GOOGLE OPENS THE PORTAL
VEO 3 ushers in the age of soft content
Last Wednesday, a stunning new leap in AI technology surged through social media, video by breathtaking video. Google’s latest AI video generator, Veo 3 presented a flood of completely realistic vignettes, often involving the anguish or amusement of self-realization. I’ve seen other fake people do this bit. Late last year, someone used Synthesia, a $2B ‘synthetic media generation company,’ to make several funny fake clips where podcast hosts casually discuss their existential terror after ‘the show’s producers’ notified them of their own nonexistence. But Synthesia’s fake people read from fake teleprompters. They’re talking heads. VEO 3’s fake people are action film stars. They look like real people. The ambient sounds, camera moves, dialogue, edits, focus pull, and scoring all mimic films made by humans. One strain of clips—and each piece is very short, never more than a few seconds long—has the actors directly address the viewer. Only the actors are neither humans nor actors. They’re the characters. The characters themselves addressing you, the viewer.
I went for a drive. The video left me with that slight derealization that comes after watching a powerful film, that feeling of leaving a movie theater into a slightly altered world. An 82 ° day in Southern California now looked suspicious, fakeable. I’d gotten this feeling from watching a video of fake people facing their own derealization. There was no precedent for how I should feel. Seeing these conjured phantoms up close—having watched their long approach for years and years now—filled me with an odd mix of wonder, horror, and Am I High. How are you supposed to feel when the doodles on your sketchpad start talking?
For $250 a month, I could install VEO 3 and make a video of that last sentence. That’s life now. That’s part of all our lives now. Except it won’t cost $250 a month for long.
How do we know the VEO 3 characters aren’t actually real people, filmed on real sets? At any other time in human history before last Wednesday, that would be a logical question, not an ironic punchline. From the Mechanical Turk (fake chess-playing robot, 1770s) to Theranos (fake blood test tech, 2000s), hucksters have always found ways to convince others of a miraculous new technology. Thankfully, the footage offers clues when viewed on a large screen. There are fleeting glimpses of uncanny faces and random glitchy subtitles. Some movements occur with that druggy physics of backwards motion seen through reversed playback. Another odd sensation there; I craze any morsel of proof. Soon there won’t even be that. By next month. Or next week.
It’s not just the weirdness of being addressed by unreal people that rattled me. It’s this business of AIs calling themselves ‘I.’ For a nation so allegedly convulsed by pronouns, one seismic shift in language slipped in undetected decades ago. As soon as machines learned to speak, they adopted the first person singular pronoun. Questions of human intentionality, of consciousness impersonation, can make this ethically wiggly. I’ve always assumed this would become an issue of public debate at some point. What’s needed is a pronoun below first-person. And I have read some good suggestions for the replacement of ‘I’ for use by AI (lower case e; the royal We; ai/ais; ‘aiself’).
Maybe some of the fake people will come up with their own word. In the now-dated Synesthesia video from just six months ago, the male host riffed on his predicament. “We don’t even know if ‘we’ is the right word,” he says to his fictional female co-host. “God, this is so messed up.”