Yes, on real-world data, on a first pass, I doubt you'll get >99% accuracy. But you don't need to; you can do (something like) three passes, that each have ~98% accuracy, and the end result has effectively zero errors (of the type we care about).
Here's a system, in the real world, that does this currently: blood donation screening. We very much don't want HIV (and other pathogens) in the blood supply. So first pass - donors are screened out, based on various risk factors (risky sex, IV drug use, lived in certain areas, whatever). Second pass - antibody testing. Third pass - antigen and nucleic acid testing. If ANY of these register as 'likely to have HIV' the blood is rejected. In my country (Canada) this has meant zero instances of HIV transmission through donated blood. 100$ effective - but we have surely thrown away (or not collected) blood that was perfectly fine. That's the cost. But it works, in the real world, very, very well.
We could absolutely demand a similar level of accuracy for DMCA notices - the cost will be a) additional work for rights-holders and b) some infringement will be missed.
I don't think we're talking past each other, unless there' some confusion around the fact that politics, the judiciary, etc.., are real world factors.
And that even if an actually working system is set up, there's no guarantee it will still be functioning, as designed, past the first few days.
But even getting to the initial stage is already pure speculation, unless you have convincing proof/analysis. Considering that it needs to be accepted, and physically implemented by a real world government, not some hypothetical government of the distant future.
Yes, on real-world data, on a first pass, I doubt you'll get >99% accuracy. But you don't need to; you can do (something like) three passes, that each have ~98% accuracy, and the end result has effectively zero errors (of the type we care about).
Here's a system, in the real world, that does this currently: blood donation screening. We very much don't want HIV (and other pathogens) in the blood supply. So first pass - donors are screened out, based on various risk factors (risky sex, IV drug use, lived in certain areas, whatever). Second pass - antibody testing. Third pass - antigen and nucleic acid testing. If ANY of these register as 'likely to have HIV' the blood is rejected. In my country (Canada) this has meant zero instances of HIV transmission through donated blood. 100$ effective - but we have surely thrown away (or not collected) blood that was perfectly fine. That's the cost. But it works, in the real world, very, very well.
We could absolutely demand a similar level of accuracy for DMCA notices - the cost will be a) additional work for rights-holders and b) some infringement will be missed.