6 ms·

> The company identified 32 copies of the document, all of which were removed and then reinstated > A separate request Quartz made to reinstate its own version

> The company identified 32 copies of the document, all of which were removed and then reinstated > A separate request Quartz made to reinstate its own versions of the document was resolved 17 minutes after we sent it, seemingly after an employee review. I... don't see the problem here? Sure, it'd be better if their detector never got it wrong, but in the absence of the Sufficiently Advanced Technology that would require, a speedy manual review process is the best anyone's got.

pacificmint 7y ago
> a speedy manual review process Does the law require a speedy manual process, and quantify what speedy means? Because we hear complaints about the large tech companie's manual exception processes quite often, and speedy is not usually a word that is used to describe them
- EvilTerran 7y ago
  I don't know off-hand. The claim that this "is the future the EU has voted for" was made by the article, not by me. I was, for the sake of argument, taking the article at its word on that - and observing that, in that light, its histrionics about Article 13 come across as misplaced. That being said, even if a speedy review process isn't required de jure, market forces should make it a de facto requirement. Over time, a platform with overzealous takedowns & inadequate appeals process will lose out to any competition which does better on either front. (Yes, I know it hasn't worked out like that with YouTube. They're close enough to a monopoly that market forces don't really apply, and that's a whole other problem.)
gizmo686 7y ago
>a speedy manual review process is the best anyone's got. This is a policy choice. We could not have copyright, and just leave the documents up. We could give platforms immunity, and enforce copyright by going after the individuals who uploaded without the rights to do so. We could give platforms immunity, but require them to respect court orders to remove specific copyright documents. We could give platforms immunity, but require them to remove copywrote conent when notified. (A la DMCA) We could give platforms immunity, require them to remove copywrote content, and require them restore content on a counter-claim and defer the dispute to the courts. What we are doing is requiring, by policy, platforms to proactivly remove copywrote material but making no policy requirements against over-removal. If we are seeing the best possible implementation of this policy, then maybe it is simply a bad policy.
- schoen 7y ago
  > copywrote The past participle of "copyright" is "copyrighted" (note that it's derived from "right" rather than "write").
- zamadatix 7y ago
  Something can be a "bad policy" and still be "the best anyone's got". Efficient operation, speedy compliance, copyright enforcement. Pick 2 and watch the other group say policy is a choice and you've chosen badly.
mirimir 7y ago
There's no need for a bloody AI detector. If someone claims copyright, they submit a take-down request. And they must supply documentation to support that request. People who have posted stuff first see the take-down request. If they object, they supply supporting documentation. If necessary, the matter goes to arbitration, or perhaps trial. Stuff gets taken down only if the take-down request is not contested, or if the arbiter (or court) rules that it ought to be. Also, for those who submit too many bogus take-down requests, future take-down requests get black-holed.
- eridius 7y ago
  They need a detector to detect the case of someone re-uploading a copyrighted material that has previously been taken down. But there's no excuse for running this automatically against works that have not had a take-down notice submitted.
  - stale2002 7y ago
    No they don't need that either. If someone wants that document taken down another time, then they should send take down requests each and every time.
    - eridius 7y ago
      And then Scribd would get sued by publishers for making it too easy to simply repost content that they've been ordered to take down.
      - mirimir 7y ago
        Well, not if publishers had immunity.
    - Silhouette 7y ago
      If someone wants that document taken down another time, then they should send take down requests each and every time. As someone who has actually spent several days sleeping in shifts with a co-founder so we could keep sending takedown notices when the same group of people repeatedly uploaded our material to YouTube and Google apparently didn't care, I think we can safely assume that your proposal is not a viable general solution to this sort of problem.
contravariant 7y ago
The problem wasn't that their AI was imperfect. The problem is twofold. Firstly there are import decisions that are being based on an AI that's apparently not up to the task. Having no familiarity with concepts like public domain and fair use. Secondly there is an AI that is granted supreme power to censor content. Not capable of legislating copyright, but more than capable of forcibly removing certain content.
- EvilTerran 7y ago
  > there is an AI that is granted supreme power to censor content Its decisions were overruled by humans. That's not "supreme power". In effect, some uploads were pre-moderated instead of retromodded. There's nothing novel about pre-moderation, some sites apply it to all user content - if you don't have a problem with that, I don't see how it's consistent to take exception to this.

pacificmint 7y ago

> a speedy manual review process Does the law require a speedy manual process, and quantify what speedy means? Because we hear complaints about the large tech companie's manual exception processes quite often, and speedy is not usually a word that is used to describe them

EvilTerran 7y ago
I don't know off-hand. The claim that this "is the future the EU has voted for" was made by the article, not by me. I was, for the sake of argument, taking the article at its word on that - and observing that, in that light, its histrionics about Article 13 come across as misplaced. That being said, even if a speedy review process isn't required de jure, market forces should make it a de facto requirement. Over time, a platform with overzealous takedowns & inadequate appeals process will lose out to any competition which does better on either front. (Yes, I know it hasn't worked out like that with YouTube. They're close enough to a monopoly that market forces don't really apply, and that's a whole other problem.)

gizmo686 7y ago

>a speedy manual review process is the best anyone's got. This is a policy choice. We could not have copyright, and just leave the documents up. We could give platforms immunity, and enforce copyright by going after the individuals who uploaded without the rights to do so. We could give platforms immunity, but require them to respect court orders to remove specific copyright documents. We could give platforms immunity, but require them to remove copywrote conent when notified. (A la DMCA) We could give platforms immunity, require them to remove copywrote content, and require them restore content on a counter-claim and defer the dispute to the courts. What we are doing is requiring, by policy, platforms to proactivly remove copywrote material but making no policy requirements against over-removal. If we are seeing the best possible implementation of this policy, then maybe it is simply a bad policy.

schoen 7y ago
> copywrote The past participle of "copyright" is "copyrighted" (note that it's derived from "right" rather than "write").
zamadatix 7y ago
Something can be a "bad policy" and still be "the best anyone's got". Efficient operation, speedy compliance, copyright enforcement. Pick 2 and watch the other group say policy is a choice and you've chosen badly.

mirimir 7y ago

There's no need for a bloody AI detector. If someone claims copyright, they submit a take-down request. And they must supply documentation to support that request. People who have posted stuff first see the take-down request. If they object, they supply supporting documentation. If necessary, the matter goes to arbitration, or perhaps trial. Stuff gets taken down only if the take-down request is not contested, or if the arbiter (or court) rules that it ought to be. Also, for those who submit too many bogus take-down requests, future take-down requests get black-holed.

eridius 7y ago
They need a detector to detect the case of someone re-uploading a copyrighted material that has previously been taken down. But there's no excuse for running this automatically against works that have not had a take-down notice submitted.
- stale2002 7y ago
  No they don't need that either. If someone wants that document taken down another time, then they should send take down requests each and every time.
  - eridius 7y ago
    And then Scribd would get sued by publishers for making it too easy to simply repost content that they've been ordered to take down.
    - mirimir 7y ago
      Well, not if publishers had immunity.
  - Silhouette 7y ago
    If someone wants that document taken down another time, then they should send take down requests each and every time. As someone who has actually spent several days sleeping in shifts with a co-founder so we could keep sending takedown notices when the same group of people repeatedly uploaded our material to YouTube and Google apparently didn't care, I think we can safely assume that your proposal is not a viable general solution to this sort of problem.

contravariant 7y ago

The problem wasn't that their AI was imperfect. The problem is twofold. Firstly there are import decisions that are being based on an AI that's apparently not up to the task. Having no familiarity with concepts like public domain and fair use. Secondly there is an AI that is granted supreme power to censor content. Not capable of legislating copyright, but more than capable of forcibly removing certain content.

EvilTerran 7y ago
> there is an AI that is granted supreme power to censor content Its decisions were overruled by humans. That's not "supreme power". In effect, some uploads were pre-moderated instead of retromodded. There's nothing novel about pre-moderation, some sites apply it to all user content - if you don't have a problem with that, I don't see how it's consistent to take exception to this.