Computer Program Composes Music That Fools Humans
Last year, more than 200 listeners participated in two online studies where they were asked to see if they could tell the difference between music written by a human and music written by a computer program. In a classroom on the campus of Yale University, lecturer Donya Quick gave me the same test she gave participants in the studies. She played audio clips of short melodies and asked me to rank them on a scale from absolutely human to absolutely computer. Some were written in the 18th century by Johann Sebastian Bach. Some were written by Kulitta, the computer program Quick designed last year. Here’s how my test went:
I’m not the only one to be fooled. Just over half of the people who participated thought Kulitta was human. Quick said the way Kulitta and other programs compose music is very similar to the way humans do.
“Kulitta has two things to draw on for decision making. One is a random number generator," she said. "Lots of composers draw from randomness. That’s a very common thing in music, and they're wonderful to listen to. And yet they sort of get uneasy about that if a computer is doing the same thing.”
Quick said the other thing Kulitta draws on is a set of rules to narrow down its choices. Those rules include the tempo, and whether it’s in a major or a minor key. Quick said Kulitta goes further than most programs that compose music. It picks chords and notes based on their role in the piece, kind of like how we choose words in a sentence. In language, we choose words for a reason. In music, we choose notes for a reason.
“Kulitta has definitely shown that there is this very strong link between language, spoken language, and music," she said. "Taking a linguistic outlook on music seems to produce surprisingly good results, and people find that to be more natural and more relatable, and they’re more likely to think that it’s produced by a human.”
Quick said she hopes in the future, when composers need inspiration, they could use Kulitta to generate new music that could help them with their next composition.