How small a target will India aim at?
> requiring social media companies and other online publishers to detect and label AI-generated content.
Detecting this generated content is decidedly non-trivial, especially when you need to do it at scale: even "just" creating an automated system that counts the number of fingers in the picture, let alone one that detects all the continuity errors in a video clip. Written material is no easier: many human posts suffer from wandering attention spans, which is why the custard yesterday was a bit lumpy.
And as soon as any such system is built it can be attached into the training feedback loop and ta-da, indetectable generated content (the only barrier at this point is cost). You *can* apply the same technique to detectors, but that will always be on the losing side (you can't feed it known-generated-tags until you *know* it *was* generated and in the lag between getting hold of GuffGPT v19.7 and training to detect it, the election is over, the social media posts are history).
There is the option of gleefully pointing at Big Social Media and enjoying the idea that they'll have to bear the massive costs of attempting this impossible task India has set them. But they also have Big Lawyer slavering to take the dosh and tie up the courts.
Which leaves the question: will India aim at the small fry, say, anyone running a Mastodon instance? You know, all the social media *other* than the big names?
As much as we want to have AI guff labelled, this kind of law-making, putting the onus on the publishers, can it ever have any effect other than strengthening the grip of the big name players and driving out any upstart alternatives?
Biting the hand that feeds IT
