duce multiple acoustic fingerprints for every track in their catalogs to populate the databases of every competing vendor.

Media companies acknowledge that the system is imperfect but, they say, it is better than nothing. But how much better than nothing is it? In 2006, MySpace began using acoustic fingerprinting in an effort to filter copyright infringing songs from being posted to users’ profile pages. Yet even today, it takes only a few clicks to find profile pages pumping out unlicensed commercial music. Music by artists who, one presumes, that user likes and is promoting for free.

As popular as online music is, though, the future of Internet media is all about video. As the $1 billion lawsuit facing You Tube from Viacom suggests, filtering video is no easy feat.

Vendors like Audible Magic and Vobile have invested millions in an effort to reli-ably identify copyrighted video despite the possibility for wide perceptual variations in identical content. Differences in color tones, compression artifacts, frame distortion, skew, and aspect ratio can all be easily overlooked by human eyes when comparing content.

Since its purchase by Google, You Tube—whose popularity stemmed in large part by being a repository of copyright infringing content uploaded by users—has long promised to implement filters. After a series of sliding rollout dates, You Tube released their Content ID tool in October 2007. Reports vary as to how well the technology identifies content, but the technical question is almost beside the point.

For a content owner to benefit from You Tube’s filter, they need to upload their video clip for analysis. As with acoustic

fingerprinting, the filter needs a database to compare against. Multiply this effort by millions of copyright protected clips. Plus, there are dozens of alternative video sharing sites where users can upload copyright infringing content without filters. And if these sites did implement filters of their own, it would add that much more workload to submitting clips for protection.

Copyright Conundrums

Despite the seeming morass of technical challenges to filtering vast quantities of content, technological limitations might be the least of the problem. Critics point out that while pro-filter companies like movie studios and AT&T frequently invoke the phrase “copyrighted content,” copyright itself is not nearly so black and white.

U.S. copyright law provides protections to content owners, of course, but it also reserves rights for the public, too. For example, the famous “fair use” doctrine allows portions of copyright-protected content to be used without permission in a variety of ways, including comment, criticism, education, and parody. While authorization to use copyright-protected content can come from its owner, it can also come from its context, as defined by the law. How will content filters interpret context?

You Tube has tried to address the issue by including thresholds in their Content ID formula—two seconds of a clip, for example, may not be long enough to trigger the filter. The exact length of time, and whether time is the only factor, has not been divulged. At best, it would seem that any computational approach to judging copyright context will be coarse at best.

It is also important to remember that copyright protections extend to all original work created by anyone, not just movie studios and record labels. This includes Web pages, blog postings, and photographs, unless its creators choose to license their