An artificial intelligence helped to protect the identity of people who took part in a violent anti-government protest in Hong Kong by swapping their faces for those of actors. The method was used in a new BBC documentary, with filmmakers saying it protects the protesters, who were interviewed on camera, from persecution while conserving their mannerisms and the emotions conveyed in their facial expressions.
The BBC2 documentary, , interviews four protesters who took part in tense and increasingly violent clashes with police following the proposal of a law that would allow criminal suspects to be extradited to China.
Possible government reprisals meant the identity of the interviewees had to be protected. Rather than covering them with a black bar, shooting them in silhouette or pixelating images, the BBC used AI to change the interviewees’ faces while retaining all their original movement.
Advertisement
at US-based developed the AI and says the effect was deliberately made to be not entirely convincing. This means a viewer may be aware that changes had been made to the speaker’s face, but the effect was accurate enough to convey when an interviewee smiled, became tearful or frowned, he says. “We include sort of a soft halo around the picture,” says Laney.
The AI process involved an actor being filmed for around 5 minutes in lighting conditions similar to those used for the interviews with the protesters. The actor read a generic statement that covered the facial movements that are created when all the various vowel and consonant sounds are made.
This video was then fed into an AI model, along with footage of the interviewee that needed to be obscured. The resulting model was specific to that actor-interviewee pair and could take new footage of the interviewee and transform their face into that of the actor’s.
Laney says that not only the facial features, but also the shape of the skull, the folds of the ears and the skeletal structure of the face were changed to avoid Hong Kong authorities trying to identify an interviewee from biometric clues.
, who directed the documentary, says the concept of AI disguises was “tricky” to explain to cautious interviewees, but that framing it as a type of “deepfake” was quickly understood. The BBC asked for the term “deepfake” not to be used in the documentary because of its negative connotations. For example, deepfake technology can create videos of politicians seemingly saying something they never actually said.
“Some [interviewees] were in Hong Kong still, and obviously… they could literally be arrested and imprisoned the next day for the stuff that they’re telling us about having done,” says Paton. “But even those in the UK, with families in Hong Kong, they were genuinely afraid that if they were identifiable in the footage and in the film, that that could mean that the authorities will put pressure on their families.”
Paton says that protecting those taking part in the film had to be considered alongside making the documentary compelling and conveying the emotions of the interviewees that could be lost by cruder obscuration.
“[With the AI approach] you can see all the character, all the nuance of expression, the emotion is all there, kind of written on the face,” he says. “And I really wanted viewers to be able to look them in the eyes and see that the person talking and narrating these events is a very, very young adult, you know, some of them not even out of their teenage years. I think that that’s really important.”
at City, University of London, says that AI is a powerful technology with many benevolent uses, but it also has the potential to be abused.
“In this situation, though, my sense would be that it’s OK, but only if it is explicitly made super clear to viewers what has been done,” she says. “And I think it would also be very valuable if the reporter took a minute to also briefly explain why it was done. With trust in the media so low, and so many people ready to ascribe the worst possible motives to journalists, any opportunity to explain why a potentially debatable ethical decision was made should be seized, in my view.”
The documentary begins with the disclaimer: “The protesters in this film could be imprisoned for what they tell us. We have protected their identities using artificial intelligence.”
