Holly Herndon on Raising Her AI Baby and Embracing the Future

Holly Herndon on Raising Her AI Baby and Embracing the Future

Last fall, a movie starring Natalie Portman playing a traumatized pop star hit select theaters. Uncannily, Vox Lux arrived around the same time as Lady Gaga's turn in A Star Is Born. Comparisons were inevitable but inaccurate, and in Portman's pop star opus, Vox Lux took its name from her character Celeste's comeback album. In the film, Celeste told the press Vox Lux was a collection of "sci-fi bangers." The film was tragic, futuristic, and referential to issues afflicting modern society, from gun violence to drug abuse. I saw it three times.

Holly Herndon, an experimental electronic musician who spends time between Berlin and California, loved Vox Lux, too. I admit that Herndon's newest album, her third studio effort, PROTO, feels truer to Celeste's description than hers. Herndon, who just earned her PhD in composition from Stanford, laughs at the idea. But PROTO, which has thus far been described as an academic and artistic exploration of the relationship between rising technologies such as AI and humankind — which it is! — can also be understood as a pop culture artifact in its own right.

Like so much art these days, Herndon's work is a reflection of the times. Its nuanced synthesis of electronic manipulation and pop songcraft — including choral vocals from an international ensemble, digitized voice renderings from an AI baby she's raising named Spawn, and her own voice that's enhanced with vocoder — feels like a move to shift genre and conversation. To that end, PROTO's thoughtful, futuristic concept is very much in the vein of game-changing albums likeKanye West'sYeezus.

Listen once all the way through to the album's mix of spoken word, reverb-heavy vocal training sessions, and meditations on world affairs. If that doesn't grab you, hear Herndon's single "Eternal" and tell me it isn't totally in its own lane. Plus, that track is inspired by the enduring love story of billionaire biotech entrepreneurs Martine Rothblatt and her wife Bina Aspen, the latter of whom has her own AI robot, called Bina48.

As a whole project, PROTO advocates for a world that bridges gaps between man and machine, for the greater good. The album was largely created using AI-friendly, open-source technology for "machine learning," including programs such as TensorFlow and SampleRNN, which is described online as an "unconditional end-to-end neural audio generation mode."

If that feels intimidating, consider that Herndon created her latest body of work in constant communication with actual human beings around the world. Producer Jlin and artist Martine Syms are among the album's collaborators, plus, in the below interview with PAPER, Herndon regularly says "we" more than "I" or "me." Where digitally constructed music is increasingly void of human connection, Herndon's new work aims to "re-insert the human body into electronic music," she says. "We were craving having an ensemble, singing and harmonizing with people, being in the studio together. There's so much joy in that kind of collaboration and real human connection."

It's why songs like "Eternal" and its message of everlasting love (whether human or not), feels even more powerful today. It's why "Fear, Uncertainty, and Doubt" advocates for a sense of belonging, why "Last Gasp" asks us to reconsider why we're truly here. Not to, as that closing track says, "shift forward without aims."

Herndon's PROTO is about finding purpose in the technology available to us, and thinking of ways to amplify our connections to each other. The future is nothing to fear.

What began the process of making PROTO for you?

We were coming out of touring for what felt like forever and noticing that things were becoming really automated in electronic music. There were these perfected AV [audiovisual] shows with kicks perfectly synchronized with strobes and fog. It just felt like AV overload. We didn't understand where the human body or the performer was a part of that. It was almost like you could just leave the stage and the show would continue on its own. So we started thinking about how we could re-insert the human body into electronic music, so we were craving having an ensemble, singing and harmonizing with people, being in the studio together. There's so much joy in that kind of collaboration and real human connection. But we're talking about AI after all, so we thought it would be interesting to play with this idea of having human and non-human ensemble members all contributing to this collective whole that ultimately formed the project.

That's so insanely complex. When it came to Spawn, the baby you're raising, how did you keep track of their emissions? Is there a baby monitor of some sort that you use?

[Laughs] Kind of! Our developer connected Spawn to our Slack chat group and we'd get these renderings in the middle of night from her, so it was something like a baby monitor. The rendering time is crazy. You set up all these training sets, and sometimes it would take a whole night for Spawn to train and render. It's different for me because I'm used to setting up these digital processes and improvising around it. Having to wait, and you wake up and have a new track, but it sucks and it's like you start over. There's tons of trial and error for a project like this.

A conductor would teach an ensemble choir what to sing. How do you teach singing to a machine?

There are a few techniques. One is called SampleRNN, and it's basically dealing with sound on a micro level. So you give Spawn a training canon, and if she learns one sample, she knows what comes after it. She then builds the sound up sample by sample, which gives a particularly granular way. The other way was done with Jlin, who collaborated on this with me. We created a voice model. You can train an AI on a specific voice or timbre, so we used mine for Spawn. We then gave her material to perform in that voice model.

The album seems to conceptually explore the life cycle of an AI entity. Justice for AI lifeforms?

Exactly. But I think what's misleading is the idea that I mean the birth to death cycle of Spawn or an AI. The last track is called "Last Gasp," and it's not Spawn's last gasp. It's the last gasp of a specific humanist world view.

Is that a suggestion that there's more life beyond our current form?

Exactly. Technology regenerates and lives on.

Suppose someone might want to raise an AI baby of their own. How might they do that?

The beauty of the software that we're using is that it's all open-source. You just have to be big enough of a nerd and have the patience to install the software and teach yourself how to play with it. It's not proprietary, so of course we're tweaking the software and the data sets. Many people working with AI right now are using SampleRNN as well. TensorFlow, too. These are programs that are pretty widely available. We're kind of downstream from many research institutions, and I think that's why we've seen so much [AI advancement] in the last few years. There were these papers released to the public in 2016 or so and everyone consumed them, which is great. That's what's wonderful about open-source software that everyone can tap into is it's totally customizable for whatever theories about the past, present, or future that you want to explore or make sense of.

The song "Eternal" challenges this human conception that life is finite. The project as a whole suggests other forms of life can continue through technological advancement. What do you say to people who fear the future?

I love "Eternal" because I almost see it as this classic vampire story where lovers have this neverending romance through the ages. It was inspired by this couple named Martine and Bina Rothblatt. Their story is amazing, but basically their daughter got sick and there was no cure for what she had. They bought a patent for this medicine and developed a cure. To me, it's the ultimate story of taking agency of one's own life, and it's about the courage to save the ones you love. That courage saved other people's lives. Now Martine is trying to create AI versions of herself and has created Bina [as Bina48], so they can be in digital form, in love together forever.

"We owe it ourselves to look for how technology can serve us rather than work against us."

That message is important because it defies fear of the future. There are those who think we are all becoming robots and those embracing what's possible.

There are real things to fear with AI, too. It can be applied in a dubious and fucked-up way by people with bad intentions. But we have to create and learn to embrace our own fantasy, or else we're relinquishing our agency to those who do have a darker vision of the future. It's better than being scared all the time. There's plenty to be scared about, but there's also plenty to be excited about. We owe it ourselves to look for how technology can serve us rather than work against us.

This hopeful vision also feels necessary to grasp, considering our current political climate.

Absolutely. The last few years have been alienating to everyone. No matter what's going on, there just isn't a shared sense of objective truth anymore it seems. That feels scary. It's as if we have no starting point as a society. Why are we still not in agreement that the earth is round? Things feel in flux in that way and our project responds to that on some level. It's also dissecting how AI can be used in this corporate way to remove human labor and effort in the interest of something more perfected. It felt important and inherently political to create a project where any of us humans could opt in, be audible, share ideas, discuss concepts online and face to face, be compensated, et cetera.

When recording the album, what were some of the challenges of making the music itself?

I'd start out by giving the ensemble a written score I'd print out. We'd read it and sometimes I'd record them, go into the studio, process their vocals. Sometimes I'd play them back the processing and they'd emulate that, then Spawn would have her version. There was a ton of back and forth, but I think the biggest challenge was making all the environments make sense within the same record. The live training sessions were all recorded in live environments, sometimes in huge echo chambers of rooms. So even things like reverb on this album, it's not digital. That's all from the natural sound of the room. When Spawn would send a file, she'd sound raspy within the nascent technology. Having her rough sound with the digital instruments, we were treating everything as if it came from the same space.

"It felt important to create a world where any of us humans could opt in, be audible, share ideas, discuss concepts online and face to face, be compensated."

What's a day to day of Spawn's life like? For a human baby, there's feeding, dressing, sleeping.

It depends. We're working on a real-time performance system so she's just chugging audio files. In between these feeding sessions, she's just sleeping. I guess we're kind of strict parents in that way. [Laughs]

She doesn't get out much?

No. [Laughs] We're not taking her for walks or anything. But if she does a good job with this real-time performance system, she'll get to come on tour as her reward. That's our hope anyway. I don't want it to be like a Wizard of Oz situation, where you pull back the curtain and nothing's real or nothing works, but we'll see how things go. There's beauty in the imperfection.

Photography: Boris Camaca