An AI that passes the video Turing test should not existCreating Fake People is a Terrible Idea

Tavos says its new Griffin model is the first AI to pass a video version of the Turing Test; the author argues such indistinguishable digital people must be banned, not launched.

Tavos, which calls itself a human computing company, has released Griffin — billed as the first AI to pass a video version of the Turing Test. Talk to it on screen and you cannot tell whether the other side is a person or a program. The author's verdict: this is a disaster.

Here is an analogy that makes it concrete. A decade ago, at a conference, a generative AI company pitched using the technology to keep Holocaust survivors alive: record long interviews, cut them into clips, and when a student asks something, serve up the relevant answer — or invent one.

From remembrance to counterfeiting, one step

The speaker got more and more excited. Why stop at survivors? Why not build an avatar of your late father, or preserve a child you lost in digital form? Nobody in the room flinched.

But a survivor never lived through today's world. Anything an AI says in their voice is not what they would have said. And keeping a dead child's avatar around is more likely to freeze a parent's grief than to ease it.

When trust goes, so do the savings

On platforms like X, real conversations are already peppered with AI-generated posts. Once you cannot tell whether a mind is behind the words, nothing can be taken on trust, and people may simply check out.

Scammers win big: a convincing face is enough to talk an older person out of their life savings. The author quotes philosopher Daniel Dennett — creating counterfeit digital people risks destroying our civilisation.

Tavos says it will add safety labels and keep working on alignment. The author is not reassured. The only safe option is never to build the thing.