YouTube is expanding its AI-generated similarity detection technology, which identifies deep fakes, to a pilot group of the company’s government officials, political candidates and journalists was announced Tuesday. Pilot team members will gain access to a tool that detects unauthorized AI-generated content and allows them to request its removal if they believe it violates YouTube policy.
The technology itself launched last year to around 4 million YouTube creators in the YouTube Partner Program, following earlier trials.
Similar to existing YouTube Content ID systemwhich detects copyrighted material in user-uploaded videos, the similarity detection feature looks for simulated faces constructed with artificial intelligence tools. These tools are sometimes used to try to spread misinformation and manipulate people’s perception of reality by leveraging the fake personalities of notable figures – such as politicians or other government officials – to say and do things in these AI videos that they wouldn’t do in real life.
With the new pilot program, YouTube aims to balance users’ free expression with the risks associated with AI technology that can create a convincing likeness of a public figure.
“This expansion is really about the integrity of public discourse,” Leslie Miller, YouTube’s vice president of government affairs and public policy, said in a press briefing ahead of Tuesday’s launch. “We know that AI impersonation risks are particularly high for those in the urban space. But while we provide this new shield, we’re also careful how we use it,” he noted.
Miller explained that not all matches found would be removed when requested. Instead, YouTube would evaluate each request under its existing privacy policy guidelines to determine whether the content is parody or political criticism, which are protected forms of free speech.
The company noted that it supports these protections at the federal level, with its support for the NO PLASTIC Law in DC, which will regulate the use of artificial intelligence to create unauthorized representations of a person’s voice and visual likeness.
To use the new tool, eligible test pilots must first prove their identity by uploading a selfie and a government ID. They can then create a profile, see the matches that appear and optionally request their removal. YouTube says it plans to eventually give users the ability to prevent uploads of infringing content before going live, or potentially allow them to monetize those videos, which is how its Content ID system works.
The company did not confirm which politicians or officials would be among its initial testers, but said the goal is to make the technology widely available over time.
These AI videos will be tagged as such, but the placement of these tags is not consistent. For some, the tag appears in the video description, while videos that focus on more “sensitive topics” will apply the tag to the front of the video. This is the same approach YouTube takes with all its AI-generated content.
“There’s a lot of AI-generated content out there, but that distinction isn’t essential to the content itself,” Amjad Hanif, YouTube’s vice president of Creator Products, explained about the label’s placement. “It could be an AI-generated cartoon. And so I think there’s a judgment as to whether it’s a category that maybe deserves a very visible disclaimer,” he said.
YouTube is not currently sharing how many such AI deepfake takedowns have been managed by this deepfake detection technology in the hands of creators, but noted that the amount of content removed so far has been “very small.”
“I think about a lot [creators]it was just being aware of what’s being created, but the volume of takedown requests is really, really low because most of them turn out to be quite benign or incremental to their overall business,” Hanif said.
This may not be the case with deepfakes of government officials, politicians or journalists.
Over time, YouTube plans to bring its fake detection technology to more areas, including recognizable spoken voices and other copyrights such as popular characters.
