Can computers compose music?
They will soon be able to drive better than us, they can already beat us at chess and they can out-think our best investment bankers, but will computers ever be able to compose music? Surely Bach is beyond even the best AI? It cannot be possible that a machine like Watson will ever compare with a surly and clever genius like John Lennon. It is again surely not possible that a place like Paisley Park might one day vibrate to the playful modulations of Prince’s computers as they start to play of their own accord?
This is the new line of defence. People say ‘yes, Artificial Intelligence can even beat the sharpest tiger kid at Math, but it will never create like an artist.’
Well don’t be so sure. Here is Sony’s CSL project making a passable attempt at The Beatles, and here is the same system mimicking Bach. When her compositions were performed live, ‘Emily Howell’ fooled classical music critics who thought her a decent new composer. Pass a few parameters to Jukedeck and its AI will generate a soundtrack to your video.
Like it or not, music has always been technological. Its form and boundaries are given by the available technology of the day and musicians have often been deeply interested in the very latest instrumentation. The strings, wood and varnish of a Stradivarius were hi-tech in 1700. In 1810, when Chopin was born, the piano was already a century old but only just coming into the domestic settings of the well-to-do. Why did so many great bands originate in the 60s? Well, because electric guitars and amps made it to the high street in the 50s. Synth bands followed Casio and Roland in the 80s. Hip hop followed the popularisation of DJ mix decks. Toshiba led digital sampling into the market, and what would 2Pac or Snoop Dogg have created without it?
It is impossible to believe that music composition will not be affected by something as powerful as Artificial Intelligence. Even if one does not cede total control to the machine, one can imagine AI working out the string parts, figuring out the piano and editing the song into three, five and eight minute versions. It will suggest twenty-thousand different ways to follow your middle-eight and help you pick the most impactful option.
Go to a ‘live’ performance today and many, many times what you will actually hear is a kind of hybrid, where the band plays to a click-track and that click-track synchronises lights, backing vocals and additional musical parts. How long before the computers extemporise and jam live, adding colour and mood to a performance according to the playing of the band and the responses of the crowd?
AI as your live jazz, jam partner? It is going to happen.
Remember one thing about AI: it learns. It gets better. Here is that Sony system again, this time trained by Kumisolo with a few of her favourite songs.
Written by Peter Kawalek