The Revolution Will Not Be Sora’d

why I don't expect video deepfakes to ever cause serious trouble
Feb 19, 2024

Video deepfakes remove the law of artificial nature that says, “If there’s a video of it, then it really happened.” We already don’t have a rule that says, “If there’s a photo of it, then it really happened,” and never really did; doctoring photos was always possible in one way or another, especially since photo-editing technology improved at about the same rate as photo-taking technology. This has had no effect on society-level trust to my knowledge.

Will video be different? In my opinion, no.

One thing that obviously won’t happen: an equilibrium where people now go around thinking “surely this video must be real!” and then it isn’t, with the exception of elderly people who were already getting scammed by the likes of emails and phone calls. There will of course be a brief period where scams happen before the news that deepfakes are good propagates to everyone, but I think there’s basically zero chance this becomes the permanent state of affairs. When “if there’s a video of it, then it really happened” becomes false, then people will stop believing it; it will become easily verifiable common knowledge that it isn’t true, and so it would be extremely implausible for everyone to go on believing it in the face of clear personal incentives not to.

But you could imagine the loss of that video-veracity assumption being problematic; people won’t blindly trust fake videos, but they won’t be able to conclude much at all either way from “a video of this thing exists.” Will that cause issues? Well, how often in real life do you see some incident whose veracity is disputed and ultimately confirmed only by video? I basically never hear of stuff like this.

The loss of video does mean that the last method of non-interactive verification of physical truth over the internet is gone. Even if video is rarely used as verification in the status quo, maybe it’s load-bearing as a backstop of last resort? And yet, thinking about it, I believe it isn’t. The only real problem I anticipate from fake video, in fact, is slightly hobbled sybil resistance. Trust propagates over the internet all the time, but crucially, it’s not tangible proof that’s propagating. It’s webs of interpersonal trust. Most verification of real events comes not from the spread of video of the event (almost zero percent of everything that ever happens gets filmed and posted on the internet), but from the spread of people talking about the event.