YouTube's Content Identification Failure Raises Eyebrows

MSNBC is carrying a story looking at YouTube's failure to follow through with a promised 'content identification system' by the end of the year. The article goes on to discuss the possible impact this failure will have on the site's (so far) good relations with television, music, and movie studios. From the article: "If the delay lasts for more than a week or two into the new year, suggesting more than just a slight technical hitch, 'this is certainly going to be a serious issue', [Mike McGuire, a digital media analyst at Gartner] added. Leading music companies have already made clear they see completion of YouTube's anti-piracy technology as an important step in any closer co-operation. Failure to build adequate systems to protect copyright owners could also add to the risk of legal action against the site."

Google and Youtube aren't that dumb

[Salvance]

It's hard to believe that Google hasn't already discussed the delay and any consequences with the movie, television, and music studios. Google had such intensive conversations with them before purchasing YouTube, that it would be silly if they went quiet and just let things slide.

Lawyers Shouldn't Set Tech Deadlines

[spike2131]

I pity the developers who are making this product. They have been given a complex task and an arbitrarily chosen deadline, probably pulled out of the air by marketing/legal/upper management. Since September they have been on a death march to meet this date, sacrificing family time around the holiday season.

But you know what? It just ain't ready because it was a fools errand to begin with. My guess is they are working off of half-assed specs that weren't even ready before Thanksgiving. Maybe in a few more months they can have something good. But media partners getting pissy about it isn't going to help the code mature any faster.

Re:Lawyers Shouldn't Set Tech Deadlines

[Herr Ziffer]

The technology isn't there yet. There are other companies working toward the same goal of media fingerprinting for much longer than YouTube has. For a sufficiently long media clip, it can be done. There serious problem, though, is with smaller clips. 30 seconds just isn't enough material, currently, to get a good match. Add to that the fact that the original clips get resampled and distorted and overdubbed. YouTube may be getting a break from media companies simply "because" it is so easy to make the argument that this was never feasible in the first place.

Re:Is it possible?

[Rob86TA]

The thing is, the MPAA and etc don't care if there are false positives, they only care that they are no escapes. Youtube could probably deploy a solution that would make the MPAA happy, only to have its own users leave as valid content was always accidently being blocked.

Enforce That !

[leftcase]

Given that the media and entertainment industry has made such a miserable job of enforcing copyright since the emergence high speed internet, perhaps their efforts would be better spent figuring out ways to capitalise on the presence of sites such as youtube and myspace.

If businesses such as Red Hat can make a living from open-source software, surely there's a more refined way for said media businesses to realise capital from their assets without being so 'grabby'!

No, it's not possible.

[twitter]

... is it really possible to do an automated content identifier/filter solution?

To take away your fair use they would have to fingerprint both the audio and video content. That's possible for whole works at a given frame size, rate and audio quality. Already, you can see the problem because there's an almost unlimited choice of those. Couple that problem to every length variation and you have an impossible task for any single work. The database of fingerprints would be infinitely large. You can multiply this infinite sized database time the hundreds of thousands of works the crackpots want to "protect" for a result thats that many times less practical. Policing for original works based on someone else's "intellectual property," such as a Star Wars parody, is clearly impossible. The already impractical task of making fingerprints of each submission is trivial by comparison. Even if they could fingerprint all submissions, there is no way they can match it to their satisfaction. Policing will require AI or a human inspector because the "crime" is sharing the details of a story, something only a person can recognize. If they do make it work, the first thing it will do is point to the blatant theft of concepts by every movie ever made, such as Star War's liberal use of "Triumph of Will", "Forbidden Planet" and several WWII films.

Re:Is it possible?

[MindStalker]

I'm betting they go with a computer/human pair system. If it matches close to 100% to a known video treat it as if it were the known video. If it matches greater than 50% have a human look at it. If it matches less keep it and wait for a user to flag it. Realistically most youtube videos are near carbon copies of other videos on youtube already. This would greatly decrease dups at least.

Re:It's all Utube Has

[symbolic]

I totally disagree. I rarely pay any attention to the copyrighted stuff, because that's exactly what I'm trying to get away from. The only way that I'd agree with you relates to situations where someone has used a copyrighted work to produce something derivative - like a spoof of a music video, or some music in a home-made video trailer.

Youtubs is a threat - I don't think it's a threat because people use copyrighted material in this manner, it's a threat because it moves the entertainment decision-making process from the few that used to have nearly complete control, to the end user. It's another paradigm shift that will be fought tooth and nail by the old guard.

Re:No, it's not possible.

[Jerf]

I Am Not A Mathematician either but I'm closer than the vast majority of people on Slashdot. (I've studied this stuff in a formal setting and done some limited work in the field of handling wildly multidimensional data.)

This reply is much more reasonable, and much closer to the truth. One of the missing pieces of your first post is the problem of making attacker-resistant fingerprints. Fingerprinting is actually not so hard when you haven't got people actively trying to hurt the fingerprint and you can accept a reasonable (and small) rate of false positives. It's not even that hard to make it fairly stable under certain easy transforms like a volume modification.

Making it attacker-resistant is as hard as you say; it's not that a fingerprint function can't be created for each of the attacks you mention, it's that covering them all at once is hard. The easiest thing to do is simply make the fingerprints cover more stuff ("fuzzing" the fingerprint is a pretty good mental model), which definitely increases the false-positive rate on audio. (Video doesn't suffer from this quite so badly because it has much more data to work with, therefore videos are "farther apart", and can tolerate much more "fuzzing". The flip side is dealing with this extra data can be a pain and it does open up some other attack avenues.)

Re:Easiest code EVAR

[ifrag]

That's just a joke about the MD5's right? Even the most simple edit to a clip would cause it to change, such as clipping blank frames from the start or end.