It’s fairly well established by now — thanks to court decisions like Viacom v. YouTube and UMG v. Veoh — that online service operators have no legal duty to proactively police their services for potential copyright infringement.  But that doesn’t mean that some services don’t do it anyway.  The biggest example is Google’s Content ID system for YouTube, which uses fingerprinting technology to flag uploads that contain copyrighted material.

The reason why Google implemented Content ID (in 2007) is simple: Google figured out a way to make money from it.  Copyright owners can choose to allow their content to be uploaded and take a share of revenue from ads that Google places in or alongside the video clips.  The major record companies participate in this arrangement for the vast majority of their content.  They aren’t thrilled with the per-stream revenue they are getting, but both sides agree that it’s better than having YouTube just block everything that matches.

Enter Facebook. Over the past couple of years, Facebook has become a bigger and bigger video-sharing service, one that is starting to rival YouTube in audience size and arguably exceed it in audience engagement.  This has led to “freebooting,” or capturing video streams from YouTube and re-posting them on Facebook.  And not just major record label or Hollywood studio content, but any popular YouTube video.

After a crescendo of complaints from native YouTube stars as well as the music industry, Facebook announced that it would be building a “Rights Manager” system based on the Audible Magic fingerprinting technology that it has been using for years to scan uploaded audio.  (YouTube also used Audible Magic before Content ID was implemented based on its own technology.)

Facebook’s Rights Manager allows copyright holders to “claim” their content and decide what they want Facebook to do with it.  Yet unlike Google’s Content ID, Rights Manager ultimately allows only two options: just allow the upload (and offer usage statistics to the copyright owner), or report it to the copyright owner as a potential violation.  There is no option to block the upload automatically.  Instead, rights holders must receive notices of matched content and then issue takedown notices to Facebook using Facebook’s DMCA process; then Facebook will “promptly remove those videos in response to valid reports.”  Facebook offers an API so that rights holders that have the wherewithal to do so — or to pay third-party copyright monitoring services — can build tools that receive reports of matching content and issue takedown notices automatically.

More recently, there have been reports of Facebook enhancing Rights Manager in the context of negotiating license agreements with the music industry.  Facebook will need to do three things in order to make record labels and music publishers comfortable.

First, Facebook will need to add a monetization option to Rights Manager — an option to allow uploads and pay rights holders royalties on them.  They could emulate Google’s Content ID and pay percentages of ad revenue, but the linkage between ads and content on Facebook is less straightforward than it is on YouTube, so this would be a complex scheme to negotiate and maintain.  An alternative would be to pay rights holders fixed royalties whenever someone views copyrighted content — an arrangement similar to the one Pinterest probably has with Getty Images for photo licensing.

Second, Facebook will need to provide an option to automatically block material that matches its fingerprint database, as Content ID does.  Major record labels use this option on YouTube with a small amount of current “frontlist” content where they want to direct attention to other sites — such as Warner Music Group and MTV.  The major labels will expect an automatic blocking option.

Although the deal terms may be complex, these two changes to Facebook’s Rights Manager are not difficult to implement.

The third one is a lot harder.  The NMPA (National Music Publishers Association, the trade group for U.S. music publishers) has raised concerns about the growing amount of videos of cover versions of copyrighted songs (compositions) being uploaded on Facebook without licensing.  Recording a cover version of a song that’s in copyright normally requires a mechanical license from the songwriter’s music publisher.  Under the law, the publisher can’t refuse to grant the mechanical license, but the performing artist must notify the publisher (if the publisher is known), and the artist must pay a standard royalty.  If the song is used in a video, then it could require a sync license, which the publisher is not obligated to grant.

Detecting unlicensed cover versions is a major undertaking.  First, it requires that Facebook detect cover versions of musical compositions.  Audible Magic can’t do that.  Acoustic fingerprinting technology is good at matching recordings, but it’s not designed to match cover versions of compositions.  Research has been done on this — in fact, there’s an annual contest on accuracy of cover-matching algorithms — but it’s a much harder problem, and even the best solutions are not accurate enough to be reliable in this context.

The NMPA has presumably been relying on metadata (e.g., song titles) to determine the extent of cover version videos uploaded to Facebook; meanwhile, Facebook has presumably been claiming that only the artists are responsible for getting licenses and that the DMCA 512 safe harbor shields it from liability.  Song title matching schemes are poor tools by themselves, as users can easily fool them and/or they will detect lots of false positives.  (Not to mention that song titles are far from unique; for example, the Harry Fox Agency lists 67 different songs called “You and I.”)

Then there’s the problem of checking whether the cover is licensed.  Because we’re talking about videos, in most cases this means checking that the publisher has granted a sync license for the cover.  The simplest way to do this is to require the performer to supply information about the song’s publisher — or a claim that the cover doesn’t require a sync license because the composition is in the public domain or carries a Creative Commons license, or that the recording is somehow a fair use.  Yet this is a process that is easy to fool and hard to verify.

In short, it’s unclear whether Facebook will solve the cover-video problem to the satisfaction of both its users and the NMPA in the foreseeable future.

These are the technical considerations going into the music industry’s negotiations with Facebook.  Beyond that, Facebook’s Rights Manager system — and the enhancements to it that are likely to result from those negotiations — show how the media industry’s relationship with file-sharing services has evolved over the years.

Around 2005, the music industry decided not to sue peer-to-peer (P2P) file-sharing services if they implemented fingerprinting technology to detect and block uploads of copyrighted material automatically.  The first service to come to this arrangement with the major record labels was the Israel-based iMesh.  iMesh made deals with some of the major labels in which labels could choose to substitute a paid, DRM-encrypted version of a track uploaded onto the network if it matched a fingerprint in iMesh’s database.  Audible Magic supplied the fingerprinting technology.

After the initial deal with iMesh, a few other P2P networks made similar deals.  None of them were popular with users, and by the time Google had launched Content ID, they had largely disappeared.  Content ID showed that the combination of free streaming, ad revenue sharing, ease of use, and ubiquity was better for everyone than direct revenue from users and/or blocking uploads — even if the music industry doesn’t love the payouts from YouTube.  Now, with Facebook gearing up to implement a similar system, this combination is well on its way to becoming a de facto — if not de jure — standard for online content sharing.