How we got Emma Thompson to make a pro-Brexit speech
... thanks to chilling new video technology which can literally put words into someone’s mouth. So how long before we can’t trust ANYTHING we see online?
THE face is unmistakable even if the words coming out of her mouth would cause her politically correct friends to choke on their quinoa. ‘Do we have a plan for Brexit? We do,’ she appears to say. ‘Are we ready for the effort it will take to see it through? We are!’ Arch-luvvie Emma Thompson is known, of course, for being an ardent Remainer, having described Brexit as ‘madness’ and Britain as a ‘cake-filled, misery-laden grey old island’.
So this latest astonishing footage of the millionaire Labour supporter regurgitating a speech by Prime Minister Theresa May could come as something of a surprise.
But all is not as it seems. The remarkable clip is, in fact, a video commissioned by The Mail on Sunday to illustrate the more insidious powers of cutting-edge artificial intelligence (AI) technology – which can now be used to create frighteningly realistic fake videos.
Known as ‘deepfakes’, they are so convincing – and salacious – that they can spread across social media in minutes, proving Winston Churchill right when the great wartime politician said that ‘a lie gets halfway around the world before the truth has a chance to get its pants on’.
The chilling potential of the technology was illustrated to great effect last month when Donald Trump shared a clip on Twitter, apparently showing Nancy Pelosi, the Democrat Speaker of the House of Representatives, slurring, stumbling and repeating her words at a press conference. There was one problem: the video had been doctored to make her appear drunk and incoherent.
Technically, the Pelosi film was a ‘shallowfake,’ made by simply speeding, slowing and altering the pitch of a real video of her speaking. Despite the crudeness of its production, the clip has been viewed millions of times on Facebook.
Deepfakes, however, can make anyone appear to do or say anything. A perfect recent example was a clip of Facebook boss Mark Zuckerberg which appeared online after the social media giant refused to take the Pelosi video down.
It showed a sinister-looking Zuckerberg gloating about his power and a made-up organisation called ‘Spectre’. ‘Imagine this for a second,’ the deepfake Zuckerberg says. ‘One man with total control of billions of people’s stolen data, all their secrets, their lives, their futures. I owe it all to Spectre. Spectre showed me that whoever controls the data, controls the future.’
Today, in an investigation that should terrify us all, The Mail on Sunday reveals just how easy it is to turn Emma Thompson from Remainer to Brexiteer with the technology. If we can no longer believe what we see, the
implications are far-reaching not just for people like Thompson, Zuckerberg and Pelosi, but for democracy itself. Making computer-generated images, known as CGIs, of real people is, of course, nothing new. A recent advert was spookily realistic, appearing to show Audrey Hepburn eating Galaxy chocolate while being driven along the Amalfi Coast. The advert was created by technical wizards who applied stateof-the-art CGI computer graphics typically used by Hollywood to enhance old footage. Deepfake technology, however, is different. It uses AI, computer software that effectively ‘learns’ how to do complex tasks, to quickly and cheaply manipulate images fed into the machine and make them appear to do or say something else. The Mail on Sunday commissioned a team at the University of Albany in New York to fuse a clip from several randomly chosen Emma Thompson videos and a Brexit speech made in March by the PM in which she called on MPs to back her Withdrawal Agreement. The software analysed the mouth movements of both women before generating the final clip.
Professor of Computer Science Siwei Lyu, at the University of Albany, who created the clip with student Yuezun Li, said: ‘To make these types of videos requires computer units that cost between $5,000 and $20,000 (£4,000 to £16,000).
‘There is no issue of accessibility to such technology, however, as there are plenty of services online where people make these videos on request. All you need to do is get high resolution images from YouTube and input them into the computer, which then generates the deepfake clip using algorithms.’
Prof Lyu said within five years, everyone could have access to the technology on their home computer, smartphone or tablet. ‘It will be doable in just a few hours,’ he said.
It is already advancing. At present, most deepfakes feature the apparent speaker lip-synced to someone else’s voice, often an impersonator. But soon, the technology will be able to deconstruct the various distinctive elements of any voice and put them back together to create phrases, or whole
speeches that the original speaker has never uttered.
It could be used for nefarious domestic purposes, such as making someone look like they are having affairs or carrying out illegal activity. But of more concern is its ability to put words in the mouths of political leaders which could deliberately set out to spark widespread fear, hatred and panic. For this reason, it is already being dubbed ‘the next generation of fake news’.
Digital expert Rafe Pilling, who works as a senior security researcher at digital security firm Secureworks, said: ‘It is likely to be available to the mainstream within five years, as a computer program or an app.
‘It could be as easy as applying an Instagram or Snapchat filter [which alters faces, and can appear to swap genders] is today. As soon as the first consumer application comes out, it will rapidly become ubiquitous.’
Professor Anthony Glees, Director of the Centre for Security and Intelligence Studies at the University of Buckingham, added: ‘The prospect of this is Orwellian, and then some.
‘It’s not just fake news, it’s giving out fake news in the images of people who we believe.’ Professor Glees is in no doubt what effect it could have.
‘This kind of Frankenstein technology will have a devastating impact on our politics,’ he said.
‘In the hands of an enemy that wants to sow discord and undermine our way of life – be it Russia, Iran, North Korea – the potential is terrifying. We’ve become used to treating words and images with caution, especially online, and lots of people see video footage as the only evidence they can trust.
‘As these deepfakes show us, in a few years even that will be gone.
‘The end result will be a collapse in the trust in political figures, which we need in order for our democracy to function.’