4 ms·

“YouTube: "identifying copyrighted material can't be an automated process." Startup disagrees” Things have changed a little since 2007!

wyclif 5y ago
Also, that "The Desktop Is Dead" story on Wired is pretty funny in retrospect.
O_H_E 5y ago
Loool wow that is hilariously ironic
codetrotter 5y ago
The article is gone now, but here’s an archived copy of it on IA https://web.archive.org/web/20070221023808/http://www.iht.com/articles/2007/02/19/business/piracy.php https://web.archive.org/web/20070221023808/http://www.iht.co... Supposedly there was a page 2 of the article, but even though the first attempted capture of page 2 by the IA was only a few days after the article was pusblished, page 2 was returning 404 to IA.
southerntofu 5y ago
For people who don't understand: What changed is that Youtube acquired such startups, and started developing automated censorship based on copyright which has both a lot of false positives as well as venues for manual abuse when someone claims to be a rightsholder when they're not. Turns out Youtube was right in the first place: this can't be an automated process, unless you accept more than half the videos that will be deleted have nothing to do with copyright infringement.
- throwawaynay 5y ago
  it can if you're not cutting corners
  - oarsinsync 5y ago
    If you can build it and prove this to be true, you'll likely become a multimillionaire when you get acquired.
  - southerntofu 5y ago
    No, it's impossible. AI-based algorithms are fairly bad at recognizing copyrighted works. It's hard to strike a balance between what is an original composition and what's a remix/cover/copy: - if you make it too strict you'll just find bit-by-bit copies: reencoding or slightly altering the data would pass screening easily - if you make it too lax, you will find that many "original compositions" are in fact very similar to or inspired by other works (oh really that bird singing i just recorded is copyrighted?!) Beyond that, when there's an actual copy involved, an algorithm is fairly bad at determining whether this copy is legal: - if it looks very similar to another user's video which claims to have copyright on the content, however what proves the copyright is owned by that other uploader and not by a 3rd party? how can you make sure the license was not copyleft or public domain in the first place? - how do you know whether the content is used in a context of lawful criticism? parody? - how do you know whether the usage is tied to an education or accessibility exception to copyright? - how do you know if the usage of the copyrighted content is legal in the uploader's country? in the copyright owner's country? in the host's country? how do you decide which is more important? If censorship is bad, automated censorship is even worse. There's no way to make it right.
    - hnlmorg 5y ago
      Even courts struggle to agree on all of the edge case complaints about "copied" work. There's no way an algorithm would manage it
      - hutzlibu 5y ago
        Since you can apparently copyright "silence", no one should wonder that an algorithm cannot understand those decisions as they are quite arbitary.

wyclif 5y ago

Also, that "The Desktop Is Dead" story on Wired is pretty funny in retrospect.

O_H_E 5y ago

Loool wow that is hilariously ironic

codetrotter 5y ago

The article is gone now, but here’s an archived copy of it on IA https://web.archive.org/web/20070221023808/http://www.iht.com/articles/2007/02/19/business/piracy.php https://web.archive.org/web/20070221023808/http://www.iht.co... Supposedly there was a page 2 of the article, but even though the first attempted capture of page 2 by the IA was only a few days after the article was pusblished, page 2 was returning 404 to IA.

southerntofu 5y ago

For people who don't understand: What changed is that Youtube acquired such startups, and started developing automated censorship based on copyright which has both a lot of false positives as well as venues for manual abuse when someone claims to be a rightsholder when they're not. Turns out Youtube was right in the first place: this can't be an automated process, unless you accept more than half the videos that will be deleted have nothing to do with copyright infringement.

throwawaynay 5y ago
it can if you're not cutting corners
- oarsinsync 5y ago
  If you can build it and prove this to be true, you'll likely become a multimillionaire when you get acquired.
- southerntofu 5y ago
  No, it's impossible. AI-based algorithms are fairly bad at recognizing copyrighted works. It's hard to strike a balance between what is an original composition and what's a remix/cover/copy: - if you make it too strict you'll just find bit-by-bit copies: reencoding or slightly altering the data would pass screening easily - if you make it too lax, you will find that many "original compositions" are in fact very similar to or inspired by other works (oh really that bird singing i just recorded is copyrighted?!) Beyond that, when there's an actual copy involved, an algorithm is fairly bad at determining whether this copy is legal: - if it looks very similar to another user's video which claims to have copyright on the content, however what proves the copyright is owned by that other uploader and not by a 3rd party? how can you make sure the license was not copyleft or public domain in the first place? - how do you know whether the content is used in a context of lawful criticism? parody? - how do you know whether the usage is tied to an education or accessibility exception to copyright? - how do you know if the usage of the copyrighted content is legal in the uploader's country? in the copyright owner's country? in the host's country? how do you decide which is more important? If censorship is bad, automated censorship is even worse. There's no way to make it right.
  - hnlmorg 5y ago
    Even courts struggle to agree on all of the edge case complaints about "copied" work. There's no way an algorithm would manage it
    - hutzlibu 5y ago
      Since you can apparently copyright "silence", no one should wonder that an algorithm cannot understand those decisions as they are quite arbitary.