A.I. musicians are a growing trend. What does that mean for the music industry?

Image used with permission by copyright holder

The most prolific musical artists manage to release one, maybe two, studio albums in a year. Rappers can sometimes put out three or four mixtapes during that same time. However, Auxuman plans to put out a new full-length album, featuring hot up-and-coming artists like Yona, Mony, Gemini, Hexe, and Zoya, every single month. How? The power of artificial intelligence of course.

Before this goes any further, don’t worry: You’re not hopelessly out of touch with today’s pop music. Yona, Mony, Gemini, and the rest of the bunch aren’t real musicians. Well, at least not in the sense that you could meet them and shake their hands. They’re A.I. personalities, each with their own characters and genres, which have been created by Auxuman, an artificial intelligence startup based in London.

Machines making music

“The music is generated through a few engines that create the words, melodies, and a digital singing voice,” founder Ash Koosh, a British-Iranian music producer, told Digital Trends. “At the core of songs that have lyrics, the text is generated using models that are trained on articles, poems, and conversations related to the subject of a song. Yona and her Auxuman friends are, in a sense, a reflection of human life on the internet. Expression on each song comes from stories we have told, ideas we have generated, and opinions we have shared.”

It’s certainly an interesting idea — and one which is absolutely on the money here in 2019. The first album can be listened to on Soundcloud here. While it’s all broadly electronica, the genres within this range from the more soothingly melodic ambient sounds of Yona’s “One” to the harder-edged sounds of Hexe’s “Try Me.”

MONY with his song MARLENE as part of @auxuman compilation album VOL.1 https://t.co/daOGjJtP9T pic.twitter.com/uFYZ9F06yJ

— Ash Koosha (@AshKoosha) September 30, 2019

But why try and create A.I.-generated music at all? After all, as proponents of A.I. repeatedly point out, the real commercial promise of artificial intelligence is to automate those jobs that are too dull, dirty, or dangerous for humans to want to do. There’s no shortage of people who want to be professional musicians, so why focus on this as a task for automation? “There is always [a shortage of people] for giving birth to a new genre,” Koosh said. “Due to the economic nature of the act of making music for humans, we are naturally submissive to forms that have been successful. We believe machines [can] blend and merge forms and styles [and, in the process], find the next exciting sound.”

That’s certainly a tall order (and one which, at least based on the first released album, hasn’t been achieved.) However, Koosh may be onto something. Creative A.I. has already been used to help formulate and test new drug hypotheses and design new satellite components. In an age of online niches, why couldn’t it be used to create new musical genres?

A genre like alt-horrorcore salsa (my own invention) may never build up enough of a critical mass of listeners to make it commercially viable. But if music could be cheaply generated by machine and then distributed to a small dedicated audience willing to pay for it, it might just represent a new type of musical distribution.

Is Auxuman genuinely creative?

Whenever a story like this breaks, the first question a lot of people ask is, “yes, but is it actual creativity?” It’s a tough question, and the answer is both yes and no.

The problem with creativity is just how difficult it is to define. There are very few metrics that determine whether something is creative and, if it is, how creative it is. Last year, a painting generated by an A.I. art collective sold at a Christie’s auction for $432,500. That was far more than both its estimated sales price and the amount that most artists will ever sell a painting for in their entire lives. Do we therefore assume that the A.I. is one of the more accomplished artists out there? Probably not, in the same way that movie buffs will frequently talk up the artistic value of an underground indie movie while downplaying the work of a filmmaker like Michael Bay who, objectively, earns far more money at the box office.

I’ve got a two-year-old daughter who constantly displays creativity by learning to solve problems. But these are problems that everyone learns to solve as a two-year-old, and so they aren’t considered by society to be especially valuable.

There are also double-standards when it comes to machine creativity. A machine that learns its knowledge of the world from datasets, rather than “original” inspirations, is not considered creative. A human artist who has a deep knowledge of his or her field and references it in their work in new ways is considered creative. A machine is considered non-creative because it isn’t trying to express a particular emotion and isn’t imbuing its creativity with deeper themes. But literary critics have long talked about the “death of the author” and how it is up to the audience to interpret a piece of work.

Cyborg music

The bigger question with this kind of machine creativity is how much of it is actually done by humans. The answer here? A reasonable amount.

“We call this the puzzle process, a process in which multiple engines — text, melody, and voice generators — create valuable pieces, such as poetic verses [and] melodic sequences … and the human curator puts these pieces together,” Koosh said. “These pieces follow rules that are directly reflecting values from each Auxuman’s personality case file, resulting in pieces that will always match and build a new interesting body of work. The final production is informed by our engineering skills as humans. Decisions we make shape the final outcome.”

In other words, this kind of machine creativity is more along the lines of something like the “cut-up” technique of writer William S. Burroughs. Starting in the late 1950s, Burroughs began exploring ways that new text could be generated by cutting up existing pieces of writing and rearranging them to form something new. Artificial intelligence makes these kind of remixes more interesting than ever. It adds a level of complexity that takes away more control from the creator. It’s as if another quasi-intelligent entity is playing a role in the production process.

Thomas Hobbes once referred to imagination as “decaying sense,” meaning that it is a sort of obscured memory of a thing, often mixed in with other memories. This is what makes A.I. creativity so interesting. A.I.s which create jazz riffs, computer games, imagined monologues from TV shows, and more take fragments of something recognizable to us and then mix it up and reassemble it using computer logic. It’s a fascinating example of humans and machines working together to produce something new.

The hype conundrum

Some people will dismiss this kind of A.I. creativity as hype. There’s probably an element of that. The moment that paintings created by machines start selling for half a million dollars at auction you know that there’s a vested financial interest in building “creative” machines. But if you view this as what it is: a fascinating new tool that can augment human creativity rather than replace it, the whole thing becomes a whole lot more compelling.

“Machines are creative, but in a supplementary manner,” Koosh continued. “Human creativity will be fundamentally different due to our biological nature, needs, and intentions. Our memories are recorded and played back on a very complex system in our brain, but machines can begin with rules and iterate towards possible tangents that a human creators might take months to reach. Machine creativity comes down to maximizing possible outcomes that we humans as curators could look at and choose to apply in a creative process.”

What creatively minded person, interested in pushing the envelope, could possibly find something wrong with that?