Open your eyes to the ‘deep fake’ world – where seeing isn’t necessarily believing

collapse of trust among Americans in their public institutions. No wonder MediFor is funding research it hopes will lead to easy detection of fakes.

Prof Hany Farid, a computer scientist whose own image analysis work is helping expunge child abuse pornography from the internet, says the kind of tech Darpa is pinning its hopes on is “decades away”.

While we wait, the ramifications of video fakery are already becoming clear. In February this year, 17 students and teachers were killed at Marjory Stoneman Douglas High School in Parkland, Florida. Afterwards, one survivor – Emma Gonzalez – posted a video in which she tore up a paper shooting range target. But the video was then doctored to make it appear that she had torn up a copy of the US constitution. The fake went viral on YouTube, and Gonzalez later said she had received death threats.

So is deep fake technology already good enough to meddle with, say, America’s 2018 midterm elections. The answer is yes, probably. “There’s no reason why a well-resourced state player could not have taken the technology that’s already out there and improved it to the point where they could use it to make something that is potentially quite damaging,” says Gibson.

He and others say context is as important as quality. The perfect strategy would be to fake a video in which a political leader is made to say something explosive in an apparently unguarded moment – rather like Gordon Brown’s “bigoted woman” comment about Gillian Duffy.

It could be presented as scratchy mobile phone footage captured by a member of the pubic, say, and so could be grainy or blurry in parts, making the faked video component easier to pass off as real.

Even today, says Gibson: “It is fairly easy to imagine circumstances in which you could synthesise Theresa May saying something about Brexit, off-guard after a dinner, that would cause her to lose the leadership of her party fairly quickly.” And Gibson is a former staffer at No 10.

If deep fakes do emerge in time to play a part in the midterms, however, Donald Trump may not mind. Because there is a flip side to the damage that fake video and audio can do to honourable politicians. Once the technology is out there, those in power caught doing something they wished they hadn’t will simply be able to dismiss hitherto indisputable video or audio evidence of their misdeeds as faked – even if it is true.

The upshot is that video, for so long the unsurpassed signifier of truth, is now on the cusp of becoming just another weapon in today’s “he-said, she-said” communication wars, where every fact is disputable, every lie convincing.

Eventually, consumers may get better at spotting these deep fakes. But it is likely to take time. It has taken almost 30 years for readers of glossy magazines to become aware that images can be doctored using Photoshop. No one knows how long it will take for a similar process to occur with video fakery. Nor exactly how the transition period between now and then will play out. “It will be rough,” says Ilya Feige, head of research at ASI. “But that roughness can be mitigated.”

He wants to set up websites where children are encouraged to experiment with deep fakes, the better to learn about them. He also thinks deep fakes spell a new, potentially golden era for media brands. “Maybe in future you just have to trust brands with ground truth content like video too”, says Feige. “I trust the words I read when I read The Telegraph even though words are totally fakeable.”

That sounds like one small step, but is in fact a giant conceptual leap. From the moon landings to 9/11, video has been the guarantor of credibility, playing an irreplaceable role in convincing audiences that extraordinary events are true.

Now mankind must adapt to a new world: one in which video can be as deceptive as it is ubiquitous; in which presidents, generals, celebrities, sport stars and countless other figureheads cannot be taken at their word; one, ultimately, in which seeing is not necessarily believing.