The deluge of convincing AI-generated images has only just begun

Graphic fake images of Taylor Swift began spreading across X this week. Picture: Fernando Leon/TAS23/Getty Images

Joe Biden, for all his many fine attributes, is not super at talking. He slurs, he stumbles, he drifts. At times, he threatens to go the full Grandpa Simpson. So you might think an AI-generated fake that sounds like him but also gets all the words right could do him a favour.

Not so in New Hampshire. Over the past few days, a deluge of fake pornographic images of the pop star Taylor Swift has attracted huge global condemnation, but they should not have been the big deepfake story of the moment. That honour ought to go to an automated call sent to voters shortly before the primary last week, in which an artificial voice that sounded as though it belonged to the president told people to save their vote for November. “What a bunch of malarky,” it began. What followed was stilted, robotic and uninspiring. But then, isn’t it always?

US President Joe Biden, for all his many fine attributes, is not super at talking. Picture: Saul Loeb AFP

It is possible, sometimes, to see something coming a long way off and still be flummoxed about what to do about it. I was made into an AI avatar myself for a magazine feature last year, an experience that was more eerie than terrifying, although only just. I first started writing about the threat of fake news shortly after the 2016 US election and I have learnt since that the great challenge is persuading people to care.

Partly this is because the roots of disinformation, even when you can find them, tend to be convoluted and bizarre. On rare occasions you can trace fakery back to malign state actors, such as the Russian Internet Research Agency troll factory. A more typical example, though, back in 2016, was a preposterous story about Biden threatening to barricade himself in the Oval Office and never leave, shared on Facebook more than 100,000 times. Created by a Canadian satirist, it was then mocked up as real by Macedonian teenagers chasing advertising revenue clicks, to be circulated by older, internet-naive users who had not read it properly. Suggest that 60-plus million Americans voted for Donald Trump because of stuff like this, and you sound preposterous. Yet suggest that it made zero difference and I think you are missing something.

Another problem is that once our decisions are made we are all incredibly poor at properly comprehending why we have made them. Tell people they have been fooled and they are apt to both bristle and bargain. In New Hampshire, if there were any Democrats who opted not to vote after hearing Robot Joe, I would wager that by now they have convinced themselves that they were not going to vote anyway.

Or consider the horrible image of a baby amid rubble that circulated at the start of Israel’s current assault on Gaza. Spread thousands of times, it has appeared on placards at demonstrations and, through that, into newspapers worldwide. Like an awful lot of viral Gaza images, it was in fact AI-generated, and fairly obviously so, not least because the baby had too many fingers. For those who believed it, though, so what? Doubtless many would now counter that there have been many babies crawling through rubble these past months, just not that one.

Fakery is already playing a role in elections. Last year in Indonesia, spoof footage of monoglot presidential candidates speaking fluent Arabic was circulated, presumably by their supporters. In Bangladesh, conversely, one female politician was hit by mocked up pictures of her posing in a bikini, and another in a swimming pool.

It is not hard to envisage how something like the New Hampshire call could affect our own elections. Imagine a fake audio recording of Sir Keir Starmer in his lawyer days, expressing sympathy for terrorist suspects. Or of Rishi Sunak speaking to the American healthcare industry and agreeing to turn the NHS into an app. Eventually they would be debunked but anybody fooled would be likely to keep on telling themselves that these guys probably would say those things, even if in these cases they had not. As in, even when the lie is exposed, do not imagine that minds will change.

Britain's Prime Minister Rishi Sunak and Joe Biden. Picture: Leon Neal/Getty Images

Smarter laws would help. In November, faked audio spread of the Sadiq Khan, the London mayor, saying “I don’t give a flying s*** about the Remembrance weekend” and suggesting that it should be postponed so as not to clash with the more important business of marching for Palestine. According to the Metropolitan Police, this was not a crime. Yet even if the Electoral Commission gets its way and laws are toughened up, the puzzle remains of whom to blame.

Back in the first wave of Trumpish and Brexity fake news, one bleak lesson learnt was that this stuff could be coming from almost anywhere. At the time Facebook even let you pay for western political adverts in roubles. Probably you remember Cambridge Analytica, the cowboy electioneering firm that was, for a time, everybody’s favourite villain. When they meddled in Nigeria’s 2015 election the client was neither candidate but a supposedly unconnected local businessman.

It is true that the world did not end after 2016, much as it may have got weirder and nastier. Laws were tightened and social media firms became more discerning about what they would share on behalf of whom. More importantly, even the most naive internet users have probably swapped maximum credulity for paranoid scepticism, at least when it comes to text. This brings its own challenges, because trusting absolutely nothing is not hugely more healthy than trusting absolutely everything. Either way, the most optimistic long-term picture is probably of that same painful learning curve, this time with video and voice. In the shorter term, I think you won’t believe what you’re about to see and hear. Or, worse, you will.

The Times