Four Paths to the Future: II. Artificial Intelligence

It is four years now since I visited computer scientist François Pachet in his office at Sony’s blue-sky Computer Science Lab in Paris’s 5th arrondissement. At that time he was working on a project called Flow Machines, a system he believed could be “the next generation of interactive systems.” The software was intended to work as a partner for improvising musicians, able to follow even the most unexpected changes in timing and key.

François Pachet's Reflexive Looper, a loop pedal that uses AI

Effectively, he was trying to teach jazz to a computer. He told me he wanted to take it busking on the Paris Metro (“I don’t expect much money…” he admitted). So I was intrigued to see in the news just a few weeks ago that Pachet had apparently been poached by Spotify. Intrigued, but not altogether surprised.

The Flow Machine Project

The history of artificial intelligence approaches to music is almost as old as artificial intelligence itself. It was Ada Lovelace, Charles Babbage’s own “enchantress of number,” the first computer programmer and first theorist of algorithmic art, who insisted that though Babbage’s engine might be induced to create music, it had “no pretensions whatever to originate anything.” The twentieth century has tended to perceive this more as a challenge than a rule to be bound by.

"Mechanical Brain Takes Up Composing Music" was the headline in the Champaign-Urbana News Gazette when Lejaren Hiller and Leonard Isaacson premiered their Illiac Suite – a string quartet, named after its "composer," the Illinois Automatic Computer (or ILLIAC). Like many of the projects to succeed it, The Illiac Suite employed a combination of random numbers and Markov chains, strings of probabilities that tell the computer whether, say, A is more likely to precede B or ⻬. One of the most celebrated music AIs of the twentieth century, David Cope’s Experiments in Musical Intelligence, was essentially a more sophisticated version of the same principle.

The Illiac Suite for String Quartet

"Bach-style Chorale" was composed by the Experiments in Musical Intelligence computer program created by David Cope.

More recently, the focus has shifted from Markov-based systems to neural networks. There are now countless different AI music projects, several of them – like Google’s Magenta, the Baidu AI Composer, and whatever Spotify are planning on doing with François Pachet – are funded by the major social network companies. The question is, why? What would the parent company of a notoriously royalty-averse platform like YouTube want with a program that could create music at the touch of a button, without human intervention?

The Baidu AI Composer

"Daddy's Car" is a song composed by Artificial Intelligence in the style of the Beatles.

Robert Barry is the author of The Music of the Future, published by Repeater Books earlier this year, a history of speculative musics from the Enlightenment to the present day. In "Four Paths to the Future," a new four-part series for Q2 Music, he examines the past, present, and future of music through the lens of a set of topics normally reserved for science fiction or futurological think tanks. "Four Paths to the Future" will provide an imaginative glimpse of where we're at and where we might be heading, through the thoughts, sounds, inventions and ideas of the composers, thinkers and technologists who could be taking us there.