The Summer of "Deep Drakes": How Generative AI Is Creating New Music and Copyright Issues
With the summer season around the corner, people are wondering which artist will have the coveted "Song of the Summer for 2023." Last year, contenders included Harry Styles' earworm "As It Was," Lizzo's anthem "About Damn Time" or even Kate Bush's 1985 hit "Running Up That Hill," which experienced a resurgence following "Stranger Things" Season 4. Could this year's summer crooner be …a computer algorithm? Or maybe not, if record labels have anything to say about it.
Pop music fans were wowed last week when "Heart on My Sleeve," a new duet purportedly featuring megastars Drake and The Weeknd, hit the Internet – except the song featured neither Drake nor the Weeknd. The song is instead the latest – and the most popular, as well as perhaps most convincing – work product of generative artificial intelligence (AI) technology. That tech was used to recreate Drake's and the Weeknd's familiar voices.
Despite its popularity (the track had hundreds of thousands of views and streams on YouTube and Spotify), "Heart on My Sleeve" was quickly scrubbed from streaming services at the request of Universal Music Group (UMG), Drake's and The Weeknd's record label, exposing a growing rift between this nascent technology and its ability to create new works on the one hand, and artists and the record labels holding the copyrights in their artists' back-catalogues on the other. This technology is already having a profound impact on how music is consumed and created and calls into question how well equipped copyright law is to handle AI's growing impact on the music industry.
What Is Generative AI?
Generative AI is a type of artificial intelligence that is capable of creating text, images, sound recordings videos and other media in response to prompts. One example is ChatGPT, which uses a model called Large Language Model (LMM) to process natural language questions or prompts from its users to produce a narrative answer. One could use ChatGPT to write a short story about virtually any topic, come up with an interesting subject line for a marketing email, create a travel itinerary for an upcoming vacation, write code or even draft a brief. Generally speaking, ChatGPT uses LLMs to "fill in the blanks" for a prompt by scouring endless amounts of data to predict responses.
Generative AI can be deployed in a number of ways to create (or help create) a new piece of music. For instance, the technology could be used to study or "scrape" millions of input music points to generate a new composition. Or, as is the case with "Heart on My Sleeve," the technology can listen to and analyze various songs and other sound recordings to create a realistic reproduction of an artist's voice, right down to his or her signature vocal tics and styles, and use that voice in a new song.
Ghostwriter977's "Heart on My Sleeve" is hardly the first song created using this technology. Wishing the Gallagher brothers could finally put aside their sibling rivalry to reform Oasis? You got it. What would it sound like if Paul McCartney sang "Imagine," rather than John Lennon? Imagine no more.
In 2015, Sony's CSL Research Lab used AI called FlowMachines to analyze a collection of Beatles songs to create a new song called "Daddy's Car." (The song – reminiscent of the Beatles' mid-'60s psychedelic/Revolver/Sgt. Pepper era – is not all AI; French composer Benoît Carré arranged the song and wrote the lyrics.) More recently, mental health organization Over the Bridge used Google's Magenta AI and a neural network to examine dozens of Nirvana songs to create a new song called "Drowned in the Sun," a new song that evokes Nirvana's loud-quiet-loud style. The vocals were sung by Eric Hogan – a member of an Atlanta-area Nirvana tribute band.
"Heart on My Sleeve" uses AI to actually mimic Drake's and the Weeknds' familiar voices and styles – and may just be the bridge too far for record labels.
Copyright Law Implications
Who owns the copyright in these AI-generated tunes? It depends. Leaving aside cases of works made for hire, copyright in a work vests initially in the author or authors of work fixed in any tangible medium of expression. But the U.S. Copyright Office has made clear that the author in cases of works created using generative AI must be a human – not a computer. See , 16190 Federal Register, Vol. 88, No. 51 (Mar. 16, 2023) ("In the Office's view, it is well-established that copyright can protect only material that is the product of human creativity. Most fundamentally, the term 'author,' which is used in both the Constitution and the Copyright Act, excludes non-humans.").
The "human author requirement" does not mean that works consisting of elements partially created by AI cannot receive copyright protection. The Copyright Office has explained that it will consider works including material generated by or with the assistance of technology for protection, asking essentially whether the work at issue is essentially the work of a human with the computer or technology "merely being an assisting instrument," or whether the traditional elements of authorship "were actually conceived and executed not by man but by a machine." See Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence, 16190 Federal Register, Vol. 88, No. 51 (Mar. 16, 2023) (quoting U.S. Copyright Office, Compendium of U.S. Copyright Office Practices sec. 313.2 (3d ed. 2021)). As it specifically pertains to AI-generated works, the Copyright Office will consider whether the AI contributions are the result of the author's own original mental conception, or are merely a mechanical reproduction of output from the computer or technology. See id. In other words, the registration analysis is made on a case-by-case basis.
For example, last year, the Copyright Office issued copyright to Kristina Kashtanova for her partially AI-generated graphic novel titled Zarya of the Dawn. Kashtanova relied on AI image generator Midjourney to create the images for the graphic novel. In February 2023, however, the Copyright Office clarified that Midjourney users are not "authors" for purposes of copyright. The Copyright Office concluded that Kashtanova's overall work, including the graphic novel's text and arrangement of images as whole, was protectable, but the individual images generated by Midjourney were not.
Whether any of the aforementioned AI-generated songs are copyrightable is similarly a case-by-case inquiry. For example, a piece of music created by a generative AI tool after an analysis of Nirvana's back-catalogue would not be copyrightable, but lyrics written and performed for the song would be copyrightable (assuming the lyrics were written by a human, of course). The problem with AI-generated songs (or really any content generated by an AI-based analysis of copyrighted material) is whether such works infringe the rights of the copyright owners used to create the work (i.e., the right to create derivative works). Proponents of AI-generated content may argue fair use – that the work was created by essentially scraping publicly available works to create something new and transformative.
The artists themselves could claim that such songs are essentially deep fakes that run afoul of their rights to publicity, while copyright owners – such as UMG – may well argue that AI-generated works are unauthorized derivative works based on copyrighted material. That's not the approach UMG took with "Heart on My Sleeve." Ghostwriter977 left a portion of a different copyrighted song in the music of "Heart on My Sleeve," rendering the track a victim of takedown notices. Yet, UMG's statement issued after the song was pulled from streaming services suggests legal battles on the horizon:
The training of generative AI using our artists' music (which represents both a breach of our agreements and a violation of copyright law) as well as the availability of infringing content created with generative AI on DSPs [digital service providers], begs the question as to which side of history all stakeholders in the music ecosystem want to be on: the side of artists, fans and human creative expression, or on the side of deep fakes, fraud and denying artists their due compensation.
Some artists are embracing this nascent technology. Independent artist Grimes has agreed to allow anyone to use her voice (or AI to recreate it) without penalty, and further agreed to split any royalties earned on a successful AI-generated song 50-50 with the song's human creator(s). While Grimes (who is not signed to a label) is willing to work with AI-wielding individuals to create new music, time will tell how other artists, rights holders (such as UMG), content hosts (such as YouTube or Spotify), the Copyright Office and the courts will handle the deluge of AI-generated works in the future, and whether such works will be copyrightable or treated as unauthorized, infringing derivative works.