Just about everything in the world is awful, except this fake Coachella lineup generated by a neural network, which is perfect.
The fake lineup was created by Botnik Studios, a company that uses computational tools like neural networks (a subset of artificial intelligent that tries to imitate the way a human brain works by identifying common trends from lots of examples) and predictive keyboards (like the one on your phone, but more complicated) to create digital art. Botnik Studios puts the systems behind all of their creations online in their Facebook group, Botnik Gallery, for anyone to use and create with.
You may have heard of Botnik before, or, if not, you’ve probably seen their work. Quite a few of the community’s other projects have gone viral over the years. Some personal favorites are an episode of Seinfeld, a list of Halloween safety tips, and a new chapter in the Harry Potter series. All of these projects were done using a predictive keyboard system, which combines the narrative-aware input of human comedy writers with the creative content-generation abilities of a neural network, however, the Coachella lineup was the first to be created differently. Since a list of band names doesn't have to conform to grammar rules, or contain a coherent plot line, they were able to generate the lineup purely from the neural network.
I called up Botnik’s creative director, Michael Frederickson, in order to learn more about the creation of the Coachella lineup and Botnik Studios as a whole.
The Outline: So, how did you come up with the idea for the fake Coachella lineup?
Michael Frederickson: Coachella hasn't happened yet, but the lineup is out, so it’s recognizable, and we wanted to play on that feeling that like, no matter how culturally in touch you are, when a festival lineup comes out [and] you're reading through it, [you’re] inevitably gonna be like “Who the fuck is that?” It just sounds like blank verse poetry or some shit, and you're just like “How do these fucking people name these bands?”
So with this one, we got a list of every band that Pitchfork had ever reviewed. I don't actually know how. Someone just found one somewhere, so, we seeded the neural net on that, and then, in this particular case we did something kind of crazy: We had it generate 100,000 band names.
Wow, that’s a lot. How did you go from 100,000 to what’s on the flyer?
We divided up that 100,000 into chunks of a thousand names that we called an “acre,” and then assigned the acres to people on the team to go through this thousand, send us back the ones that appealed to [them]. Because the neural net doesn't know why they're interesting, but I as a person who's looked at band names, or have these comedic sensibilities, am applying that to it, and think it's good.
So then, those all got sent back to us. Then we like whittled that list down even more ... I went through like a maniac and just counted how many headliners and stuff there were all these different tiers. “How many Scandinavian folk singers do we need?” Like, that was all just kind of manual.
Then we just took all those things and then we slotted them into this big spreadsheet and just pitted them all against each other — like “Is this funnier than this?” and so on — till we landed at like, a hundred and sixty-something, and then ordered them all so you didn't see the same letter or sound repeating next to, you know, similar ones nearby, and shit like that.
Dang, that sounds like a labor of love if I’ve ever seen one.
Then I sat there for all of Sunday putting all those little dots between all the names and my freaking hand is still sore.
One of the things I’ve always found so interesting about Botnik projects is that your use of both human and machine input gets rid of what I’d consider the worst part of neural network projects, which is the meaningless nonsense that is inherent in any 100 percent machine-generated work.
Our whole jam is trying to hit this balance between something that the computer just spit out and something that a comedy writer wrote in a kind of vanilla way, and trying to split the difference. We found that people who get really excited about the algorithm sometimes are not as in tune with what human artistic intuition brings, and someone who's been practicing comedy their whole life [isn’t] examining what might be artistically interesting about elements of tech.
I think [there is] this kind of a fallacy [in] computer science education in general that you absolutely need to know how every single element of something like this works in order to be able to say what's interesting about it, or use it.
Tell me a little bit about Botnik. When did this all start?
So I've been involved with it for a little more than a year, but Jamie Brew, who is the founder, he was the head writer at ClickHole. We didn't know each other in college, but we both we both went to Brown and we wrote for the Brown Noser, which was the comedy paper there. Somebody introduced the two of us to each other last year because they said we were both independently doing similar tech/comedy/art stuff, and we talked for like a million hours. It was like, “Oh shit, we're actually on the same page about this.” So Jamie had done this as a side project when he was at ClickHole, and had made a Tumblr called Object Dreams, where he was putting all the stuff he did with it.
And when I saw that I was just like, oh my god, this is like exactly what I care about. And, my day job is at Pixar, where I do a lot of like combining art with tech type stuff, but nothing with machine learning or anything like that, or just like pure comedy. So, this is just like just an immediate, “Oh, we've got to do it.” So Jamie and I talked for a long time and then like towards the end of the year was when we actually started structuring a little bit of a team and started setting up a comedy writers room format, where we pitch every week and schedule, so we can keep churning stuff out.
This interview has been condensed and edited for clarity.