Fair Use vs. Algorithms: What the Dancing Baby Did to Copyright
Contrary to popular belief, a landmark case is unlikely to slow down the everyday flurry of DMCA takedowns.
The Digital Millennium Copyright Act (DMCA) is the easiest way to get something taken off the internet. Small wonder, then, that it's so frequently actively abused. DMCA notices are often sent over instances that are obviously fair use—in one notorious case, Lionsgate took down a remix video that had been cited by the Copyright Office itself as an example of transformative fair use.
It happens all the time. Exact numbers are hard to come by, but Twitter and Wordpress report that they decline to enforce about a third of all DMCA notices that come their way. It's not as though Congress didn't anticipate DMCA abuse. Section 512(f) of the DMCA allows targets of wrongful DMCA notices to sue the sender if there's been a "knowing misrepresentation." But over the years, 512(f) has dwindled down into nothingness as various court decisions made it harder and harder for ordinary internet users to bring lawsuits against corporate rights-holders. It's prompted some notable commentators to call it "dead."
On Monday, the Ninth Circuit Court of Appeals handed down a decision in Lenz v. Universal, also known as the "dancing baby" copyright case, declaring that rights-holders must consider fair use before sending off a DMCA notice. It's a nice message, but one that's unlikely to slow down the everyday flurry of DMCA takedowns.
In February 2007, Stephanie Lenz uploaded "Let's Go Crazy #1" onto YouTube. It's a 29-second clip of a toddler pushing a toy stroller and bouncing happily as the radio plays in the background. The music is barely recognizable—but as the video's title suggests, it's Prince's "Let's Go Crazy." The video was taken down in June the same year in response to a request from Universal, and restored in July after attorneys helped Lenz file a counter-notice. A lawsuit against Universal was filed shortly after.
It has been eight years since Stephanie Lenz's video was taken down—eight years of paperwork, of hearings, of oral arguments, of lawyers shuffling back and forth between courthouses. No ordinary person in their right mind would pursue this course of action: a YouTube video being taken down for six weeks is not worth tens of thousands of dollars in legal fees.
Of course, Lenz is no ordinary case. It's being litigated pro bono by the Electronic Frontier Foundation. Among the countless DMCA takedowns that happen every day, the EFF picked this one to fight, probably because the facts make Universal look very, very ridiculous. But even this hand-picked test case can't win.
In the ruling handed down today, the Ninth Circuit is saying, "You have to look at fair use before sending a DMCA notice," but it's also saying, "Well, you don't have to look that hard." The court all but declared that its hands were tied by a previous decision, and because of that, Lenz would have to show "subjective bad faith" on the part of Universal in order to successfully sue the record company for DMCA abuse. It's a troubling move in a time when subjective intent has less and less to do with the DMCA altogether. We live in an age of copyright enforcement by robots. By giving its blessing to both automated takedowns and an incredibly low level of fair use consideration, Lenz retrenches the current status quo and does little to dissuade DMCA abuse.
The Lenz case is based on section 512(f) of the DMCA, which is the only offensive counter-balance to DMCA notices. Under 512(f), if a rights-holder makes a knowing misrepresentation in a DMCA notice, their target can sue them for "any damages, including costs and attorneys' fees."
But what counts as a knowing misrepresentation?
When it comes down to it, what Lenz says is that a rights-holder has to reach a mustache-twirling level of intentional villainy before the target gets to hold them accountable. Even a fuzzy 29-second clip of a baby bouncing up and down to the tinny, distorted notes of "Let's Go Crazy" isn't enough for a summary judgment on 512(f). This case has to go to trial, instead.
"I struggle with, how anyone looking at this… could view it as anything other than fair use," Judge Milan Smith said of the video during oral arguments on July 7th. He dissented from the majority opinion, concluding that Universal had knowingly misrepresented that the video was infringing.
But the majority opinion refused to make up its mind on that count. Universal claimed its system for issuing takedowns did take fair use into account. Lenz claimed Universal hadn't formed any subjective belief about fair use whatsoever. To settle the score, a jury trial would have to be held.
Digital copyright enforcement was once bureaucracies of people talking to other bureaucracies of people. Today it's robots talking to other robots.
This is where the majority opinion and Judge Smith's partial dissent diverge. Universal's policy was to issue a takedown if the copyrighted work "was the focus of the video." The phrase "fair use" never came into play, but that, Universal contends, doesn't matter. This description of Universal's procedure was adequate enough that the majority was willing to pass the buck onto a jury, but to Smith, the record shows that Universal never considered fair use at all.
This difference is huge. By leaning this way, the majority gives its blessing to copyright enforcement in the age of robots.
"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," says the court. The opinion even goes as far as to suggest criteria for an acceptable algorithm—for example, if "nearly the entirety" of a copyrighted work is detected.
But just because nearly the entirety of a song or a video is copied doesn't mean it's not fair use. Fair use is a four-factor test, and all factors have to be considered. "[T]he record does not disclose whether these programs are currently capable of analyzing fair use," Judge Smith wrote in his dissent. "For a copyright holder to rely solely on a computer algorithm to form a good faith belief that a work is infringing, that algorithm must be capable of applying the [fair use] factors enumerated in § 107."
In both the oral arguments and in the written opinions, the judges showed a keen awareness that times have changed since 2007. Today, rights-holders are increasingly dependent on algorithms to detect infringement, and platforms like YouTube use Content ID to proactively police infringement on their own ends. Digital copyright enforcement was once bureaucracies of people talking to other bureaucracies of people. Today it's robots talking to other robots. What Lenz does is that it holds the robots to a pretty low standard.
This isn't to say the decision is all bad for proponents of fair use. For the first time, a court is saying that fair use is not an affirmative defense to infringement, but rather, that it's not an infringement at all. It seems like a technicality, and maybe in practice, it doesn't make a huge difference. But this takes fair use from being an affirmative defense—like self-defense in the case of homicide—and closer to a defense like, "Actually, the person I killed isn't dead at all." Doctrinally speaking, that's huge.
The other thing is that even though 512(f) seems to now only apply to cases of mustache-twirling villainy, there are actual cases bordering on mustache-twirling villainy out there.
In Raven v. Molyneux, the men's rights activist Stefan Molyneux sent DMCA takedowns against a YouTube channel dedicated to criticizing the men's rights movement for misogyny, and then went on a podcast and admitted his actions had "nothing to do with copyright or anything like that," but rather he had issued the takedowns because he "felt listeners were being acted against in a negative way."
In Automattic v. Steiner, blogging platform Wordpress was hit with a DMCA notice by homophobic organization "Straight Pride UK," after a blog published an unflattering press release that Straight Pride UK had, itself, put out. (Wordpress won a 512(f) action against Straight Pride UK, but this may have been in no small part due to the fact that Straight Pride UK was nowhere to be found once the lawsuit was filed).
Extreme cases do exist, and because the new Lenz decision allows for the possibility of collecting attorney's fees and other kinds of damages, the adjusted financial incentives might encourage people to go after bad actors. But these considerations are just some very pretty flowers atop the grave of section 512(f). Sending a pile of poorly-vetted DMCA notices is just business as usual, whereas fighting those notices in court can only be an act of principle.