Holly Herndon, “Spem in Alium,” and new frontiers in polyphony.
Earlier this year, Pitchfork launched "The Song I Wish I Wrote" video series, where musicians discuss songs that have influenced them to the point of envy. My favorite so far has been Holly Herndon talking about Thomas Tallis's 1570 polyphonic masterpiece “Spem in Alium," which she describes as an "early, multi-channel surround experience."
Often cited for using AI to take her work in new directions, Herndon imagines how "Spem in Alium" might be conceived of today:
I think a machine-learning algorithm could write this now. If this was the canon, then a machine learning tool could emulate something similar. And that's why you can't take music out of context. We are really obsessed with past canon, and then training on it and then repeating it. That's not what's beautiful about music and musical expression. It's about responding to your place in time and your environment and your own personal expression of that.
This really is the essential difference in how we use artificial intelligence, in any capacity. Are we using it as a tool to expand our own capabilities and create something new and unexpected, or are we instead using it to produce vast quantities of something?
In a new video for Dropbox—which, OK—Herndon describes her process for making music with her AI, which she's named "Spawn":
Here she describes the creation of Spawn:
Spawn is our AI baby, and when I say that it's basically a metaphor for the experiments that we've been doing with music using neural networks. One of the reasons why we chose a child or baby metaphor when talking about Spawn is that we really see it as kind of nascent, baby-like technology. So it's at such an early stage that, you know, we're trying to imagine what if this community, what if our approach to raising her could really kind of have some sort of impact on how she grows up.
So there are a couple reasons why we decided to create our own datasets. I mean, at the end of the day artificial intelligence is just us. That's kind of the dirty secret of AI is that you have this vast amount of human intelligence that goes into the training sets, and so we really wanted to take an approach where we would make audible and name and give attribution to those who went into training Spawn.
In a previous "Spem in Alium" experiment, on June 6, 2006, conductor David Lawrence led more than 700 singers—as the People's Choir—in a massive performance of "Spem in Alium," which some of the singers had never even performed before.
A dispatch from one of the singers:
So: eight five-part choirs, one to four in the stalls, five to eight up above. When all sang, it was possible to sense for the first time the music's spatial quality, its flow from left to right and back again.
Even so, we thought the conductor David Lawrence had flipped when he instructed us downstairs to turn and face those upstairs. He gave a couple of beats to bring in the altos of choir one and then left us all to get on with it, to sense the pulse that runs through the work.
But it was a sensational moment of time outside time, with we basses of choir one embraced by the sound drifting down from the sopranos of choir eight. The temptation to stop singing and just listen was huge.