“Even from the amount of words that you’ve emitted in this conversation, I could create a voice model of your voice,” Holly Herndon explains calmly from across the hotel room, a mischievous glint appearing in her ice-blue eyes. “I could capture your voice, and then I could make it speak whatever I wanted it to speak. So what does that mean for the sovereignty of your personhood?”
Just ten minutes into our meeting and I’m seriously contemplating the moral implications of A.I., not to mention the practicalities of never uttering another word aloud ever again. If the aforementioned scenario sounds ludicrously dystopian, it’s chastening to realise that these are the capabilities – and very real ethical concerns – of the technologies that Herndon has dedicated the best part of a decade to exploring.
In the process of completing her PHD at Stanford, the Tennessee-raised, San Francisco-based sound artist has produced three LPs that, in part, document her research. First came 2012’s RVNG INTL-released Movement, a feast of laptop-based experimentation that first established Herndon’s fascination with vocal manipulation. Platform followed in 2015 on 4AD, developing those ideas further both practically and thematically via experimentation in the field of autonomous sensory meridian response (ASMR) and – what subsequently proved to be – an eerily prophetic critique on surveillance culture. These themes continue to underpin latest LP, PROTO, which structurally is Herndon’s boldest experiment yet.
“Somebody has to take a risk and come up with something new in order for music to progress, otherwise we’re stuck in the now forever.”
At the heart of the record is an “A.I. baby” named Spawn, which Herndon and her partner Mathew Dryhurst built from “a souped-up gaming PC running some machine-learning algorithms.” Spawn works by mimicking sounds from the specific training sets she’s exposed to by Herndon, Dryhurst and their collaborator Jules Le Place, be it recordings of individual voices or choirs, or any other sound design. The end result is what Herndon describes as “an ensemble member who can interpret and perform,” drawing on very specific data sources.
“Most machine-learning systems are drawing on a vast pool of data,” Herndon explains of the difference. “So when Facebook and Google are doing machine-learning, they’re hoovering up all this freely available information that we’re constantly giving them, and then creating models on these oceans of data. We wanted to hone in and make the human labour really visible – which is something often made opaque by more corporate A.I. structures – to see if [the music] had a different aesthetic quality to it.
“And the reason we used the child metaphor is that we wanted to think about it with the gravity and weight of raising a child; seeing ourselves as involved in that process and having a voice in it. Because this nascent intelligence doesn’t have context beyond what we’re feeding her, kind of like a child being raised by a parent or a community.”
Spawn appears throughout PROTO, interwoven with – and often indistinguishable from – the array of human voices that Herndon assembled at her second working base in Berlin. On the sublime, synth-driven symphony ‘Eternal’, there’s an angelic quality to the intricately-layered vocals. On Jlin-collaboration ‘Godmother’, Spawn’s garbled, guttural performance is borderline nightmarish. Herndon laughs when I mention that Dryhurst described the latter effect as “ugly”.
“At first I was a little embarrassed, because it’s kind of like Spawn is beatboxing. I trained her on speech – reading 10-minute scripts and also singing phrases to bring in different timbral variety. Of course it makes logical sense that, when presented with percussion music, she would combine those things to sound like beatboxing. But ugly’s the wrong word. Maybe more rough, and honest, rather than this sci-fi, glitzy sheen. And that’s one of the reasons why we wanted to use sound as material specifically: to show that this is what audio sounds like from a neural network.”
As Herndon explains, this raw quality is at odds with the dominant narrative in A.I. music. “Most people take music, abstract that into a score that has pitch and rhythm material in the form of MIDI data, then train the A.I. on that so that the A.I. can create infinite works in the style of that canon. So we’ll see a lot of generic, Hans Zimmer-like film scores being created, a lot of muzak, a lot of passive listening music. Certainly, the economic incentive is there for people to take a more conservative approach.
“This is really problematic for many reasons. Firstly, you’re drawing on a shared musical canon and recreating something that’s being monetised for the benefit of one person. And second, when we are using the existing canon as training sets it gets us in a recursive loop aesthetically. Somebody has to take a risk and come up with something new in order for music to progress, otherwise we’re stuck in the now forever.”
Taking inspiration from the musique concrète movement, Herndon’s creative interests lie in sculpting sound to create new aesthetics entirely. “What can music reproduced by an inhuman intelligence reveal about the structure of the sound itself?,” she gushes excitedly. And yet for all her and her collaborators’ enthusiasm for the project, Herndon readily acknowledges the ethical issues and potential pitfalls in the future of A.I. development.
“Certainly there’s room to be critical, because we don’t have a great history of shared resources,” she laughs wryly. “I mean, look at the way the Internet’s gone. We were just like, ‘Information wants to be free!’ and now we’re in a giant shopping mall, and advertising logic is dominating my interpersonal relationships. That’s fucked up! And so I think we should not fuck it up this time, you know?
“I don’t wanna be hyperbolic, but I do think it’s fundamentally groundbreaking technology that we have. And it’s still early enough that we can make some decisions as a community as to where we stand on it and how we can use it in a way that’s symbiotic and beneficial to everyone. If we can encode our values and ethics into the protocol layer from the get-go, then maybe we can foresee some damage down the line. I mean, I think you have to have some optimism if you want to feel agency in the world, otherwise you’re giving up and allowing someone else to determine the future.”
With PROTO, Herndon has begun defining this future undaunted, assembling a like-minded team of “nerds” to assist her. And of all the mental somersaults required to create this third album, which aspect does Herndon cite as the most challenging? “The humans,” she laughs without hesitation. “They all have their own lives, their own schedules, their own ideas… They’re complicated.”