Contentious AI Composes New Songs Based on Dead Music Legends

AI composes songs to raise mental health awareness but leaves behind existential questions.

AI can now use machine learning to produce music that is virtually indistinguishable from that of famous artists. (Image courtesy of Stock.)

AI can now use machine learning to produce music that is virtually indistinguishable from that of famous artists. (Image courtesy of Big Stock.)

The enduring power of music is akin to immortality. Many of the songs played across the world today are from artists that died years, even centuries, ago. However, with this legacy comes the duality of the what-ifs. What if Beethoven lived longer and created just one more symphony? What if Elvis Presley lived long enough for another album? What if the Beatles never broke up?

Too often, these questions take a somber turn when we’re forced to wonder not just what kind of music these artists would have produced, but also what their lives would’ve been like had they not died so tragically young. What kind of song would Janis Joplin have written had she not overdosed? What kind of album would Nirvana have released had Cobain not shot himself?

These are exactly the kind of questions that an organization called Over the Bridge is trying to answer. By using artificial intelligence (AI) and machine learning, the Toronto-based group has created “new” songs for the most famous members of the 27 Club: Jimi Hendrix, Kurt Cobain, Jim Morrison and Amy Winehouse.

Three members of the 27 Club, a group of artists—including famous musicians—who passed away at just 27 years of age due to substance abuse or suicide. From left to right: Jim Morrison, Amy Winehouse and Kurt Cobain (Image courtesy of Rolling Stone.)

Three members of the 27 Club, a group of artists—including famous musicians—who passed away at just 27 years of age due to substance abuse or suicide. From left to right: Jim Morrison, Amy Winehouse and Kurt Cobain (Image courtesy of Rolling Stone.)

Creating the Music

Over the Bridge is a non-profit organization dedicated to helping people in the music industry with mental health issues. Since its inception in August 2017, it has helped more than 2,000 people in the industry receive training, education and emotional support for their struggles.

Outside of their legendary status in the world of music, Over the Bridge chose Hendrix, Cobain, Morrison and Winehouse due to their large and eclectic catalog. Up to 30 songs from each artist were analyzed by the Google AI program, Magenta. Using machine learning, Magenta’s long short-term memory (LSTM) isolated all the distinct note choices, melodies, rhythms, passages and riffs that gave these artists their signature musical style.

Once isolated, Magenta was able to predict the composition style of a potential new song that each artist might have written had he or she lived. Once translated through a synthesizer, musical elements were generated that an audio engineer composed into a coherent piece.

A similar process was used for the lyrics. As many as 30 songs were entered into a generic artificial neural network (ANN), which isolated passages and phrases that were used to predict the style, tone and substance of new lyrics that the musicians might have written.

Both the music and lyrics were then reviewed by the creative team at Over the Bridge for coherence and recognizability. Finally, singers from famous tribute bands were recruited to create vocals with the perfect blend of inflection and cadence to make the songs as indistinguishable from the real thing as possible.

A History of Blues

Aside from their talent and vision as musicians, what links all members of the 27 Club is a history of mental health issues. It is the gutting yet romanticized aspect of their legacies. Drug addiction, alcohol and substance abuse, severe bouts of depression, and anxiety are par for the course for the 27 Club and many musicians in general. A 2016 study revealed that 71 percent of musicians report panic attacks and elevated anxiety levels, and 68.5 percent experience depression—three times more than what the general public reported. These findings are even more galling when adjusted for age since musicians between the age of 18 and 25 are 80 percent more likely to experience depression and anxiety. Suicide attempt rates for musicians are believed to be twice that of the general public. Similarly, self-harm and eating disorders are amongst other issues that adversely affect musicians, specifically younger musicians.

Despite all this tragedy, the 27 Club still endures as a symbolic Hall of Fame—a cult of personality for the once-in-a-generation musicians who shone too brightly and paid the ultimate price. The mystique attached to dying young is perhaps symptomatic of how normalized—even glamorized—mental health struggles are in the music industry. Nothing is more telling of this mindset than Cobain quoting Neil Young’s lyrics in his suicide note: “Better to burn out than fade away.”

You Know Your Copyright

Over the Bridge makes it clear on the Lost Tapes of the 27 Club website that it is neither affiliated with nor endorsed by the Cobain, Morrison, Hendrix and Winehouse estates. They also make it clear that the music produced as part of the project should not be considered new songs by the artists.

That said, the possibility of using AI to produce such music may become prevalent in the near future cannot be ignored. If this were to happen, present-day copyright laws may prove insufficient to address the legal ramifications of using AI in this way.

Currently, there are two copyrights for a song. The first copyright is for the lyrics and musical arrangement, which is owned by the artist who originally recorded the song or their publishing company. However, any other artist or band who performs a cover of this song will also hold the copyright to their particular sound recording. For example, the song “Man Who Sold the World” was originally written and performed by David Bowie, but its cover by Nirvana is arguably more famous. In this case, David Bowie owns the copyright for the lyrics and musical arrangement while Nirvana, or its record label, owns the copyright for their rendition of the song.

With AI, however, the issue is complicated because it is not a homage or tribute as in a traditional cover song. The AI software is using pattern recognition to identify a musician’s proclivities, idiosyncrasies and style, and then predicting how they would write a new song from scratch. As such, the line where these songs can be considered new songs by the same artist becomes blurry. It begs the question: Should AI songs like the one created by Over the Bridge be considered legal at all?

Philosophical Implications

AI creating music also forces us to address a larger philosophical quandary. For eons, creative fields, such as art and music, have been considered an extension of our humanity—an outward expression of the human capacity to feel and have complex emotions.

Yet, in 2018, an AI-created painting sold for $432,000. In the same year, a poetry-composing AI program named Deep-speare composed sonnets that were virtually indistinguishable from those written by Shakespeare—the man who represents the pinnacle of human achievement in English literature. Where does this leave human creativity if AI can produce works rivaling the best writers, artists and musicians in considerably less time and with minimal effort?

AI created paintings that are as visually striking as they are unsettling. (Image courtesy of The Atlantic.)

AI created paintings that are as visually striking as they are unsettling. (Image courtesy of The Atlantic.)

Ethical Implications

Growing up in the 90s, I can attest to the fact that Nirvana was as ubiquitous as Discmans and Nintendo. As much a fan of the band then as I am now, the news of Cobain’s death is still a watershed moment in my life. I was eleven, and it was the first time I’d encountered words like suicide, depression and addiction. It also forced me to contemplate ideas like death—the what-ifs, the lack of continuity—in a much more defined way than before. After all, Cobain was gone, and with him went all the music that he was ever going to compose.

As time went by, I began to hate the fact that I’d never hear a new Nirvana song—that I’d never listen to Cobain’s absurdly evocative lyrics. I felt robbed of the opportunity of ever seeing the band perform live or one day even meeting them in person. It was a selfish and irrational feeling, that I ultimately resolved by accepting that—whether justified or not—Cobain had made his decision. He had chosen to end his life. In doing so, he chose to put an end to his creative life force.

One might argue that using AI to create new songs for Nirvana is tapping into the same kind of selfishness I felt as a child. After all, what gives anyone the right to write new songs in Cobain’s style and musical identity? What gives them the right to reanimate his creative life when he decided to end it?

While Over the Bridge is using his music to raise awareness for exactly the kind of issues that led to Cobain’s demise, we also need to raise the all-important issue of consent. After all, regardless of the organization’s intentions, there is no way for Cobain and others to consent to their musical style being exploited this way. Is it then even ethical to use AI to write new songs for them?

Too Much, Too Soon

Today, deep-fake services, such as Deep Nostalgia, can animate photographs of dead relatives to make them move and smile as if they were alive. Chatbots, with which one would be able to speak with dead relatives, are not just a possibility. Microsoft got as far as winning a patent for it, though no further plans were made once it became public knowledge. While such services can be helpful in overcoming the loss of loved ones, they are too powerful a tool to wield in a cavalier fashion. In my opinion, that same principle should apply to AI-generated music for artists who are no longer with us. While the legal aspect of this can be resolved by updating copyright laws, the philosophical and ethical conundrums are still a long way from being addressed satisfactorily. As long as this remains true, I believe our ventures into this field might not just be premature but also morally dubious.

To hear the songs generated by AI, click here.