Synthetic Symphonies Part II: ABBAtar

Last year, The ABBA: Voyage live show launched in London. The show features four digitally generated ABBA-tars - life-like holograms - of Benny, Bjorn, Agnetha and Anna-Frid who front a live, human, ten-piece band.

They glitter, sequence and disco their way through the picks of the back catalogue, from The Visitors to Thank You for the Music. ‘Dancing’ tickets on the ABBA Arena floor are priced at a pretty reasonable £53.

Very reasonable, in fact, when compared with the £325 that ViaGoGo would have me pay to see Bruce Springsteen.

Though I suppose Bruce Springsteen would have a pulse.

But how much does that matter?

In an interview given outside the ABBA Arena, Bjorn discussed the audience’s reaction to the show and was amazed by the ‘emotional connection’ between the crowd and the avatars.

I can’t help but feel as though the whole thing must have been a surreal experience for him.

To sit in an Arena - so saturated with light and sound that the show’s producer describes it as space that is ‘neither physical or digital’ - to stare into the uncanny valley of the stage where synthetic projections of your own past play to the adoration of thousands.

How would you consider yourself in his shoes? Superstar? Artefact? Spare-part?

Now, one could argue that the ABBA-tars are a step removed from the true human art of ABBA: that magnificent trans-generational, trans-national, unforgettable songwriting. They are just vessels communicating something that is still wonderfully, wildly real, like Ode to Joy emanating from the demo-store on a child’s first mini-keyboard.

And digital performers are nothing new: the ABBA-tars are, in some real sense, the latest manifestation of a tradition that includes The Gorrillaz, The Wombles, The Smurfs, all those animals from all of those Disney Movies. I’d even be tempted to throw in some of the boy/girl bands of my own childhood in the late 90s/early 00s - in a sense they were their own sorts of performative avatars, lip-syncing to someone else’s thoughts and feelings.

But what about the now very conceivable world of the near future, in which digital pop-stars perform music of their own creation? In which the human being has been removed entirely from the equation?

It begs the question: just how central is the human experience to the creation of art?

It’s a fairly weighty question for a weekly blog post.

Last week, we discussed Dall-E2 and Chat GPT: programmes developed by Open-AI (though other terrifying, cyber-net style brands are available).

Open-AI are also responsible for MuseNet, a machine learning algorithm which - you guessed it - composes its own music.

The music isn’t entirely its own: currently, as far as I can tell (the system isn’t currently available for users to experiment with themselves) it works by having a user feed it a musical prompt - four notes or so - and requesting a style (a genre, or a recognisable composer/songwriter). The AI is then able to extrapolate this into a four-minute piece of original music.

In 2019, MuseNet performed a concert live on Twitch (a live-streaming video platform) consisting of entirely original pieces in various styles that even its own software engineers were hearing for the first time.*

And, listen - it’s not good.

The music is jarringly predictable. It finds itself in loops of rhythm and melody that it can’t quite jump out of and this can be unsettling.

But the music does make sense. There are even moments, phrases, within it that catch your eye and illicit that most inexplicably human of all reactions: interest.

What stunned me the most, however, on the MuseNet site, wasn’t the music. It was this brief paragraph of exposition near the top:

“We’ve created MuseNet, a deep neural network that can generate 4-minute musical compositions with 10 different instruments, and can combine styles from country to Mozart to the Beatles. MuseNet was not explicitly programmed with our understanding of music, but instead discovered patterns of harmony, rhythm, and style by learning to predict the next token in hundreds of thousands of MIDI files.”

In layman’s terms, MIDI files are just music expressed as digital information. They aren’t recordings of physical sound but, instead, transferable computer-speak for pitch, tone, volume, note-length and all the other characteristics of music that a system like Garage Band on your MacBook would need to understand in order to digitally regenerate a piece. Picture the difference between a Polaroid picture and a Jpeg.

The amazing this about that last sentence is that it tells us how MuseNet learned to compose.

It wasn’t taught; it was self-taught. It soaked up the history of music and trained itself, by trial and error, to be able to predict where melodies, rhythms and chord progressions should go, based on everything it has ever heard before.

In other words, it developed that ghostly intuition, that sixth sense that human composers have: musical instinct.

Because don’t we humans do the same?

From birth we are surrounded by music and our own neural networks begin to form, guided by circumstance. We learn to associate scales, sounds, keys and patterns with one another. We recognise harmony from discord. We hear Mozart, The Beatles and everything else beyond and in between, just like MuseNet. And our subconscious brains are indelibly shaped by what we hear; but not just what we hear: in what quantities, and at which moments we hear it.

My own methods of composition are, undeniably, similar to MuseNet’s. I catch a snippet of a song on the radio or in the gym and before my rational brain realises it I feel a pang. The melody didn’t go where I was expecting it to. The lyric didn’t quite fit. I stop, and consider. What if they’d have done this? Or that? Oh - that’s good. I grab my iPhone and whistle something into it. Something to come back to later and transform, hopefully into something else - something new.

And where I take that idea is neuroplasticity in action. It isn't really up to 'me'. I’m just running with it down creative corridors in my brain that exist despite my conscious desire to expand or escape them. It’s tough to imagine what you haven’t heard. My years strumming Bob Dylan songs in my bedroom, my strange affinity for the Bee Gees, my brief obsession with Tom Waits means that my own neural network has a unique shape consisting of unique connections in unique quantities.

Take the same snippet or radio song and ask Dan, our guitarist, to do something with it and the end result will be entirely different. He runs it through the processing machine of his own brain and it takes on a different shape.

Like MuseNet, we’re bound by our own neural networks. We’re bound by the nature of our hardware and our software.

The trouble is, that that MuseNet’s hardware and software is potentially limitless in its complexity.

Let’s remind ourselves what MuseNet’s sister, Chat GPT, told us last week: ‘AI algorithms can analyze vast amounts of musical data and generate new compositions based on that information, leading to the creation of unique and diverse musical styles.’

Unique? Diverse?

Sounds like a tall order.

For a human.

For now, let's return to our question from earlier on: in light of all this, just what is the relationship between music and the human experience?

It strikes me, having met MuseNet, that If a machine designed by humans is composing in the most human of ways, shaped by a human-generated musical history, then it's tough to argue that, in a very real sense, the human experience isn't still inextricably woven into the fabric of the AI-generated music that is bound to be everywhere before long.

But if that's the case, why does all of this still make me feel so uncomfortable?

More on this next week.

Keep dreaming,

Rob

* https://openai.com/blog/musenet/ head here to listen to MuseNet for yourself!

‍