AI-enhanced video fails as evidence in court

The court in question was investigating the case of defendant Joshua Pulock, who wounded two people and killed three others with his firearm outside a Seattle bar in 2021. In part, these were random bystanders who were not involved in the incident. According to the defence, the defendant was ambushed by one of the people he hit and the shooting was done in self-defence. The hitting of the other innocent persons was, according to them, an accident. The man whom Pulok identified as the mastermind of the incident was also fatally struck.

The shooting, including the hitting of the victim, was recorded on a cell phone video. However, Pulok’s lawyers had the video edited by special software using artificial intelligence before it was submitted to the court.

How does the Anglo-American legal system work?

A key element of Anglo-American law is the precedent-based way of deciding cases. In practice, this means that if the parties do not agree on a legal qualification, the courts look for a judicial decision within a given case that typically resolves the issue. However, if the court finds that the current case is fundamentally different from all previous cases, the judge has the power and duty to establish precedent. Thus, de facto law is created. It is not its only source, but it is the most important one.

In this context, the plaintiffs in the case argued that as of February of this year, there is no legal precedent in King County allowing the use of AI technology in U.S. criminal court. Simply put, therefore, it was primarily up to the discretion of the court to decide how to approach the issue (while respecting the legal rules in that area of the law).

The defendant’s lawyers said the edited or enhanced version is still a “faithful representation of the original”. However, the prosecution countered that the edited video was misleading, inaccurate and unreliable as evidence. The called forensic expert confirmed that the enhanced version, on the one hand, contained data and video content that was not in the original. At the same time, some of the data had been removed from the original.

Judge Leroy McCullough took the above-mentioned negative view of the possibility of using the edited video as evidence in court, stating that the edited video relies on opaque methods to represent what the AI “thinks” it should show. Further, it was said that admitting the AI-enhanced evidence would lead to confusion as well as confuse or undermine eyewitness testimony.

In this regard, NBC News in its article quoted experts from Amped, a company that, among other things, analyses photos and videos to detect tampering. Its representative explicitly stated that artificial intelligence is not reliable enough to be used in a legal setting for image enhancement.

Leaving that case aside, AI is increasingly becoming not only a necessary ‘good servant’ that can, for example, suggest an illustration for an article or its headline, but also, unfortunately, often a ‘bad master’ that can, for example, create a ‘photograph’ of a scene that never happened or put sentences in your mouth on video that you never said. In response to such incidents, the White House recently released a government-wide policy that seeks to mitigate precisely these negative aspects of AI. Similar legislative measures will surely soon be necessary in Europe, and are already on the way.