Virtual humans – lifelike replicas of people that are indistinguishable from the real thing – have been a science fiction staple for decades. Think the stuttering weirdness of Max Headroom in the Eighties, Blade Runner 2049‘s Joi or even the floating head of the dysfunctional Holly from Red Dwarf.
And thanks to advances in artificial intelligence and processing power, technology companies now say they are at the point where they can create photo-realistic imitations of human beings that can move, talk and smile with such authenticity that it is impossible to tell they are computer generated.
On Monday night, Neon, a highly-anticipated venture funded by Samsung, unveiled an “artificial human”, which it claimed was capable of displaying emotions and intelligence. The company said its creations would “exhibit human capabilities” such as speaking, learning and even being able to form new memories.
If further evidence was needed of Silicon Valley’s God complex, Neon’s founder, former Samsung executive Pranav Mistry, said the synthetic humans represented a “new kind of life”.
There are millions of species on our planet and we hope to add one more.Pranav Mistry, NEON founder
“There are millions of species on our planet and we hope to add one more. Neons will be our friends, collaborators and companions, continually learning, evolving and forming memories from their interactions,” he said.
More than just Alexa
Mistry’s claims are bold, to say the least, especially as Neon has thus far been purposefully vague about how its artificial humans actually work. But the company seems adamant that this project is more than just Alexa with a CGI face, claiming that it can recognise different people and can respond to questions in “milliseconds”.
And like many technological advances, Neon is out to take jobs. The company says its creations will be used for video customer service, and in the future replace financial advisers, GPs, teachers and news anchors. A virtual yoga instructor, for instance, would be able to remember a client’s name and how often they come to class.
The company also expects them to become virtual companions and will start working with businesses on introducing them later this year.
Neon’s efforts drew scepticism from the tech community, especially when it emerged that the demonstrations of its tech had itself been altered to make the technology appear more impressive than it really was.
But advances in computing mean that creating lifelike digital doppelgangers with artificial intelligence – and using them to replace humans – is an increasingly realistic proposal.
Social media and technology companies have spent the last year grappling with “deepfakes”, videos that have been digitally manipulated to make a subject – such as a high-ranking politician – appear as if they are doing or saying something else. On Monday night, Facebook banned deepfakes, saying they were likely to “mislead” the average viewer. The move comes amid fears that Facebook allowing the videos would lead to allegations of permitting misinformation to spread during this year’s US election.
New “deepfake” challenge
But while deepfakes involve taking a video and altering it, often for sinister reasons, synthetic humans represent a new type of challenge, because they are completely artificially generated.
In 2018, Chinese state media outlet Xinhua broadcast using what it said was the first AI-generated news anchor. Although few would say it could pass as a human, that could soon change.
Oren Aharon, the chief executive of Hour One, a company that generates synthetic avatars for use in advertising and ecommerce, says neither humans nor deepfake detection technology can discern its videos from the real thing.
A subject’s likeness is captured in a studio in a process taking 20 minutes and, from there, AI can generate new videos of the person realistically speaking, laughing and making facial expressions, merely by feeding it text and configuring a personality. The company says it eventually wants the process to take two minutes using a consumer smartphone.
We think it’s already passed the Turing test.Oren Aharon, Hour One chief executive
“We think it’s already passed the Turing test,” says Aharon, referring to the challenge posed by Alan Turing of fooling an observer into thinking a robot was human.
Saving time and cost
Hour One’s service is used to create videos that would be too laborious and time consuming for human subjects. For example, a large second-hand car seller could generate individual promotional videos for thousands of vehicles instantly, simply by feeding text to the video generator.
Alternatively, video adverts could be targeted directly at a viewer, addressing them by name, or a weatherman could deliver a forecast for your street. “We can create 100 million videos for 100 million sets of eyeballs,” Aharon says.
Eze Vidra, a former Google executive who now runs venture capital firm Reimagine Ventures, says: “You would never bring a human to a studio for something like this because of the cost and the logistics, but synthetic video allows you to do it. It brings the human element to this very cold digital experience.”
While it currently uses unknown individuals to create its videos, Aharon said the company is now working with social media influencers and celebrities, which could mean them appearing in adverts without having to spend the time it takes to film them.
And because virtual actors never age, they could stay in work long after they otherwise could. Forget the de-ageing technology used by Martin Scorsese to wind back the clock on Robert De Niro in The Irishman: a synthetic De Niro could be any age, and star in hundreds of films a year.
A virtual doppelganger for everyone
In future, synthetic humans may not merely be for celebrity replicas, however. Soon we may all have virtual doppelgangers.
Facebook’s Reality Lab in Pittsburgh has been working on ultra-accurate virtual reality technology to capture what it calls “Codec Avatars”. This uses machine learning to capture, learn and recreate human social interaction.
The technology is aimed at eventually building an accurate digital copy of a normal Facebook user, so they might make calls or take meetings in virtual reality. Even now, VR headsets can pick up the minutiae of a person’s movement via head-tracking, but the expressions of a digital avatar would be recreated with AI.
This rise of synthetic video has led to questions about whether it could be used to deceive viewers in a way that goes beyond the altered “deepfake” videos that have already been used for political purposes.
Deepfakes currently manipulate source content to create its videos, so determining a fake is often as simple as finding the original. Full virtual humans created using AI could be more flexible and tougher to identify.
Hour One says it requires all of its videos to carry an “altered visuals” disclaimer saying that it contains computer-generated footage.
The company also says it gives the real people from whom it creates its likenesses the chance to choose what kind of videos their digital replicas appear in and that its licences to use an individual’s likeness expire after four years.
“We have a great technology that can save time and cost but you need to do it in a responsible way,” says Aharon. Others might not be so upstanding.
Source: Thanks smh.com