I Know it When I Label It: Artificial Intelligence as a Solution to Unpredictable Musical Copyright Litigation

Angelyn Gemmen & S. Sean Tu

Ed Sheeran  recently returned to court after being pulled into a third copyright lawsuit.[1] Already suffering through suits regarding his songs “Shape of You” and “Photograph,”[2] Sheeran approaches this fresh suit (comparing his hit “Thinking Out Loud” to Marvin Gaye’s “Let’s Get It On”) with exasperation.[3] The pop song genre, Sheeran claims, use only the same three or four chords arguing that he only used the musical building blocks that should be available to all artists.[4] Inevitably, there will be some overlap.

The Copyright Infringement Test

 A natural consequence of any overlap in the music industry is copyright litigation. A large problem for copyright infringement suits is the traditional test of “substantial similarity” between copyrighted and allegedly infringing work.[5] Legally defining the phrase “substantial similarity” is essential with the millions of dollars on the line[6], but it seems that the only consistent standard found in the courts is the “I know it when I see it” test.  These tests rely in large part on a judge or jury’s intuition combined with conflicting testimony of expert witnesses.[7] Human intuition, however, has only yielded unpredictable results, creating extra stress on litigants and stirring criticism of the poorly regarded test.[8]

 What’s the current answer to this predicament? Some litigants choose to settle and avoid this gamble where the odds are too difficult to calculate.[9] But that’s not the way that everybody wants to play it. Why take the risk of being accused of copying[10] before the public and your fans in the first place? Ed Sheeran’s new plan is to record every one of his songwriting sessions, so he has documentation that his work is his own before releasing it to the public.[11] But what if there was something better, a solution to the copyright test uncertainty that would work both in court and before you ever got there?

AI Could Be A Better Answer

 A better answer lies in technology: having AI act as an “expert” when it comes to musical copyright infringement.[12] An Artificial Intelligence system can be trained to reach consistent results based upon human intuition combined with expert opinion, the very backbone of the copyright test.[13] This AI, or “machine” learning, occurs in these contexts in either supervised or unsupervised form.[14] Supervised learning occurs when AI is given data with known results—in this case, giving AI a dataset of songs that human experts have pre-labelled as “similar” or “not similar.”[15] Unsupervised learning occurs when AI is given a dataset with no labels and is free to explore songs on its own.[16] Because the AI’s output follows a consistent formula, its output itself should also be consistent— or at least more consistent than the unpredictability litigants face in notoriously conflicting copyright cases.[17]

 This human intuition, unpredictable when coming from multiple quarters, would become predictable when merged across relevant stakeholders into a consistent algorithm.[18] Algorithms could be designed to account for differences in genre, style, and intended audience.[19] Once created, the algorithm could be used by the industry as an ex ante test for copyright liability.  Additionally, insurance agencies could use this tool to help ensure those musicians who want added protection, with premiums tied to the degree of similarity calculated by the AI tool.

 Sheeran himself raised the concern: “There’s only so many notes and very few chords used in [the genre of] pop music. Coincidence is bound to happen if 60,000 songs are being released every day on Spotify—that’s 22 million songs a year—and there’s only 12 notes that are available.”[20] An algorithm would be a better solution to this problem: it could be trained to ignore the intrinsic similarities within a genre to avoid false positives.

 This flagging system would also help artists, who could run their work through the algorithm before it is released to the public. If a song is flagged for copyright issues, it can be reworked until the flag disappears. Then, the artist could release their songs without fear—or, at least, with full knowledge of the extent of the risk taken.

 This proposed method is similar to the Turnitin system familiar to high school and undergraduate students in the United States. Students have the opportunity of running their essays through the same plagiarism checker their professor will use before the professor uses it. This way, the students can revise each flagged line into their unique styles before being accused of cheating when they never intended to copy at all.

Unintentional “Copying”

 Just like students occasionally stick too close to a source’s phrasing without realizing this habit may make subject to plagiarism, a piece of music can stick in a musician’s subconscious without the musician recognizing the piece as belonging to someone else.[21]  Even though the act is purely unintentional, the artist could still be liable for infringement.[22]

 The possibility of liability, not to mention the mere accusation of copying, can severely restrict an artist’s creativity. Referring to the seminal Bright Tunes case,[23] Sheeran said: "George Harrison [was at a] point where he said he's scared to touch the piano because he might be touching someone else's note. There is definitely a feeling of that in the studio. . . . You find yourself in the moment, second-guessing yourself."[24]

 Even artists whose works are alleged to have been copied can be sympathetic to alleged infringers, acknowledging that a person may not even know he has infringed.[25] To artists who know not what they do and who need reassurance, a neutral, pretrained, AI test could be the consistent, predictable answer they need to proceed creatively and with confidence.

Going Forward

 Regrettably, the Supreme Court has not weighed in on interpretations of the “substantial similarity” test,[26] despite the fact that the test appears in multiple variations across circuit courts.[27] However, this poses the perfect opening to begin using AI as an expert witness in copyright cases, demonstrating to courts the usefulness of quantifying similarity with predictability.

 Copyright infringement suits are currently on the table when someone does something as simple and predictable as paying tribute, sampling a track, or making music in a shared genre. The musicians who make artistic choices deserve predictability in copyright litigation—at least enough predictability so that they can gauge the amount of risk they’re willing to take.