SEEING ISN'T BELIEVING

AI video generators are now so good you can no longer trust your eyes

- BRIAN X. CHEN

This month, OpenAl, the maker of the popular ChatGPT chatbot, graced the internet with a technology that most of us probably weren’t ready for.

The company released an app called Sora, which lets users instantly generate realistic-looking videos with artificial intelligence by typing a simple description, such as “police bodycam footage of a dog being arrested for stealing rib-eye at Costco”.

Sora, a free app on iPhones, has been as entertaining as it is has been disturbing. Since its release, lots of early adopters have posted videos for fun, like phony phone footage of a raccoon on an airplane or fights between Hollywood celebrities in the style of Japanese anime. (I, for one, enjoyed fabricating videos of a cat floating to heaven and a dog climbing rocks at a bouldering gym.)

Yet others have used the tool for more nefarious purposes, like spreading disinformation, including fake security footage of crimes that never happened.

The arrival of Sora, along with similar Al-powered video generators released by Meta and Google this year, has major implications. The tech could represent the end of visual fact — the idea that video could serve as an objective record of reality — as we know it. Society as a whole will have to treat videos with as much scepticism as people already do words.

In the past, consumers had more confidence that pictures were real (“Pics or it didn’t happen!”), and when images became easy to fake, video, which required much more skill to manipulate, became a standard tool for proving legitimacy. Now that’s out the door.

“Our brains are powerfully wired to believe what we see, but we can and must learn to pause and think now about whether a video, and really any media, is something that happened in the real world,’ said Ren Ng, a computer science professor at the University of California, Berkeley, who teaches courses on computational photography.