by the AI model.
The court's decision Proca This is another example of judicial skepticism towards AI-enhanced or generated evidence helping to decide cases. ProcaCourts have noted on several occasions a lack of transparency regarding the inputs of AI models, and unless courts are provided with more information about such inputs, they are likely to continue their tendency of omission.
