Regarding the discussion on deep fakes: There’s an assymetry here.
Deepfake capable models seem to be a done deal. They are small, can be hosted on a single GPU, even hosted by phones and laptops with efficient cores designed for AI. Regulating them so they won’t fake any image, voice, or some types of video with temporal stability seems difficult.
Digitally signing all the “real” content and/or registering hashes of real video when it happens to a public ledger or block chain may be ooms more viable.
(Ooms meaning you might need to invade every country on earth and control every computer to stop them)
When vcrs and mp3 players came out, the media companies were unable to meaningfully ban or regulate either technology. They were forced to change their business model to adapt.
You could make the same argument—that maybe the vendors of mp3 players should have withdrawn their product until they were unable to play pirated music. But that’s not what happened.
Regarding the discussion on deep fakes: There’s an assymetry here.
Deepfake capable models seem to be a done deal. They are small, can be hosted on a single GPU, even hosted by phones and laptops with efficient cores designed for AI. Regulating them so they won’t fake any image, voice, or some types of video with temporal stability seems difficult.
Digitally signing all the “real” content and/or registering hashes of real video when it happens to a public ledger or block chain may be ooms more viable.
(Ooms meaning you might need to invade every country on earth and control every computer to stop them)
When vcrs and mp3 players came out, the media companies were unable to meaningfully ban or regulate either technology. They were forced to change their business model to adapt.
You could make the same argument—that maybe the vendors of mp3 players should have withdrawn their product until they were unable to play pirated music. But that’s not what happened.