You can’t shake it off – deepfakes are here

THE saying “seeing is believing” was coined in the 17th century, but is an anachronistic oddity in the 21st one as we gingerly pick our way around deepfake images of Taylor Swift and the Pope. Now, artificial intelligence (AI)-generated mimicry is coming for the everyman too.

Last month, a high school athletic director in the US was arrested for framing his principal as a racist, by using AI to simulate the latter’s voice saying antisemitic things in a recording. Police eventually found that the accused had done it as payback for a poor work evaluation.

The deepfake recording, which went viral, showcased the exponential improvement of the technology. It had a deliberately tinny quality and additional background noise to recreate the illusion of having been surreptitiously recorded. The principal’s deepfaked voice also had the cadences and effects you’d expect in natural human speech.

It wasn’t as if the technology used to do this was weapons-grade stuff buried on the Dark Web. There is a plethora of them available to just about any garden-variety numpty with an axe to grind. With one service, US$11 a month gets you two hours of cloned audio in 29 languages from a perfectly legitimate firm funded by brand-name venture capitalists.

And even as AI generation tools are cheaper now, the length of voice samples needed to create a convincing fake is also shorter. Current commercial offerings require as little as one minute of an audio sample. OpenAI has a version that only needs a 15-second voice sample, which it has not unleashed on the public because of all the mayhem that might ensue.

Eventually, society does develop immunity to the pathogens in their midst. Already, people are establishing secret pass-phrases with their loved ones as a verification measure in the event of a deepfake kidnapping scam call. Internet users are now much more likely to look for the uncanny valley in implausible video and audio clips online.

This immunity, however, enables an unfortunate byproduct: the liar’s dividend. Here, liars reap the benefits from the plausible deniability of all the doubt that is being sowed. Since it is unclear what is true or fake, it is easy to disclaim even authentic evidence of wrongdoing.

It is also when the stakes are higher that the players involved have the clout and resources to milk a liar’s dividend. A powerful person running for public office could, say, insist that he never told anyone to grab anybody by the anything, because that recording was so fake, you know. Tremendous. Tremendously fake. Just very, very fake work, by some very, very bad people.

This completely hypothetical person will be believed, too, if for no other reason than that people fundamentally want to believe things that fit their worldview. The human condition does not do well with perpetual uncertainty and cognitive dissonance. For sanity’s sake, the average person is wont to pick one narrative out of sheer fatigue and stick to it come hell or high water.

The saying, “seeing is believing” is apparently an incomplete one. The full quote, purportedly by English writer Thomas Fuller, goes: “Seeing is believing, but feeling is the truth.” This might suffice for promoting religious faith, but since we are butchering quotes, here’s another: “In God we trust; all others, digitally watermark your AI output.”