Again, this is separate from the fair use question. I'm saying if the owner of Star Wars files a DMCA notice against some guy reading Hamlet in an empty room, we should have little patience for that.
If they wouldn't be penalized, why? If it's because Disney owns the rights to Star Wars, how would that be discovered? Making a complaint when you don't have a good faith belief you own the copyright to it already caries the penalty of perjury. Would the cost of this new penalty be more than that?
If they would get penalized, then how is the existence of fair use not relevant as far as the risk to copyright holders is concerned?
It seems like it could only confuse the discussion, and passing readers, and not help in any way.
Here's a sketch: an automated system finds (alleged) copyright infringement, a human watches the content to check if it does in fact contain e.g. Star Wars, a second human finds the matching content in their library, and a third human verifies the work by human 1 and 2 and writes up the DMCA notice.
The current system seems to have zero penalty for submitting false DMCA notices, imposing significant costs on anyone a rights-holder decides to target.
What's your proof that such a system has been realized anywhere on the internet, with a '~100% certainty' rate?
Or is this entirely conjecture?
The copyright/DMCA notice system currently has a lot of Type 1 (false positive) errors, because it costs rights-holders nothing to send a notice. Lots of legitimate and non-infringing videos/content/urls/sites are getting taken down because of absolute bullshit.
I'm suggesting the world would be better if we moved to a system with more Type 2 (false negative) errors; where occasionally an infringing piece of media is left up, but we see fewer false positives.
If literally zero types of information have such a system set up, then there simply is no reason to believe it will actually work in real life.
Unless you have some comprehensive proof or analysis that incorporates all known, real world, factors?
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.
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.