Originally published at: https://boingboing.net/2019/05/20/false-claims-with-impunity.html
…
5 Likes
I never have understood why this system is set up as it is. Even if Warner Chappell has rights to some of the audio content, that is a small fraction of the content, most of which they DON’T have rights to. If one copyright holder owns 1% of the content, and another owns 99% of the content, why does the 1% holder get 100% of the revenue?
7 Likes
To any machine learning experts here: How hard would it be to produce adversarial music that false-positives with ContentID despite sounding nothing like the matched audio? Or even better, a system to make any song imperceptibly false-positive?
Could be fun to mess with Youtube that way, complaining about ContentID (though probably unproductive, I admit)
5 Likes
Bullies. Fucking goon squad.
2 Likes
Such a fucked system. Guilty or not, it’s always the accused who has to suffer when bad faith actors “decide” to fuck with their income. “Decide” being a strong word, of course, as content ID requires no input or effort on the accuser’s part. Cunts.
3 Likes
Because they are the ones that will sue YouTube over those samples, even if they are fair use, and the content creators are explicitly barred from suing YouTube in the TOS. Changing any aspect of this system opens Google up to nearly unlimited legal liability from outfits like Warner Chappell. Remember that judgements in these music copyright cases can reach the billions and trillions of dollars for a small handful of offenses.
This sort of nonsense is what killed all of the early video streaming services, and YouTube has only managed to avoid it by bending over and dropping trow for any media cartel that comes knocking.
4 Likes
Or the other way around? An algorithm that makes any music (sample) sound completely unaltered to human ears, but won’t trigger a positive by the content-ID… That’d be more useful I guess.
3 Likes
It seems there’s a need for a robo-disputer system to rapid-fire back at them. Catch-22, while Warner Chappell can make baseless claims with no penalty, the robo-disputer would have to be accurate.
4 Likes
The problem is, it wouldn’t “mess with Youtube,” it would only screw you for uploading the videos, because Youtube doesn’t care if your video gets flagged as having content that “belongs” to someone else - it just switches ownership over to that entity. (And fighting the claim, even if a false positive, is near impossible. It would just be creating a headache for you.) As @Ghost says, it works better to fool the system with false negatives, though even in that case it doesn’t create much of a problem for Youtube unless someone manually makes a copyright claim. (Which, in a case like this, probably wouldn’t ever happen.)
1 Like
I understand the initial violation, but now Warner Chappell is receiving revenue from content THEY don’t own. Other than the samples in question, everything else in those videos is copyrighted by the vlogger. I think some sort of proportional split would make the most sense and everyone would get paid for their own content.
1 Like
Oh, I imagined creating an account and just uploading random innocuous stuff to trigger ContentID. The reason I suggested false-positives is because ContentID is definitely hardened against false-negative, that’s it’s whole thing, but I doubt false-positive is a major consideration.
A higher false-positive rate could spur the company to reduce the algorithm’s sensitivity as well.
I want to go watch his videos to support him, but I’m afraid the money will go to Warner Chappell instead. Damn.
This would be a case I would love to see before the courts, but the precedent is not in the favor of the Youtuber.
1 Like
There are almost zero repercussions for false positives, so they have little incentive to prevent them. Especially since that could increase the chances of a false negative and billion dollar lawsuit.
4 Likes
But, again - it’s no skin off their nose when there’s a false positive. It just means more videos’ revenue streams get hijacked by the media companies that already trigger tons of false positives as it is (by, for example, laying claim to birdsong, or all performances of public domain songs). Which is more or less how it’s supposed to work, as far as the corporate entities are concerned.
2 Likes
Copyright filters can’t always distinguish between different performances of the same work.
This suggests that subtle adversarial modification to make a sample sound different to ContentID but the same to humans will not work.
Does seem stupidly easy to fix…
For example, a small charge of a few pence/cents to register a copyright claim, repaid after 6 months if not/unsucessfully challenged and a greater fine if something is sucessfully challenged…
At the moment it’s weighted near 100% to potential copyright holders, whereas youtubers bear all the costs and live under a ‘guilty until proven innocent’ disaster area. Plus the holders can write bots to copyright-claim with absolute abandon on even the most tenuous of claims with no negatives whatsoever for them…
The big money has brought the law, again…
1 Like
This topic was automatically closed after 5 days. New replies are no longer allowed.
