A random guy capable of writing a few AI prompts and a subscription to a bot—that’s all it takes to make political deepfakes now, as the world’s biggest democracies head for national elections next year. Without regulation, artificial intelligence based on machine learning could precipitate an unprecedented crisis of confidence in elections in India and the US.
An AI-generated deepfake blitz would accelerate the ongoing erosion of public confidence in the news, wearing out by sheer volume people’s capacity to identify machine-made lies. Indifference to truth and falsehood would increase polarisation, allowing electoral contenders and officials to believe or disbelieve facts according to expediency. The need to acknowledge the truth of an opponent would vanish, and with it, the need for the dialogue itself, the basis of democracy.
The general view of AI is dystopian, but occasionally, there are supportive arguments like California law teacher Orly Lobel’s book, The Equality Machine, which has closely followed the fortunes of AI. It was released in the US in October 2022, just a month before OpenAI unveiled ChatGPT-3, roiling the digital industry and raising public forebodings. The book, whose Indian edition is now out, dismisses dystopian projections as alarmist and suggests that within a framework of public policy based on equity, artificial intelligence and machine learning can be forces of good—and should be seen as public goods. Of course, that’s a reformulation of the old observation that technology is value-neutral and will pursue the moral imperatives of its maker, whether it’s swords or ploughshares.
The book’s India launch is equally timely, just after Aaj Tak powered up the country’s first AI news anchor, raising concerns about job losses and propaganda poisoning news at scale—though some human journalists have prospered in their jobs precisely by doing propaganda. If you’re writing about AI, it’s hard to go off-key because every day, someone, somewhere, is launching something shiny like Aaj Tak’s AI, a PYT named Sana. Her minder describes her as “bright, gorgeous, ageless, tireless”—more perfect than humans, precisely how AIs are described in dystopian sci-fi. Guess they don’t read much of that over at India Today because this was not said in irony.
Despite her minder’s enthusiasm, Sana is a trivial headline vocaliser. It is even less revolutionary than the AI that Bloomberg has been using to turn stock ticker data into small news items in real-time, a job that used to be handled less efficiently by sub-editors. They naturally hate such work, and it’s easy to teach a machine pink paper jargon like ‘realty heads north despite global headwinds’.
But at about the same time, popular music was shaken up by the arrival of a song by an AI that had learned to channel Drake. Alarm bells rang in the corridors of commerce as music industry executives and composers privately admitted that the AI Drake was perhaps more engaging than the human Drake. The machine has evoked a response in listeners, and some say it has passed a Turing test couched in the language of music, which is a fancy way of saying that it can compete with humans.
Meanwhile, ChatGPT has been caught making up fake Guardian articles. And under the nom de plume of Aidan Marchine, journalist and writer Stephen Marche has published the novella Death of an Author, which was actually written by three AIs. The New York Times responded with the obituary of formula storytelling, as in romcoms and space opera. That’s where publishing and entertainment make their money, not with the modern classics, so this is going to hurt.
This is not how technology is supposed to play out. Its perceived role is to liberate humans from drudgery and leave them free to pursue a life of creativity. So there was a clamour for regulation in the music industry, and the AI Drake number was taken off platforms. At about the same time, US screenwriters, who have the advantage of being unionised, laid down the picket line to reserve the right to write TV shows.
But unions have rarely been able to resist replacing manual jobs with technology, especially if it offsets job losses with new opportunities. Desktop publishing and digital entertainment have been slashing jobs and closing down departments since the early Nineties, laying off a range of workers from typesetters to cinema projectionists. But it also enabled small presses and indie entertainment. It remains true despite the recent failure of Buzzfeed and Vice, digital media which tried to supplant old media and became as vulnerable to the caprices of capital as their predecessors. AI will accelerate this phenomenon by, for instance, making computer code and art from verbal guidelines. What will people do with such new abilities? Partly dystopian stuff. That’s human nature.
Technology problems are not solved by more technology but by laws and codes—by regulation. The question of intellectual property is deep: why is it normal for an acolyte to train with a master but a violation for an AI to be trained on Drake’s music? But the question of deepfake news—fiction written by machines for electoral advantage—assumes urgency as national elections near the US and India. Since free speech absolutism has become widely accepted, it may be difficult to step back to when it was agreed that liars deserve to be penalised, not merely exposed, for the greatest good of the greatest number—including the liars themselves.
Pratik Kanjilal
The India CableEditor of
(Tweets @pratik_k)
