The 1956 composition "Illiac Suite for String Quartet" is a pleasant enough sounding piece of music – for the first three movements, that is. It's when you get to the fourth and final movement, that things get...weird. The notes sound random and dissonant. It doesn't sound much like music at all. But the peculiarity of "Illiac Suite" makes a little more sense when you realize how it was composed. This was the computer's first algorithmically generated song.
Programmed in binary by Lejaren A. Hiller, assistant professor of music at the University of Illinois, and Leonard M. Isaacson, a former research associate on the school's Illiac computer, "Illiac Suite" was nevertheless a revelation. That a computer might one day compose music indistinguishable from that of a human artist became an irresistible pop culture trope – for better and for ill. In his New York Times obituary, Hiller is said to have joked that "he would have computers compose all possible rock songs, then copyright them and refuse to let anyone perform them."
Luckily for us, computers are nowhere close to realizing that humorous albeit dystopian vision. And yet "Illiac Suite" remains an impressive feat, even today.
We can actually trace the beginnings of "Illiac Suite" back to none other than the British mathematician and computing pioneer Alan Turing. In 1951, Turing published a book on programming for an early computer known then as the Ferranti Mark I*. The machine had a loudspeaker, sometimes called a "hooter," that was used primarily to issue warnings or during debugging. But Turing found that the loudspeaker could also be used to produce solid tones – notes, if you prefer.
It didn't take long before programmers began to exploit this functionality to playback simply melodies and songs. But two programmers by the name of David Caplin and Dietrich Prinz decided to take things a step further.
Caplin, a mathematician, had access to a Mark I* as part of his work for Shell Chemical – yes, the oil and gas company – while Prinz was a research scientist at Ferranti (and author of one of the earliest chess programs, too). In their spare time, they used the Mark I* for musical pursuits – specifically, one of the earliest attempts at algorithmic music composition with a computer currently known.
Algorithmic composition, it turns out, has been part of the Western composition tradition for at least 1,000 years.
But here's the crazy part: this wasn't really all that revolutionary. Algorithmic composition, it turns out, "has been part of the Western composition tradition for at least 1,000 years," writes Michael Edwards in his 2011 paper Algorithmic Composition: Computational Thinking in Music. The program that Caplin and Prinz wrote was actually based on a musical composition algorithm used by Wolfgang Amadeus Mozart, nearly 200 years prior.
Musikalisches Wurfelspiel, often described as one of the best known early examples of algorithmic composition, was Mozart's musical dice game. It could be used to compose music – for a piano in this case – with harmonies in the left hand and melody on the right, by using dice to select from 176 pre-composed measures. This was, in essence, still an algorithm, even if it was computed manually. In 1955, Caplin recorded a Ferranti Mark I* performing a Mozart dice game program written with Prinz.
The output was very basic, of course, limited by the range of Turing's Mark I* loudspeaker sound generation technique. Only the melody was audible, and the pair had little control over tempo, dynamics or octave. It was really just a succession of tones – and this is where Hiller and Isaacson come in. The pair, who were researching similar work at the time, decided to sidestep this output problem completely by having humans perform the generated notation instead, allowing for the creation of more complex music at the same time. A full suite of music, in fact.
Hiller and Isaacson's efforts at composing "Illiac Suite" were documented in the 1959 book "Experimental Music: Composition With an Electronic Computer." Using the Monte Carlo method of random data sampling, the pair could essentially train their Illiac computer to understand certain rules of musical composition. They built models that knew, roughly, what notes sounded good or bad together, for example, and the basics of time and pitch. One of the many challenges, according to Hiller and Isaacson, was deciding just how random these musical decisions should be – the balance between order and the so-called "chaos" from which infinite musical possibilities could be plucked.
The composition's first three suites are based mainly on generating music that falls "within [these] traditional framework of compositional procedures," the pair wrote – basically, generating music with simple, recognizable melodies with varying rhythms and dynamics. But the reason that the fourth suite sounds so cacophonous is that...well, here Hiller and Isaacson set out to create music using algorithms and computational techniques that weren't based in musical theory, to see if decent music could still be produced. They called the result "Markov chain music" – after the algorithm used – with notes generated randomly based on the note that came previous. It sounds about as good as you might expect.
At the time, the idea of using a computer to create music was unheard of – criticized, even, according to the foreword of Hiller and Isaacson's book. But there's a reason that "Illiac Suite" is regarded today as the "first substantial piece of music composed on a computer." The composition – along with the development of the modern synthesizer soon thereafter – would spawn a whole field of music research and development known as computer-aided algorithmic composition that is still taking place today. And though we still haven't managed to program a computer that, of its own intelligence and control, can compose a listenable piece of music completely on its own, maybe that's okay. Even when the computer composed its first song, it was really a human-computer collaboration that brought it to life. Mozart meet Multivac.
I like that the computer composed it, but humans performed it. It does work as music, although it feels like a succession of snippets (which it is), with no theme development and no conventional ending
Some of the music Wendy Carlos wrote for Tron isn't even based on the European 12 note scale, but it's still solidly based in the harmonic tradition. Incidentally, she got her masters in composition from Columbia University working at the Columbia-Princeton Electronic Music Center, so is well versed in algorithmic composition. Another pioneer in computational music was Raymond Scott. His Electronium could compose and play music in real time in 1969!