The Best AI Music in the World, Vol. 4
This Retro series catalogues the AI-powered production of “new” music “sung” by great artists we can never again hear from—as well as experiments in sending still-living artists “out-of-genre.”
Note: The essay section of this long Retro report is provided for free, along with one AI track. To access the full article, click the button below to sign up for a free seven-day trial of Retro.
Introduction
This professionally curated Retro series features five AI-created songs in each volume, offered in no particular order. The titles appearing above each track were supplied by the artists who used AI to produce these songs. None of these songs are endorsed by their listed performers. They are intended solely for education, as they illuminate important uses of a new technology.
In the first edition of this series, I offered some preliminary thoughts about AI music. If you’re interested in following along as the “Best AI Music in the World” series expands and evolves within the Music section of Retro, I do hope you will read that—admittedly lengthy, far-ranging, and alternately personal and theoretical—essay.
One of the points made by the essay is just how rapidly AI music is developing. And nothing could be a better indicator of this than just looking at the last year of major-media news. It seems that everyone is now talking and writing about AI music and doing so with enormous urgency. On April 11, 2023, The Financial Times reported that
Universal Music Group (UMG) has told streaming platforms, including Spotify and Apple, to block artificial intelligence services from scraping melodies and lyrics from their copyrighted songs, according to emails viewed by the Financial Times.
UMG—which controls about a third of the global music market—has become increasingly concerned about AI bots using their songs to train themselves to churn out music that sounds like popular artists. AI-generated songs have been popping up on streaming services and UMG has been sending takedown requests “left and right”, said a person familiar with the matter.
The company is asking streaming companies to cut off access to their music catalogue for developers using it to train AI technology.
More recently, the New York Times reported that “[state and federal] lawmakers are beginning to contemplate questions about authorship and ownership around creative machines [because] the stakes for creative businesses are high.”
According to the Times, the three key questions being asked right now are these:
“What is owed to the creators of the original material?”
“Does ‘[the] fair use [doctrine]’ apply?”
“Who owns the output of generative A.I.?”
To Retro, the answers to these questions seem straight-forward—though it will no doubt take a good deal of time for state and federal courts to work through them.
The Future of AI Music
Ultimately, AI cover versions of non-public-domain songs will have to give royalties to the original artists, just as cover versions and their artists have done for decades now.
The fair-use doctrine will be applied or not based on the usual considerations for its application, including whether an existing song has been sampled but materially altered and whether there’s a parodic, critical, and/or educational purpose in play. And while courts currently don’t recognize “algo jockeys”—the digital artists who, like early graphic designers, use text and their technological know-how to make artifacts they couldn’t create using their own motor skills—as copyright holders, it’s likely AI music will at some point in the future become, at least in some fashion, copyrightable.
In the meantime, the three questions asked above will have the short-term effect of making it that much harder to locate AI music online—unless you habitually go deep into the bowels of YouTube and TikTok or locate obscure user-made Spotify playlists.
Indeed, after the scores of “takedowns” relating to the Drake & The Weeknd AI song “Heart on My Sleeve”, it has already become notably harder to find full AI-generated tracks online (particularly those that are produced in the style of a famous artist, and even more so if that artist is still alive). Far more common, now—a phenomenon we will see in this particular edition of the “Best AI Music in the World”—are AI covers of existing tracks by popular living artists, usually but not always by artists who have passed away or are, for various reasons, particularly unlikely to issue takedown notices on YouTube.
The chief remaining question for AI music, however, is whether and when it will be embraced by working artists on the grounds that it could be tremendously lucrative for them. One can imagine a near-term future in which—in the same way artists now release singles or EPs featuring a single song alongside some remixes of it sanctioned by the artist—new singles are accompanied instead by AI tracks of other artists singing the same song. These professionally produced AI tracks would of course be made with the permission of these second (and third and fourth, as the case may be) artists, with these other artists expecting royalty payments even though they’ve not done any work themselves. Indeed, while of course we’ll continue to see collaborations between big artists in perpetuity, AI could transcend the limits of time, space, money, scheduling, and other logistics to permit virtual collaborations that couldn’t be orchestrated otherwise. (Imagine, for instance, foreign releases of U.S.-published music that allow local listeners to hear how the songs might be sung by local artists they already know.)
The Upshot
In short, in business—and entertainment is a business—profit always finds a way.
The anger that accompanied the emergence of “covers” in the 1960s faded away once America’s legal system developed a way for artists who had had their songs covered to benefit from it. Suddenly, artists were not only okay with cover versions of their songs but starting writing new songs, more than they ever had before, that they from the jump intended to see hit big with other artists. They might record their own version for posterity and for their own self-satisfaction, but they knew that the real money lay in having someone else with either a different voice or a different fan base or a different appeal or a unique connection to the lyrics of the song perform it for a mass audience.
Is AI much different? So long as a track is performed with the permission of the artist whose voice is used (who presumably doesn’t have the time or inclination—for either reputation or timing reasons—to do a live cover version)? Many artists would relish the idea of passive income of this sort not just to get rich but to help fund their own future projects. And because AI tracks can help songs and singers jump genres, they create an opportunity for, say, a country singer to reach a hip-hop audience or vice versa, something that ordinarily would only be possible with an enormous investment of time and energy by the artist(s) involved in such attempts and/or collaborations.
Just so, the estates of deceased singers must know that when a singer who has passed on goes viral for having their “voice” used in an AI track that millions hear, that (a) keeps the artist’s name and story and reputation alive and relevant; (b) could lead to new fans for (and purchases of) the original music by the said artist (which will almost always sound better, and more suited to the artist, than the AI track that people have already come to love, meaning a better chance of new-fan adoption); and (c) expands the cultural impact and artistic oeuvre of a deceased artist in ways that are exciting and showcase the versatility of that artist’s native talents. And as long as every AI track is transparently labeled as such—and even in these early days of AI, almost all of them are, as creators who fail to do so know they will be called out and chastised by the public and then punished by social media platforms like YouTube—no music fans are being misled into thinking a late singer or band performed work they really didn’t.
The history of technological innovation—and indeed artistic innovation as well—is that major innovations are first met with anger, then rejection, then curiosity, then bemusement, then acceptance, and finally crystallize within a culture as though they’d always been there. There’s candidly no reason to believe AI music will be any different.
{Note: On rare occasion, songs that appear in this series may be unsettling to certain viewers due to their ability to mimic human singers both living and deceased. Please know that any such upset is unintentional. These artworks are intended to surprise, delight, illuminate and amaze, not provoke negative emotions or physical responses. However, because AI-generated work product is definitionally already lodged deep in the “uncanny valley”, occasional strong reactions to the songs in this series from some listeners are to be expected. Certain tracks are prefaced by a “curator’s note” that briefly explains why I find the work to be of artistic merit.}