, with Marcus du Sautoy, Mahan Esfahani and Robert Thomas, Barbican, London, 9 March
CAN you tell when a piece of music has been written by a machine? Back in 1979, cognitive scientist Douglas Hofstadter was the first to ask that question in his classic book
Forty years on, I thought it was a rather tired question. Of course we cannot tell. Of course we can be fooled. Why worry? After all, no one stopped playing chess or Go when computers proved they could trounce the best players. If anything, the machines inspired people to play more, and better. Pitting yourself against a human adversary is the whole point of these games. And if the point of music is that it conveys emotion, it is only interesting if there is a human doing the conveying.
Advertisement
A concert on 9 March should shake up my assumptions. London’s Barbican is bringing together harpsichordist Mahan Esfahani, mathematician Marcus du Sautoy and composer Robert Thomas for a performance lecture – called The Eternal Golden Braid: Gödel, Escher, Bach – that uses an algorithm trained on the music of J. S. Bach.
Bach’s compositions have been fed through a machine-learning process created by computational artist Parag K. Mital. It will use what it has learned to create its pieces.
The audience will listen to Esfahani playing a piece that interlaces real Bach with Bach generated by AI – and be asked to look for the joins. When people think they have spotted one, they can flip a card that is a different colour on each side. They will also listen to new pieces by Thomas and another AI-savvy composer Robert Laidlow.
The point isn’t to fool anyone into misattributing music created by AI to a composer regarded by many as the greatest who ever lived. Instead, audience responses will be used to create new music that explores Bach’s sound world and vocabulary.
Musicians from the London Contemporary Orchestra will also work with the audience. But you will need to be there to find out how it will work.
“My feeling is that people find themselves stuck in a particular way of doing things, and that’s when we start behaving like machines,” says du Sautoy. “My hope is that artificial intelligence may free us from behaving mechanically, by showing us that there are new places to go.”
“As we search for new musical territories, must we confront ever stranger sound worlds?”
He cites the work of computer scientist , director of the Spotify Creator Technology Research Lab. His Flow Machine program jams with jazz musicians in real time, leading them into improvisations that feel natural – and rightly so, since they are derived from a deep learning of the musicians’ output.
How does Esfahani feel about such technology? I expected him to be either enthused or threatened. I didn’t think he would regard it as business as usual. “Every innovation has unintended consequences,” he says. “But these include positive consequences.”
For Esfahani, the world of classical and contemporary music is anything but a stable environment – it has been in a state of reinvention for centuries. “From Mozart’s birth in 1756 to Schubert’s death in 1828 is no more than a single lifespan,” he says. “Yet in that one generation, the instruments of the orchestra became unrecognisable – sometimes literally so.”
It is true that AI threatens to decentre much of human life, but this continuing reinvention of music means it is relatively safe. Feelings will run high, though. In the 19th century, for example, the German composer Richard Wagner caused great outrage with his radical style. Philosopher Friedrich Nietzsche went as far as to say: “He contaminates everything he touches – he has made music sick.” In fact, Wagner exploited and exhausted contemporary harmonic and chromatic possibilities to the point where, at the turn of the 20th century, younger composers had no choice but to abandon tonal music in a search for a sound of their own.
Will the algorithm used during the upcoming concert reveal compositions that are easier to swallow? Or, as we search for new territories, must we confront ever stranger sound worlds?
As a mathematician, du Sautoy thinks he has an answer. “When I make the mathematics-music connection, people worry that I’m taking the emotion out of music and making it very cold, clinical and logical,” he says. “What they don’t realise is that mathematics is highly emotional. It’s a response to the play of extraordinary, surprising patterns. I get the same buzz reading mathematics as I do when I’m listening to Bach.”
Music isn’t an arbitrary jumble of notes. It is iterative, generative, algorithmic. Music can be easy and banal, just as mathematics can be, and for the same reason: structurally, easy music isn’t particularly interesting.
For both mathematics and music, the point isn’t to hunt for novelty for novelty’s sake, but to look for results that are interesting and surprising, and that lead to further discoveries. Such results are always rare, and the limits of human cognition set a hard barrier beyond which the search becomes pointless. By applying AI and machine learning to the problem, beautiful surprises may await us.
