With seven albums under its belt, Yacht has never embarked upon a process like this - but the band needed to do something different. In 2016, Yacht's foray into experimentation went awry with a “sex tape” scandal where it tried to tease a music video featuring graphic sci-fi alien sex as a stolen, leaked sex tape of its own. The internet -- and the world at large -- did not respond kindly. The band has since taken accountability for it.
“We have nothing but regrets about it,” Evans says. “Obviously it was a mistake. Being in the eye of a storm like that allows you to see the true nature of the internet in a way which made us realize we needed to take a long break.” The experience humbled the band. They didn’t tour for two-and-a-half years, Evans wrote a book, and quietly, the band began working on this expansive tech project with AI.
“I think part of the motivation of this record was to re-interrogate our own intentions as artists and rediscover who we were as artists,” she says. The “sex tape” scandal humbled Evans who admits “it was a dumb project.”
But during the three years since the “sex tape” scandal, Yacht worked on bringing Chain Tripping to life. Using neural networks to break patterns into variations, the group then re-assembled into completely new songs. To generate lyrics, the band used a range of AI processes including lyrics (Char-RNN), musical data (MusicVAE), raw audio (SampleRNN), and instrumentation (a “neural synthesizer” called the NSynth). During the process, the trio wasn’t dependent on any single tool, and similarly, it wasn't dependent on one single inspiration. Generative composition, as seen in William S.Burroughs’ jagged writing style and David Bowie’s “Verbasizer” lyrics software, became a throughline of the project.
“This wasn’t an efficient process, this wasn’t an intuitive one,” says Bechtolt, perched on a chair in the Dolby Interactive space in Soho. “Or a fun one,” Evans chimes in on the navy couch across from him. The group is preparing to share what they’ve worked on for the past three years with a select group in the space’s Atmos immersive room. With multiple speakers embedded in the walls, the futuristic setting was on-brand for the band who’s found tech to be limitless.
Later on, Evans discusses the creation of Chain Tripping with a panel of people who explore and thrive in fusing technology with creativity, including poet and programmer Allison Parrish who is also full-time faculty NYU's Interactive Telecommunications Program Cristóbal Valenzuela, co-founder of Runway, a radical new tool allowing creators of all kinds to use artificial intelligence and artist and Google Creative Lab technologist Maya Man, who brings dance to AI.
“We just knew that [AI] was an interesting space we wanted to explore,” says Evans. “We pretty much historically learned things by doing them.”
Evans knows the process can sound mind-boggling. “If I’m getting too technical pull me back, if I’m getting too high-level pull me in,” she laughs. To initiate the process, the band looked for things that were approachable to non-coders, and looked for technologists and creative technologists that were willing to collaborate with them. While they ended up using a couple of different processes, the two main sides were lyrics and music.
“We were interested in the reflective quality of AI,” says Evans. “In order to teach a machine-learning model anything, you need to train it.” To do that, the band needed to use historical back-data. “We liked the idea of, ‘oh we can use this to better understand ourselves, better understand what a Yacht song is and where it comes from,’” adds Evans.
Yacht took its entire back-catalog - which is 82 songs and 17 years of music - and notated it in MIDI, later running it through a machine learning model called a latent space interpolation model. The band started out with hundreds of pieces of music that were just symbolic -- just melodies -- and had to sort through the data, pick out the most alluring sections and structure songs based on all the pieces.
The challenge didn’t end there: the band had to learn how to perform them.
“Since these models don’t consider the human body, a lot of it was very difficult to play and it would take forever to learn two bars of music which was super frustrating at times,” says Bechtolt.
While they could have made this record forever, they made it with one pass of the lyrics model. For the lyrical aspect of the project, Yacht worked with “algorithmic coder poet” Ross Goodwin.
“Everything we’ve ever loved is somewhere in the DNA of the lyrics,” says Evans. By that she means, the lyrics model is trained on the band’s favorites: everything from Patti Smith to Silver Jews and David Bowie. The machine, overall, was fed 2.2 million words.
“It doesn’t know language at all -- it would just write lyrics character by character,” says Bechtolt. What they got were pages and pages of repetitive words and phrasing, but they said it ended up being a “beautiful piece of paper.” The band used the repetition to their benefit.
One of the most jarring tracks is “Death." It was the first thing that came out of the machine and happens to be the slowest song the group has ever written. “I think that was the first [one] where we are going to leave that as is and use as much of this as possible,” says Bechtolt. The repetition of the word “stab” became a large portion of spoken word for the song, and gave the song a punk-rock ethos, while “Stab a cop” became the chorus.
With the record said and done, would the band put themselves through this one more time?
“Maybe we’ll do it again," Evans says. "Maybe we’ll do something different. The technology is evolving really fast."
Regardless, Yacht is proud of what they’ve accomplished together, and they’d love to see other people help define how to use AI to make art. “It’s this new wildcard element in the mix,” Evans says.
“But none of this would matter if we couldn’t stand behind the output," Becholt adds, "if we didn’t actually like the songs or if the songs couldn’t stand by themselves. Who cares if you did all this complicated boring stuff if you made stuff that was just ‘bleep bloops.’ Somehow doing all this hideous work, we actually found ourselves."
Chain Tripping is out everywhere on DFA Records Friday, Aug. 30.