Generative AI music is amazingly lifelike, yet somehow still uncanny.  But, considering trajectory of its meteoric improvements over the past decade, it won't be long before we can't hear the difference between not-Bender here, and the real thing.

Believe it or not, generative music AI dates back to the 1950's, with academic exeperiment Illiac Suite from the University of Illinois, pioneering the craft.  An Israeli team's Blue Jeans and Bloody Tears blew the public's mind as a more recent early AI work: "There's no life without your life in misery, blue jeans and bloody tears..."  It was featured on but contrary to popular myth wasn't entered into and didn't win the Eurovision Song Contest in 2019.  What's emerged in the space of six years since then is unrecognizable.

Suno and Udio are the two major players in today's generative music AI.  I love to play with them.  I started creating tracks like Pressure's Building in a platform using Suno:

What I do to create a piece is input my lyrics, provide a user prompt to instruct the model on genre, style, instruments, and so forth.  While I write lyrics, you could get the AI to generate those, too.  Sometimes it's better at following my instructions than others.  It provides two samples, from which to choose; otherwise, I can try again with another prompt.  It uses a freemium model, which means I can start creating for free (you can too!) and then it charges me after a certain point.  But I found it leaned country more than I wanted so I sought something different.

More recently, I've come to prefer Udio for its ability to create different movements that I can control in songs like Coming of Age.

AI scraping draws from elements in a reference bank of existing art pieces to inform the creation of new content.  Let's get one thing right: scraping doesn't involve lifting copyrighted material for republishing, in whole or in part.  An algorithm learns methods and reproduces them.

Before you get your knickers in a twist, people do the same.  An argument against generative AI in music creation cannot be made without a great deal of hypocrisy.  A person listens to many songs by many artists and lifts pieces and patterns that they find good, then assembles them into their own creation.  That's what creativity is.  The AI algorithm is the same.  It scrapes untold thousands of tracks in each genre and experiments with pieces and patterns, learning what humans approve of along the way.  If a person lifts an entire measure from a piece without credit or compensation, it's plagiarism and it's unfair.  Just like a human, an AI avoids lifting and reproducing any portion of a song; rather, it learns patterns, tones, rhythms, and so forth so it can reassemble them into something fresh.  Rather like a DJ.

Generative AI models should scrape songs from legal sources, paying appropriately for the practice, and, really, nothing indicates that they don't.  Just like an old-fashioned person of flesh and blood -- "ugly bags of mostly water", per Star Trek -- the AI should properly subscribe to a music service or otherwise compensate in a mutually agreed upon manner, or pay normally for tracks for download, and learn from those.  I'll also hasten to add that AI content should always be denoted as such.

This contention and ultimate legitimacy is not unique to music AI, that's simply the focus of this piece.  You can see a close derivation from Futurama in the AI-created banner to this page, which was created from the simple prompt: "A robot in a recording studio plays electric guitar".  It was made using Vidu and I would say, this is more of an edge-case.   There's a tool out there making every kind of AI art imaginable.  None of their output is quite as good as direct, human achievement.  Yet.

Just stay clear of the AI writing assistants; they're poison, as I point out in detail here.

Back to the silent-album artists, then.  What they're really protesting is their own fear.  AI will undoubtably cut into the market share in the years forthcoming.  The industry will adapt, just like it did when digital tracks took over the world of CD's, or when tapes encroached on vinyl records.  The technological progress marches on in lockstep with Ray Kurzweil.  Things change.  What these artists should do is compete.

As for their silent album, I say: shut up.

Listen to my full AI Rock album project She Used to Smile here, and my full AI Pop album project Robo which preceded it here.