Last April, pop star Grimes made headlines when she announced on Twitter, now X (the failing platform owned by her baby daddy) that she would split royalties 50–50 with anyone who used her voice for an AI-generated song. This whipped the music industry into a frantic tizzy until people calmed down and started thinking in a sensible manner. Grimes has always been a proponent of futuristic technology, but that doesn’t mean it’s here yet. She hasn’t released a new album in years. And while she can carry a tune, she’s not exactly Whitney Houston- her talents lie more as a songwriter and producer.
Artificial Intelligence can do a lot of things and it’s evolving to add even more capabilities. Fraud detection, astronomy research, chatbots, and Siri are a few of its uses. But we haven’t really ironed out a lot of the kinks. Recently a self-driving car plowed through a crime scene in California cordoned off with yellow police tape. Teslas have a glitch where they won’t move if you place a traffic cone on top of the hood. A Jimmy Kimmel segment featured him instructing his Echo to order pancake mix and viewers woke up the next morning to discover two boxes of Bisquick on their Amazon wishlists. Nor does AI necessarily encourage good habits. My old neighbor Evan was literally dumber than a bag of hammers but had the wisdom to take the Alexa away from his three-year-old daughter because “I don’t want to teach her to yell at technology.” (This is a person who once tried to start a winter backyard s’mores cookout by igniting piles of frozen mud with a lighter. Not a high standard, folks.)
The problem with using AI for any type of creative artistry is that these systems are trained on the output, not the process. According to AI music generator Soundful, “Deep learning involves training an artificial music generator on a large dataset of existing music. . . Neural networks mimic how our brain works when creating music.”
Except those are two vastly different things. Large Language Models (LLM) rely on the input of existing (most often, copyrighted) songs, boiled down into trillions of little 1’s and 0’s, where they can then analyze patterns and randomly generate rhythms and lyrics based on the data they’ve been fed by engineers. Is this how the human brain functions when composing? Probably not, but frankly we don’t know- neuroscientists haven’t quite figured it out yet. With very advanced medical imaging technology like MRI, CAT, and ligand-based PET, researchers can tell which parts of the brain light up and which chemicals are released, but the actual step-by-step procedure at the cellular level is anyone’s guess. Complicating matters is the fact that music doesn’t just impact one area- it’s processed by the limbic, neuroendocrine, and autonomic nervous systems, along with some cerebral cortex involvement.)
Which is not to write off the use of AI in music entirely. A 2018 paper published by Berklee College of Music and Health neurologists Dr. Samata Sharma and Dr. David Silbersweig suggested that due to music’s therapeutic effect on the brain, machine learning could one day be customized to treat functional disorders such as chronic pain, depression, and Parkinson’s.
But is it art?
Not according to the Oxford English Dictionary, which defines “art” as “the expression or application of human creative skill and imagination.” If we don’t really understand the brain’s creative procedure in the first place, how can we attempt to replicate it with technology? Sure, you can prompt AI to generate an “image” or “content,” but it’ll base the result on the data it’s been fed rather than via the cognitive process of imaginative thought. We don’t know how that last one works yet.
There’s also that uncomfortable little fact that money-hungry MBA techbros don’t like to acknowledge: great art breaks rules. You have to know the rules in order to break them, and you need artistic sensibility to know when and where to break those rules. The Wu-Tang and Avicii songs they’re blasting in their $250 AirPods at the gym? Those guys broke rules. RZA’s production techniques influenced rap for the following decade and Avicii pioneered the blend of electronica with melodic soul, funk, blues, and country structured in a traditional song format. Even Grimes- while she’s always used software to make music- has notoriously strayed from the standard electro-pop route; in 2019’s “Delete Forever,” she layers spare acoustic guitar and synths over a hip-hop beat that fades in a mournful swirl of banjo and fiddle. Could AI do this right now? Not even close. Could AI do it eventually? I doubt that, too.
I recently got into a discussion on LinkedIn with an AI enthusiast who pointed out the positives of music generators like Loudly and Soundful, and I have to admit he’s made a good case. They democratize the creative process so that you don’t need a studio, or instruments, or even a musical background- just an Internet connection. They can help overcome writer’s block or provide a good starting point. They can help students learn to compose different parts of a song: rhythm, melody, harmony, lyrics. In short, they’re perfect for hobbyists, amateurs, or beginners.
The problem comes when CEOs decide that AI can, and should, replace actual professional musicians. I understand that artists can be a pain in the ass sometimes. They’re expensive, weird, and smelly; they get drunk onstage and can’t perform; they say outrageous things in the press and cause a nightmare for the label’s PR team. But they also have the capability to create a much better product than AI systems. Most importantly, they can connect with an audience, something a bot can’t. I recently watched a YouTube clip from a concert by Norwegian indie-pop singer-songwriter/queer icon Girl in Red where several lesbian couples in the audience either got engaged during the show or attended it together to celebrate their upcoming nuptials. Which is an amazing testament to music’s ability to create a vibrant, welcoming community with safe spaces, and also demonstrates the importance of a human component within art.
But if you want to take Grimes up on her offer, here’s what you can do with AI. Prompt a generator by selecting genre, tempo, duration, key, and instruments. Ask ChatGPT for a few rhyming lyrics about Elon Musk or technofascism and run those through software that can replicate Grimes’ high, thin voice. I’ve tested a handful of products and that seems to be where their capacity ends (although some are better than others; Loudly produced a 30-second cacophony that I can only describe as “worn-out brake pads on a ’92 Ford Crown Victoria.”) You could then write a code that pairs a melody with a few simple chords or structures your composition in a traditional verse-bridge-chorus format. You can do this a thousand times if you want and produce a thousand different, but equally monotonous, four-chord pop songs clocking in at 3:35. And it’ll sound a lot like Grimes’ recent disastrous set at Coachella, but not like the music that got her famous in the first place.
A very smart friend of mine, who has worked at a major Fortune 500 tech company for almost ten years, sagely advised the other day that “AI needs to stay in its lane.” I don’t doubt that it holds great promise for areas requiring intake, pattern recognition, and prediction of large data sets: medicine, transportation, weather, etc. Unfortunately we’ve mostly used it so far to steal from copyrighted works without authors’ consent, generate fake nudes of Jennifer Lawrence, and produce glitchy website sales chatbots- while worsening our climate crisis with its rampant, gluttonous electricity requirements. The last place we need AI is on the airwaves. Or in art. Or in literature. Much like the proverbial grape, “keep it in the dark until it matures.”
