A.I. Killed the Radio Star
The trees strip the yellow leaves, it is pitch black by 4pm, and the temperature has dropped enough to freeze our bones. It is late November in London - which can only mean one thing: it is Spotify Wrapped season! (sorry Apple Music users)
As I anxiously rushed down the busy London streets, I pulled out my phone from my back pocket and flicked open Spotify to be faced with the conclusion of my 2023 boiled down to music.
As always, some artists were anticipated regulars (Lana del Rey), others entirely unexpected (Abel Korzeniowski??) and, of course, the guilty pleasures (BLACKPINK).
But the biggest surprise of them all was Xavier. Who is he, I hear you ask?
Well, I’ll just let him introduce himself:
Spotify´s own AI DJ, Xavier
For those of you living under a rock, 2023 has been the year of AI.
Since it launched on the 30th of November 2022, ChatGPT has dominated debates, conversations between friends, academia, media outlets, and most tweets.
It feels like it has been around forever, so it is quite destabilizing to think ChatGPT launched exactly a year ago. Since then, AI has only grown, beginning to encompass even more fields of culture.
It seemed quite fitting, almost ironic, to see my AI-powered DJ sitting comfortably in my Spotify feed considering that it was only earlier this month that Bad Bunny went on a furious rant online about a viral TikTok song that uses AI to replicate his voice for an unlicensed song. The track, which uses fake vocals from Daddy Yankee and Justin Bieber, amassed hundreds of thousands of listens and views.
A user uploaded the track under the handle ‘flowgptmusic’ (it is still unclear whether they are linked to the AI platform FlowGPT).
In the words of the singer: “You don’t deserve to be my friends” he wrote in his native Spanish “I don’t want them on tour either” - seemingly feeling betrayed by the fact that a fake song using his vocals without consent seemed to be such a fan favourite.
Bad news, Benito - this hasn’t been the only time this year that we have seen deepfakes of artists' voices to generate songs. In April, Drake and The Weeknd had their voices cloned by a TikTok creator known as @ghostwriter to produce the viral ‘Heart On My Sleeve’.
Just like in Bad Bunny’s case, the song went viral online and later had to be removed from streaming platforms such as Spotify and Apple Music after Drake himself said this was “the final straw”, leading the artist’s label, Universal Music Group, to complain and threaten with taking legal action.
Heart on my Sleeve led to interesting developments, however. Not long after the scandal, Canadian artist Grimes took to Twitter to dish out some of her opinions on the industry of it all: stating that she is in fact excited by the prospect of collaborating with AI.
Grimes urged fans to create songs using her new website Elf.Tech, which allows users to use her AI-generated voice. The platform allows anyone to upload themselves singing and have their voice generate the singing style and technique of Grimes herself - free of charge.
The next month, LA-based Australian artist by the stage name of Kito released a track called ‘Cold Touch’ using Grimes software. She went on to call the song a masterpiece, going on to describe it as “transhumanist industrial Taylor Swift with a smattering of happy hardcore”.
Grimes is an early pioneer in this area - predicting the influence AI would have over music in 2018; yet The Canadian singer wasn’t the first to embrace AI in this way. Berlin-based US electronic musician, Holly Herndon, is an innovative figure in computer music. She made a custom AI recording system back in 2019 for her album Proto. Her most recent recording is with Holly+, her digital twin.
Holly+’s voice was cloned from recordings of Herndon singing and speaking. In an interview with the Financial Times, Holly stated that “the first time I heard my husband [artist and musician Mat Dryhurst] sing through my voice in real-time, which was always our goal, was very striking and memorable”.
Just like Grimes, her cloned voice has been made available for public use.
But not all are as embracing of new technology as Grimes and Holly Herndon are. Irish singer Hozier said he would consider striking over the threat of AI to his profession, telling BBC Newsnight that he was unsure whether tech “meets the definition of art”.
Nick Cave called a ChatGPT an exercise in “replication as travesty” and a “grotesque mockery of what it is to be human” after a chatbot produced lyrics ‘nick-cave-style’.
And even Drake’s (the everlasting victim of AI) patience was tested once again when a vocal deep fake of him covering Ice Spice’s ‘Munch’ went viral.
The same month Cold Touch was released, the Financial Times reported that thousands of tracks had been removed from Spotify after discovering that bots were being used on the platform to artificially inflate streaming figures.
This enforcement was ushered by a letter from Universal Music Group to streaming services, asking them to crack down on the use of generative AI on their platforms.
(Ironically, that same week Heart on my Sleeve went viral on TikTok and streaming platforms)
Needless to say - the music world is divided.
But the vocal deep fakes don’t stop at AI generated songs, they can also be used to have artists simply say whatever you’d like to them too. Take, for example, this audio that went viral on TikTok around the time the Ticketmaster / Eras-Tour-tickets controversy was happening:
Even Grimes herself, ever so optimistic about technology, has set rules to protect herself when it comes to people using Elf.Tech to haver her (seemingly) say inappropriate, racist, and cancel-worthy things:
This issue, of course, raises a plethora of legal questions - starting with the following: how much ownership do artists have over their own voice? And can copyright laws encompass and protect something as murky as vocal timbre?
The discourse online seems to be dominated entirely by conversations around the legal side of this technological development, but rarely do I see people questioning or debating the morality and ethics around using someone else’s voice - even if it is with the most innocent of intentions - without their consent?
What do individuals agree with when they become public figures?
“The sound of a singer's voice doesn’t have the same protection in law as the words and melodies they’re singing. Their voice might be their prize asset, but its sonic frequency isn’t theirs to copyright”
Ludovic Hunter-Tilney
The discourse online seems to be dominated entirely by conversations around the legal side of this technological development, but rarely do I see people questioning or debating the morality and ethics around using someone else’s voice - even if it is with the most innocent of intentions - without their consent?
What do individuals agree with when they become public figures?
Celebrities have their privacy violated every day and yet no one bats an eye - we tend to excuse paparazzi, deux moi blind items, and other invasions as the ‘price of fame’ BUT does the use of someone's voice without consent fall within this parameter? Can we really excuse these instances as ‘the price of fame’?
Taylor Swift saying she “doesn’t give a fuck if her tickets are over one thousand dollars” because she “doesn’t perform for poor bitches” might seem funny now, but it is dangerous and scary - and it is important to remember that this treatment is not limited to her.
We all have online personas, there are minutes of audio floating around the internet of all of our voices - most times happily uploaded by ourselves to our own platforms. What is stopping someone from using AI software to have my voice say a compromising thing? Nowadays we are all public figures by choice.
The Flipside
Computer science student at Southampton University, Aditya Bansal, set up Voicify.ai - a software that creates voices for users. Aditya noticed the cluster of AI generated covers popping up on the internet and decided to take quick action - within a week his website was up and running.
Interviewed by Hunter-Tilney of the Financial Times, Aditya recounts his need for speed to get his product out as quickly as possible. He himself has tried out the software in the style of deceased American rap artist Juice Wrld, “but my singing voice isn’t as good so I can’t reach the notes” he says.
Reading Aditya’s statement, I am reminded of Nick Cave’s words: “writing a good song is not boiled down to mimicry” - and I can’t help but think that there seems to be a lot of confusion around AI (rightly so).
I understand that Cave was coming from a place of critiquing the practice of typing a prompt such as “write me a nick cave song” into ChatGPT and having the software spit out two verses, a chorus, and a bridge.
But Aditya’s own experience suggests that a degree of musical talent is invariably needed in the world of AI-generated songcraft.
Take Cold Touch as an example - it was written, produced and put together by humans - it is just the voice that was AI generated. And even in order for that to work, you still need a real human voice as a guide so the AI can map it to sound like Grimes.
“I never really saw myself as a true vocalist. So when I learned of machine learning and the ability to try it on my own voice, that's when I could create a digital version of my voice that could kind of do all of the vocal gymnastics that my physical voice with my own training can’t do”
Holly Herndon
This means there is a certain level of talent and expertise necessary in order to create an ‘AI song’ like Cold Touch - for all of the deep fakes brought to you by AI, the missing ingredient in ALL mimicry remains good old-fashioned talent.
It makes me wonder what the difference really is between a song made with Elf.Tech and traditional singing / songwriting?
Expertise, taste, creativity is needed for both - just like not anyone can pick up a guitar and write a masterpiece, not everyone can get behind Elf.Tech and make a good song.
The process of writing this blog post has also made me consider the relationship between technology and music a lot - and they are certainly not mutually exclusive. Almost, if not every, stage of technological development in the history of recorded music has been followed by dire forecasts of doom. In the 1920s, with the rise of radio, there came a fear about the death of live music, and the innovation of drum machines in the 80’s was nervously observed and dreaded by drummers everywhere. Even Napster (precursor to streaming) had its fair share of fear mongers when it first released.
It is 2023 and live music is still a thing - and so are drummers.
Musician Holly Herndon
Florence and the Generative AI Machine:
What does it even mean to be an artist when anyone in the world can create as you?
That is one of the questions I have found myself asking while researching AI-generated vocals. When I proposed this blog post to myself, I thought I knew what my final conclusions would be - art is meant to be by the human and for the human. I was reminded of a beloved quote of mine by author James Baldwin:
“you think your pain and heartbreak are unprecedented in the history of the world, but then you read”
I suppose music is like this too. It makes us feel connected to one another. For example, when I listen to Butterfly Net by Caroline Polachek, I have this odd feeling that she wrote the song for me, that she was thinking about me when she wrote it. That makes me feel connected to her as an artist - she felt this way too, and that’s something so powerful.
And then I look online and I see all the other people that connected with her, with that song - and I feel less alone in this world. This is the inimitable magic of music.
I suppose I am somewhat of a purist when it comes to art (I wrote a blog on a similar topic, AI in visual arts), and I deeply align with artists such as Nick Cave and Hozier who are skeptical about the promise of AI; however, as I was doing my research to write this blog I found that there was another part of me that understood artists like Grimes - the optimists.
Music and technology are not mutually exclusive, and they never have been. It is impossible to deny that the reason why music can sound the way it does today is because of incredible technological advancements.
Which led me to think about samples.
We all love the rich history behind music sampling, particularly the Amen Break.
The Amen Break is possibly the most sampled drum break in the history of music - and yes, you have heard it before and might not even know it.
Notable samples include:
The frontman of the The Winstons, Richard Lewis Spencer, called ‘Amen Brother’ “a throwaway piece” - and yet not it is the sampled song in history, featured in 4,271 songs according to whosampled.com. The record went unnoticed for almost two decades until the New York hip-hop scene picked up on it in the late 1980s. Nevertheless, Spencer had no idea how his record was being used - and just how much it was being used. The Winstons had disbanded, and he was working in the Washington DC Transit system to support his return to university when he got a call from Strut Records all the way in the UK asking for a copy of the master recording (to which he owned the rights).
He never received a penny from any of those samples, and stated he “felt as if I’ve been touched somewhere that none should touch. I felt invaded upon”.
Lack of licensing structure led to many not being recognized or compensated for their work, and it also led to lawsuit after lawsuit. One of the most famous ones being when Queen and David Bowie sued Vanilla Ice over his use of the iconic bassline in Under Pressure. Sampling created new and iconic music, but it also brought a wave of legal issues. Which is why licensing allowed for a balance between the protection of artists, while encouraging innovation and creativity.
Look at us now, not only is sampling one of the most interesting and fun parts of music (I like to think of it as a trail of Will-o’-the-wisps leasing me to discovering more and more music) but it is also safe to say that sampling was a net good for the world. What’s to say that, once the legalities of it are drawn, AI won't do the same? The truth is - music and technology have always gone hand in hand. Technological advancements have always been helpful in fueling musical innovation - take for example the layered arrangements of Les Paul and Mary Ford, or the 8-track recording machine Bing Crosby was able to develop from incredibly advanced tape recording found in Nazi Germany.
As technology advances, so does music.
Holly Herndon is, once again, at the forefront of those fighting to protect artists from AI - whilst still believing the technology shows promise for musical innovation. Her and her husband, Dan, created a website called ‘Have I Been Trained?’ which lets artists search for their work in major AI systems, allowing them to either opt in or out of their work being available for usage in AI systems.
The primary function of “Have I Been Trained” is to allow individuals to check whether their art or photos have been used as part of the training data for text-to-image generative AI tools. By simply entering their name or relevant details, users can determine if their work has been employed in this manner. The website provides transparency by displaying the art and images that have been scraped or used in AI training. In addition to identification, “Have I Been Trained” offers an ‘opt-out’ tool. This tool allows individuals to request that their work be excluded from the training data of generative AI tools.
Encore
Be for or against AI, one thing is clear - it is here to stay.
Many find the use of AI in music dangerous, and a threat to artistry and intellectual property - but most people are perfectly comfortable attending ABBA Voyage, were virtual avatars depicting the members of the Swedish band (as their younger selves - which is another interesting aspect that we do not have time to get into right now) are powered by artificial intelligence in order to interact with the crowd in real time?
So what is acceptable when it comes to AI and music? Where do artists, labels, producers feel threatened and when do they feel like their pockets are getting bigger?
Where do we draw the line?
As I said before, the musical world is undoubtedly divided about this.
Many, like Grimes, see AI as a helpful and exciting tool that will allow creatives to expand their intentions, visions and processes; and also being able to democratize the music industry - being able to jump over the heads of music labels and executives, a way revolutions not only how we make music, but how we share it with others
Yet, there’s always Side B to every record -
Instances such as Bad Bunny’s, Drake’s, The Weeknd’s and (frankly) many more to come display just how AI can also be used to simply ride on someone else's voice, talent, career and success in order to rack up a couple of streams.
Examples like Cold Touch by Kito give credit to Grimes' vision, leading to the production of creative and original music. But then Heart on my Sleeve by ‘Not Drake’ is a testament to the concerns of many artists.
My final question is this:
Do we think artificial intelligence will be a tool for creativity that leads to new and interesting sounds? Or will it be just another disappointing tool used by the talentless to feed off from all other popular music trends out there, having the machine spit out the same old song with the same old BPM that will lead to the most streams?
At the end of the day, like with most technological advancements and most of history, the outcome of this lies in how we decide to use it.
So basically, time will tell...
...in the meantime, I will leave you with this:
