Deepfakery

Election​ season in the Trapped-Together Kingdom, and people are talking about politicians and parties, sort of. The talk isn’t always talk, as such. To put it another way, when was the last time someone told you a joke? When was the last time you saw a fresh quip inked on the door of a toilet? No, I can’t remember either. But when was the last time you clicked on a link to a funny or shocking video, or – charming old world intimacy! – were handed somebody’s phone with a clip playing and asked: ‘Seen this?’ Or more likely you weren’t asked anything at all, but expected just to watch and get it. Shared video culture has reached the point of smoothness where the clip is a unit of communication in itself, like a gag or an anecdote or a bit of graffiti. Walter Benjamin imagined an ideal book consisting of nothing but quotes. A conversation made up entirely of quotes was always more likely, but the internet has made possible for everyone a feat that previously could have been managed only by two people with prodigious memories and a tedious amount of shared reading. Shared video culture is quote culture that has arrived. And what quotes! – dialogue and scene reproduced perfectly every time by anyone with a smartphone and a warm finger.

The emergence of videos shared on social media as a dominant means of communication in election campaigns has been accompanied by dread warnings of their vulnerability to manipulation by unscrupulous agents. Technology has advanced to the point where it’s almost possible to create what alarmists are calling ‘deepfakes’: fake videos of politicians and other celebrities, constructed using CGI and doctored audio, that are impossible to tell from the real thing. It’s hard to know what is most worrying about the announcement by University of Washington researchers two years ago that they had worked out how to marry – convincingly – asynchronous voice recordings and video of the same person. Was it that they accompanied their announcement with real-seeming video of Barack Obama, as president, appearing to speak words he’d actually spoken decades earlier? Or that the work was funded by Google, Facebook, Samsung and Intel? This year another set of scientists devised software, part-funded by Adobe, that allowed them to alter what a person in a recorded video appeared to be saying simply by typing alternative text into a script box.

So far deepfakery lies in the realm of premature panic. ‘Almost possible’ isn’t the same as ‘possible’. Italians were startled in September when the TV comedy news show Striscia la notizia broadcast a clip of what looked and sounded like the former prime minister Matteo Renzi offering the bras d’honneur, with raspberry, to political opponents. Some Italians, viewing the clip out of context later on social media, thought it really was Renzi. But the rendering was far from perfect; and one of the weaknesses of deepfake fearmongering is that it assumes, along with immaculate fakery, a public so sceptical of politicians that it will refuse to privilege the real politician’s rejection of the fake over the fake itself. Of course, a large section of the public is ready to make exactly such a refusal. But that section of the public has a large overlap with conspiracists also eager to believe that real video is fake – footage of men landing on the moon, for instance – or that no video evidence is needed to support the obvious conclusion that George W. Bush ordered up 9/11.

Early in the current election campaign Bill Posters, an insurgent artist and subverter of commercial propaganda whose website includes a useful guide to hacking the advertising space on London bus stops, helped the think tank Future Advocacy release bogus videos of Jeremy Corbyn endorsing Boris Johnson for prime minister, and vice versa. The intention was to highlight the danger that deepfakes pose to democracy. Posters and another artist, Daniel Howe, used technology called VDR, or Video Dialogue Replacement, from an Israeli company, Canny AI, whose website offers to ‘replace the dialogue in any footage’. Like their earlier deepfakes of Mark Zuckerberg, Kim Kardashian and Donald Trump, the Corbyn and Johnson videos are eerily good, though not quite indistinguishable from genuine clips. Perhaps the biggest danger deepfakes present to the election process at this stage of their development is what Future Advocacy calls the ‘liar’s dividend’, where the heightened discourse around skilfully faked audio, video and images helps the powerful dismiss real and damaging audiovisual records of malfeasance as bogus artefacts – as with Prince Andrew’s attempts to suggest that photo evidence of him hanging out with an abused teenager is doctored. This kind of doubt-sowing is already having a political impact. Earlier this year, questions over the authenticity of a video of the absent president of Gabon, Ali Bongo, provided the pretext for a coup. Bongo’s opponents claimed the illness that was keeping the president out of the public eye was much more serious than had been admitted, and that the video, in which Bongo delivered a new year’s message, was a deepfake. (In terms of videos that are unquestionably deepfakes, the greatest harm isn’t being done to electorates by political rogues, but by the porn industry to famous women. According to the Amsterdam-based deepfake tracking organisation Deeptrace, 96 per cent of the almost 15,000 deepfake videos online consist of female celebrities made to seem, without their consent or knowledge, to be taking part in ‘real’ sex scenes.)

Still, in politics, the ecosystem of making and sharing short videos, mainly but by no means entirely on social media, is bigger than ever. The limited impact of deepfake videos on political campaigns in 2019 doesn’t mean there isn’t a lot of more old-fashioned manipulation going on. In Britain, where the perceived personalities of party leaders dominate campaigns to a degree few of us like to admit, the artifice doesn’t so much involve the use of digital processing power to create a realistic fake of a real person as the use of the now universal medium of the shaky, supposedly spur of the moment phone video to make an essentially fake personality seem real. Real, that is, in the sense of being somebody who a voter without wealth or influence beyond family, friends and workmates can relate to. A video four minutes and forty seconds long purporting to show a spontaneous, casual interview with Boris Johnson at Conservative Party campaign headquarters, conducted as he makes himself a mug of tea, alternates political topics with cosy lifestyle questions such as whether or not he likes Marmite. He claims to start each day by walking the dog, says the thing that surprises him most about being prime minister is that he can’t order a Thai curry in Downing Street, and when he’s asked if he prefers a roast or fish and chips replies oddly that you can’t beat fish and chips ‘on a cold night on a beach’. Cybernetics researchers speak of the phenomenon of the ‘uncanny valley’, that point in human reaction to lifelike robots where the cyborg’s increasing verisimilitude ceases to provoke increasing wonder and begins instead to creep people out. I have no reason to think the person in the Boris Johnson video isn’t Boris Johnson, yet when he proclaims that his favourite band is either the Rolling Stones or the Clash, the hackles rise on the back of my human neck. The video has been viewed 4.5 million times on Twitter alone.

At the heart of the rise of video-sharing political culture, and at the heart of its most pervasive manipulations, is a system that transcends the stereotype of a news- watching society divided between oldies sitting down to watch the ten o’clock news at the time it’s aired and young people who ‘get all their news online’. The more complex reality is of a churning exchange of micro-moments between mainstream TV and social media – of non-TV video clips pushing themselves onto the bulletins and the rolling news shows, and of recorded snippets of mainstream TV shows like Question Time and Good Morning Britain decontextualised, recontextualised and re-edited on Facebook, YouTube and Twitter. In Austria, the right-wing populist Freedom Party soared to kingmaker status on the back of its mastery of the art of video culture ‘politainment’, using comedy and folksy stories to sugar-coat messages of spite and bitterness. Yet when, earlier this year, the party was badly damaged by a covert video of its leader offering a woman he thought was the representative of a Russian oligarch future state contracts in exchange for favourable media coverage in the Austrian newspaper she said she was about to buy, it was two extremely old media operators, Der Spiegel and the Süddeutsche Zeitung, that released and originally contextualised the footage.

The dominant form of fakery in politainment, and political video-sharing generally, is the misleading edit. It requires very little technical skill, doesn’t depend on fancy algorithms and, while it’s easy enough to spot, there’s always an uneasy awareness on the side of those calling it out that all video is, by necessity, edited, and that all editing requires a subjective assessment of what is important and what isn’t. Sometimes the misleading edit is blatant, as with the Conservatives’ recent release of a video, based on footage culled from a Good Morning Britain interview, that appeared to show Labour’s Keir Starmer lost for words and licking his lips nervously in response to a question about his party’s Brexit policy. In reality, Starmer gave a full and articulate answer. The Conservatives did not apologise.

Sometimes the fakery is simpler and yet more subtle, its effect intended to be cumulative. The UK internet is awash with what you might call ‘plucky bulldog bites back’ videos, usually between five and 15 minutes long, featuring Faragist reliables like Nigel Farage himself, Jacob Rees-Mogg or Johnson. Each will be placed in a situation where they appear both to be endorsed by a highly respectable institution and surrounded by vicious enemies – Farage in the European Parliament, Rees-Mogg on the BBC being asked a hostile question by a student, Johnson facing a baying House of Commons. The videos will be edited in such a way as to begin with a brief provocative point from the ‘enemy’, followed by a long speech from the ‘bulldog’ that appears – because the video is cut so that it ends when the speech ends – to reduce his tormentors to impotent silence. When they speak publicly, politicians today know there will be cameras watching, so it is more important for them to imagine how their words will come across in one of these videos – like the end of a film where they deliver the triumphant speech – than to bother about the debate they’re supposed, institutionally, to be having.

These plucky bulldog videos, put out by any one of scores of click-hungry outlets and recycled endlessly on social media, are watched by hundreds of thousands, sometimes millions, of people. The videos come with a peculiar genre of title. Rather than ‘rebuts’ or ‘argues’ or ‘claims’, they use the language of male domination and victory: ‘Jacob Rees-Mogg OWNS student’, ‘Nigel Farage DESTROYS questioner,’ ‘Boris Johnson DEMOLISHES Corbyn.’ The mood of political video-sharing culture is a distinctive one. As well as the celebritisation and comedic glaze of politainment – the Conservatives themselves titled the Johnson quiz ‘Boris Johnson’s hilarious election advert’ – and the mixed martial arts language of the video titles, you can’t delve into this world for long without becoming aware that, for many, politics is just one small, strange peninsula in a vast hinterland of shared videos about music, video games and sport.

In the last days of the Roman Republic there was a similar anxiety on the part of the ruling elite to insert themselves into the fabric of mass entertainment, not only to identify themselves with the fierce joys and hates of the people, but, in modern terms, to measure and flaunt their likes, their views and their followers among the tribes whose votes they required. As Geoffrey Sumi describes it in Ceremony and Power (2005), the hostile silence that greeted Caesar when he entered the theatre in July 59 bce was a major political event. A month after Caesar’s assassination, Cicero, who was away from Rome, sent letters begging for political news. Failing that, he wanted a measure of what, and how much, the people were applauding. ‘How,’ he might have asked, ‘is it playing on social media?’