> The distinctions between the inputs and outputs created by an AI on the one hand, and a human art student on the other, are too fine for judges to reliably analyze in the courtroom.

I don't understand why this comparison would be even looked at in the first place, it does not make sense to me.

A printing press can be used to infringe copyright - or not at all. That's why copyright uses a "black box" model: because the same user can sometimes be infringing, other times not at all and yet other times it's borderline and debatable. So copyright never had any other choice than this black box model. It could never care about the technology. It did try to attack some copy technologies sometimes but that never made sense and never worked.

A student can produce a drawing too similar and copyrighted - or not at all. Sometimes they will infringe and others not.

Exact same thing for AI output. There is no simple AI rule and there cannot be. It's still case-by-case.

> Technologies come and go, and the law adapts. But we don't get this flexibility for free. Judges and lawyers have to figure it out on a case-by-case basis, and they need a coherent theory of how that works. That theory is now the problem.
> Broadly and informally speaking, most analysis of infringement uses a black-box model...

That inability to scale has not changed. The only thing that changed is the _volume_ copyright infringement must now face. Copyright enforcement never "scaled". AI made that inability much more obvious but did not fundamentally changed the initial design flaw.

In a recent past, BitTorrent and friends already exposed another, different inability of copyright enforcement to scale. That inability was never "fixed". It was only mitigated through various means: targeting only the biggest offenders, threats, offering better, legal streaming alternatives...

One difference with today's AI challenge is not just the volume of infringements but also the very deep pockets of AI companies who can buy even more congressmen (thank you "Citizens" United vs FEC) than the copyright lobby can. Again: a scaling issue.

So maybe this is finally the end of copyright, dunno. But not because AI affected any of copyright's fundamental logic, but rather because it was never designed to scale in the first place - as you wrote yourself.