The societal benefits we've enjoyed from copyright law have been substantial, but the upside is completely maxed out at this point. The tail has been wagging the dog since the MPAA and RIAA grew into de-facto government agencies.
The potential societal benefits to AI are unbounded, but only if it's allowed to develop without restrictions that artificially favor legacy interests.
Any decision or legislation that says that training is not fair use -- and yes, that includes gaining access to the content in the first place by any means necessary -- will have net-negative effects on the society that enforces it.
> The potential societal benefits to AI are unbounded, but only if it's allowed to develop without restrictions that artificially favor legacy interests.
That's a very strong claim based on currently limited evidence.
It's in no way clear that AI has an infinite ability to scale capability, nor that that can only be done by completely ignoring compensating those who provide training data.
OpenAI and Anthropic would love that to be true... but the facts don't support it.
It's entirely possible for something to be suboptimal in the specific (I would like this thing for free), but optimal on the whole (society benefits from this thing not being free).