THE human face is being disrupted. Advanced artificial intelligence software is driving an explosion of online videos in which people’s likenesses are seamlessly swapped on to other people’s bodies.
Some are harmless fun, such as those in which the face of actor is pasted into films in which he never appeared. Others are far darker, including many in which the faces of female celebrities have been pasted over those of people in pornographic videos.
Advertisement
Hollywood has been swapping faces for years. The crucial difference is that now it can be done quickly and cheaply by almost anyone. “It has become a viral meme,” says Samim Winiger, co-founder of AI firm Creative.ai.
Meanwhile, researchers have used AI to footage of world leaders made to say words that never actually left their mouths. So in a world where seeing is no longer believing, what does the future hold?
Given existing fears about misinformation and fake news, it is not surprising that many people are worried about the potential harm that manipulated video footage could cause. “I’m usually a tech optimist but this is one where the red flags went up,” says Kate Devlin at Goldsmiths, University of London. “I think it could be really insidious.”
You don’t even need to fake a video to cause a stir – just the fact that video manipulation is so easily done is enough to cast doubt. Take Donald Trump’s U-turn concerning the in which he is seen making misogynistic remarks. “We don’t think that was my voice,” he reportedly said.
“Much of the discussion about AI-generated faces misses the broader impact the tech could have”
There are a few ways we could combat this threat. Some are calling for methods to authenticate video clips, or at least stop fake footage from spreading online. Videos could be given digital signatures that only their creators know in order to verify their authorship – the same technique is already used to secure some emails. This could be automated, built into both the web and also cameras and smartphones. “You want to know that a video was taken at a particular time and place,” says Aviv Ovadya at the Center for Social Media Responsibility in San Francisco.
People may even start recording digital alibis, to guard against videos that show them elsewhere. This could be done using the tracking data on smartphones, for example. “You’d have continuous alibi creation,” says Ovadya.
But Winiger is wary of tech solutions. “A lot of people will jump in and say we’ll solve this with blockchains and virtual reality with AI on top,” he says, but the reality is that teaching people digital media literacy and to be more sceptical of what they see online would probably be more helpful.
Lawyer up
If tech doesn’t work, lawyers might. Using AI to steal people’s faces without consent seems like something existing laws aren’t equipped to handle, but Neil Brown at UK technology law firm :Legal says courts could fall back on existing principles, such as copyright infringement. Most people do not have exclusive rights over their appearance, but if a video was produced using footage or photos that you took then you might have the right to sue, says Brown.
New laws may also seem like a good idea, but they could backfire. “There is a risk that if we legislate against the technology itself, we could outlaw beneficial applications, including those that we cannot currently predict,” he says.
It’s true that much of the discussion about AI-generated faces misses the broader impact the tech could have. Applications go far beyond viral memes and fake news: deeply immersive films or video games, novel forms of therapy and virtual humans to entertain or keep us company.
Take Stephen Rosenbaum of San Francisco-based company , a visual effects artist who has won Oscars for his work on Forrest Gump and Avatar. He is now in the business of making virtual people, and his latest project is cloning the band Abba, now in their 70s.
“Stephen Rosenbaum makes virtual people, and his latest project is cloning the band Abba”
“They don’t have the energy to go out and tour,” says Rosenbaum. “That’s why they asked us if we could recreate them digitally.”
Rosenbaum’s team is capturing likenesses of each of the four members, right down to facial tics and quirks in the way they smile. These avatars will star in a music video due to be released this summer, and next year will perform in stadiums packed with real people.
As these state-of-the-art tools trickle down into the mainstream, we can expect to see more digital doppelgängers of people, both alive and dead. Actors could appear in films without performing, or studios could cast a Frankenstein-type creation, by combining different actors into a virtual performer.
We could also enjoy personalised experiences, with virtual versions of friends and family members starring in bespoke films or video games. “You could be talking to your deceased grandmother with her face and her voice,” says Winiger.
For many people, virtual humans could simply provide comfort. “There are a lot of lonely people out there who want somebody to talk to,” says Rosenbaum. Companies like in New Zealand, founded by Mark Sagar, who pioneered facial capture techniques in films such as Avatar and King Kong, are trying to create virtual humans to be companions or even therapists.
respond favourably to different types of face. Tailoring a virtual human to individual preferences could ease therapy sessions and simply make an avatar better company. Digital therapists could look like celebrities, relatives from the past or your partner.
Or they could look like nobody. A few months ago, researchers at graphics hardware firm Nvidia released had invented after being trained on thousands of photos. The faces are eerily realistic, but none of these people actually exist (see image).
The current system can’t make a particular sort of face on demand, says team member Timo Aila, but in the future it could be possible to control things like age, race and gender to generate any face you wanted. “This would be immediately useful to law enforcement, helping eye-witnesses to create much more realistic-looking sketches of the people they saw,” he says.
Virtually here
Devlin suggests we could do something similar for pornography, creating images and video without the need for actual performers. Winiger agrees: “In a few years the porn industry will be an AI industry,” he says.
Yet even with avatars invented from scratch, there are ethical concerns. It is easy to chat to a customer service bot online without realising it isn’t human, for example. But forgetting that we are interacting with software means we let our guard down and reveal things we might not have wanted to, says Devlin. This will only get worse when we have photorealistic avatars as therapists or companions.
“There are those who go as far as saying we shouldn’t be deceiving people in this way at all and we should be developing machines and AI that are obviously computers or computer generated,” she says.
That may be wishful thinking, however. Now that anyone with a small amount of know-how is able to conjure up hyperreal likenesses of people, the future is looking a lot more crowded. Virtual humans – some digital doppelgängers of real individuals, others dreamed up by AIs from scratch – are here to stay, for good or ill. “You see something, you expect it to be real,” says Ovadya. “But that isn’t an assumption you’re going to be able to make any more.”
This article appeared in print under the headline “Rise of the nobodies”
