The dancing baby puts automated takedowns back in the box?

The big news of the day in copyright land is that the Ninth Circuit decided a meaningful appeal in the “dancing baby” case, Lenz v. UMG. You probably know the one, but if not, the basic background is that Universal Music Group had this adorable dancing baby video taken down because of the badly distorted Prince song playing in the background. The baby’s mother, Stephanie Lenz, lawyered up and tried to take advantage of the little-used and presumably-toothless provisions of § 512(f) to make UMG pay up for abusing the notice and takedown process.

The big takeaways that have many in fair use’s posse clapping are that (1) the opinion resoundingly rejects arguments that would reject fair use as a mere “affirmative defense” rather than an affirmative right, and (2) the opinion injects a little bit more of § 107 (fair use) into § 512 (notice and takedown) by requiring would-be takedown notice senders to consider fair use or risk some (probably minimal) level of liability.

With all legal opinions the devil is in the details, but—in initial my read—this is one to celebrate. There’s lots going on, but the court is trying to strike a delicate balance. Is there a workable way for the nuance of the fair use analysis to mesh with the essentially automatic (if not automated) notice and takedown process? It would be easy to decide this case in a way that favors one or the other without taking up the challenge to make them compatible. My initial impression is that the Lenz opinion does as much as can reasonably be expected to make it all work. Don’t get me wrong: there’s much to recommend the dissent’s points, but my general feeling is that it would be hard to do better.

So let’s get to what I want to talk about: Mike Masnick’s concerns about the opinion’s implications for automated notice and takedown.

In many ways, automated notice and takedown is a scourge. This is the process whereby takedown notices are robosigned to eliminate anything with even the faintest whiff of copyright infringement. It’s how rights holders manage to take down their own postings or legal content that happens to share a name with a major motion picture, to say nothing of countless fair uses. A powerful legal cudgel operated by a mindless robot: what could go wrong?

But, in an important sense, these measures are understandable. From a rights holder perspective, the DMCA is useless if it can’t scale to internet-sized proportions, and it can’t do that without some measure of automation.

So what’s the answer? Well, it probably looks something like what the Court (tentatively and nonbindingly) proposes. Here’s the language (with apparent assistance from the brief from the Organization for Transformative Works and the International Documentary Association):

We note, without passing judgment, that the implementation of computer algorithms appears to be a valid and good faith middle ground for processing a plethora of content while still meeting the DMCA’s requirements to somehow consider fair use. For example, consideration of fair use may be sufficient if copyright holders utilize computer programs that automatically identify for takedown notifications content where: “(1) the video track matches the video track of a copyrighted work submitted by a content owner; (2) the audio track matches the audio track of that same copyrighted work; and (3) nearly the entirety . . . is comprised of a single copyrighted work.”

Copyright holders could then employ individuals . . . to review the minimal remaining content a computer program does not cull.

(citations omitted). Essentially, the idea is to automate that which can reasonably be automated. Literal reproductions of entire works without additional user-provided context have the largest burden to climb to in the fair use analysis. And that part of the analysis could, conceivably, be more readily outsourced to machines.

Yes, it’s doubtful machines are in a position to handle the more difficult prongs of the fair use test. And, without that context, it’s a certainty that machines will get things wrong. But I’d rather live in the world where automated notice and takedown were restricted only to those instances where it might reasonably be capable of achieving the right result occasionally, rather than what we seem to have got now, where it’s more or less free to run amok. It might not satisfy the TechDirt set, but that interpretation of today’s decision would serve to put automated notice and takedown at least partially back in the box.