Image via Filip Preis
Michael McKinney understands the cultural importance of Kreayshawn’s “Gucci Gucci.”
Lee Gamble isn’t interested in the future. Electronic music, which the British producer has been neck-deep in for decades, is filled with ideas about alternative timelines and far-flung utopias. Techno and electro are built, in part, upon disassembling umpteen existing cultural motifs and reassembling them into gleaming, neon-coated monoliths: relics of the past transformed into barely recognizable transmissions from a better timeline. Again and again, club nights gesture towards times that have yet to happen by reaching backwards, melting our cultural detritus into something new along the way. But the future is, by its very nature, imprecise. What does it mean, and to whom? Which levers of power will move, and when, and how?
Instead, he’s focused on what sounds like now. This, in retrospect, comes as little surprise. When he was a child in the early ‘80s, his dad would blast The Police and Japan on the stereo, and his mom’s (Irish) side of the family would sing together around the piano. When he first heard Inner City‘s Paradise, he was at a fairground with his cousins. This is all material rooted firmly in a time and place: it might have been made with forward-thinking technologies or approaches, but people always came first. When he was 16, he fell into production, sampling breaks from records and piecing them together with a cousin in Birmingham.
For a time, he fell out of love with jungle and techno. He started listening to, and studying, Stockhausen, Xenakis, and Morton Feldman. He started making what he calls “experimental and multichannel computer music.” After moving to London at the turn of the century, he enrolled at the Lansdown Centre for Electronic Arts, where he studied Sonic Arts. There were no keyboards allowed in the department, and if he wanted to interface with something, he was encouraged to build it himself. That spirit carries into his work to this day: “If everything gets switched off tomorrow,” he said over a recent Zoom call, “Just give me a stick, a microphone, and a hammer. I’ll make some music with that.” Slowly, though, club culture was starting to come back into his work. He’d moved to London as dubstep and grime were rocketing in popularity and quality, and he began “inject[ing] computer music sounds with strings and pads that sound[ed] like old jungle” records.
In 2012, after years of sound art and electroacoustic work, Gamble released Diversions 1994-1996. The record sits at a relatively brief twenty-odd minutes, but it carries years of baggage. It is his reflection on “a time in my life that’s gone.” The album’s genesis was a box of cassette tapes and memorabilia returned to him by his father: jungle tapes from the years when a young Gamble was DJing with his cousins. The resultant LP sounds like classic jungle records misremembered. It takes the crackle and hiss of warped vinyl and pushes them to the forefront, every drum break nearly muted and buried under a mountain of dust. At the time, it read as a culmination of his years of parties and training in sound art. In retrospect, it looks like the start of a new chapter. He put it simply during our call: “I’m more likely to be interested in the trace of something than I am in the thing itself.” Since then, Gamble has prized feeling over form. Maybe that was always the case.
In 2020, the world shut down. Suddenly, he had a lot more time than he’d planned for. He started painting, slowed down a bit, and thought a lot about neural networks and AI. He read Kit Mackintosh’s Neon Screams, a book that explores the history of the processed voice in pop and hip-hop; he dug into Agnès Gayraud’s Dialectic of Pop, which traces the disembodiment that is inherent to the recorded voice. He listened to K Allado-McDowell, a writer and speaker whose work looks at AI in a “spiritual” way. Eventually, though, he had to put the ghost back into the machine. “I wanted to smell the f*cking petrichor rather than the inside of a Mac,” he said, laughing.
This oscillation between the digital and tangible worlds became the genesis for Models, his latest album. The name recalls large language models, but also simulations, stencils and silhouettes; it sounds like a photonegative of a pop record that doesn’t exist. In the recording process, he treated a neural network like a songwriting companion, conjuring a choir of voices without vocal chords. He fed a computer’s output into itself until any specifics were smudged away, replaced by digital detritus and half-remembered earworms. Models is the sound of Gamble reacting to a technology that is evolving faster than any one person can track.
The result is both remarkable and of a piece with the rest of Gamble’s work. It doesn’t sound like the “future,” whatever that means. It sounds like the present: beautiful and a bit eerie, informed by budding technologies that have a million unknowable paths ahead of them—terrifying, thrilling, rapturous, and unsettling at once. The club nights Gamble went to as a kid, filled with the slamming rhythms of Birmingham techno and jungle, took pre-existing building blocks and constructed a universe out of spare parts. Decades later, with Models, Gamble has done the opposite. Models takes hypermodern songwriting techniques and makes them sound ancient. It is the sound of data sets and moss-encrusted iPhones, of “Believe” and dancefloor utopianism, of streaming-era sonic dissociation and timeline collapse. What is there but the present?
In advance of the release of Models, we caught up with Lee Gamble. We explored how painting informed his latest LP, the club nights he threw in the 2000s, the ramifications of “artificial intelligence,” the slipperiness of the future, and lots more.
(This interview has been condensed and lightly edited for clarity.)
What’s some art that connected with you early on?
Lee Gamble: The first thing that I remember really hitting me was early hip-hop—electro tapes. I was really young; there were some older lads around where I lived and they were b-boy types. They gave me this tape, and I’ve still got it! I remember my dad putting it in the cassette player, and I went, “What’s this?!” There was hip-hop in the charts, but my dad wasn’t playing it; he was playing rock, and Kate Bush, and that kind of stuff.
Jungle came later, when I was a teenager. Techno, too: there was an amazing techno club in Birmingham called Que. That exposed me to that heavy Birmingham techno sound. Jungle was simultaneous. Sometimes you’d go and hear Regis, Jeff Mills, Surgeon, Claude Young—that third-wave Detroit generation sound, which I still adore. Then you’d go watch Fabio & Grooverider at The Institute. That was when I was quite young. That was, of course, the time where you might get into bands, but there was no way you were dragging me into an old smelly pub when I could go into a rave with lasers and fog and this incredible futuristic machine-music. It was a no-brainer where I was going to end up! [laughs]
I’d later branch off to jazz and things like that. I’d find who Photek sampled, and then I’d find the record: “This Pharoah Sanders record’s sick.” I’d go to the library and borrow jazz records. It all starts from that tape, really. That’s my rose-tinted version of events. That tape stuck out—it opened up a portal.
How’d you find your way towards pirate radio?
Lee Gamble: My dad had one turntable, and I nicked it and went upstairs. Then I got another: a belt-drive sort of thing. I played around with that, and I’d buy records. I was 14 or 15 [at this point]. I had some older cousins who were 6-10 years older than me, so they were already into acid house. I’m from a big Birmingham / Irish family, so there’d be a lot of people around, and people would just play tapes, so I was exposed to rave music pretty early. There’d be two people in the whole area who might have a set of turntables, so me and one of my cousins would go there and hang out, playing records and pooling our records together: I’d have four, and my cousin might have twelve. So you might be able to scratch a mixtape together.
Eventually, I got a turntable and a belt drive, and then another. Finally, you get a cassette deck, and you’ve got 20 records. All of a sudden, you can make a 60-minute cassette. I was buying records in Birmingham and Coventry. They’d eventually say, “Give us a tape,” and they’d give you a slot early in the night. I was still a teenager at this point. There was a pirate [station] in Birmingham. I didn’t play much, but I played a few times. They’d move around; they’d get shut down pretty quick. That was through my cousins; I mostly tagged along. I hopped in the car, brought my records, played, and went home.
In the late ‘90s, I got kind of bored of jungle and techno. It started to feel a bit stiff and less experimental. I started to dive into more experimental music. I studied art. I started listening to Stockhausen and the like every day. I spent a lot of time studying that and working on computer music. I did multi-channel stuff, and I did gallery-leaning things and more conceptual stuff: noise music, improvisations and things like that.
In the early 2000s, I was in London, and when grime and dubstep happened, I said, “Wow—what’s that?” That’s when it grabbed me again. That’s the short answer; it’s 10-12 years of your life! Once I dipped out of dance music, so much opened up: the history of electronic music, musique concrète, sound poetry and German electronische musik. Up until I made Diversions, I was in that space a lot. In university, I studied a sound and theory art degree. We’d still go raving sometimes, but I was burrowing deep into experimental music and the arts.
Who’s “we”?
Lee Gamble: I moved to London in 2000. It was the people I was at uni with, and people around in the scene. Not many of us had dance-music backgrounds; many of us were into dance music and something else, too. I threw a series of nights in London after I finished uni, because there were a bunch of us saying, “I really like musique concrète. But dance music is amazing, too.” At the time, they never seemed to be close together. So we put on a series of events in a small venue: we’d do a £3 cover, and we’d get people to do an experimental improv thing. We’d go on after, and we’d DJ, like, A Guy Called Gerald, Pan Sonic, and Jeff Mills, but we’d clash it on top of Merzbow. It was, like, “F*ck it—I see value in both of these things. I don’t see why they should feel separate.” We did about ten of these, and they got quite mental.
After the last few, we got banned from the venue. For one of these, we had someone playing computer music, and he did an unbearably loud set—I wouldn’t call it “drone,” because that sounds kind of relaxing. This was ear-piercing: totally hallucinogenic and alien Pretty much everyone left, and the people we were renting from said, “What’s he doing? You’ve got to tell him to stop.” I said, “I’m not telling him to stop.” It was this horrifically loud sine-tone. People behind the bar hid until it went away. When that stopped, we’d start playing, like, pounding techno or something. For £3, you can have quite a nice Tuesday night. [laughs]
When did you start painting?
Lee Gamble: I drew all the time when I was a kid. I started painting properly a couple of years ago. Before that, I never really had the space—I still don’t have enough space. I want to make bigger pieces! I lost a few people who were really close to me, and I had a hole to fill. And a painter friend of mine asked me if I painted, and I said, “Not really, but I drew all my life.” He told me he’d love to see what I’d paint. He very kindly gave me a canvas, and some paints, and told me to do it. I love it. I’ve been spending a lot of time on that in the past year.
After I’d finished Flush Real Pharynx for Hyperdub, I felt lost. Between the pandemic and losing these people, I said, “I’m gonna get grounded again.” Painting made me realize that I really like yellow! I didn’t know that. There’s things like that, where you say, “I like that color; I don’t like that color.” You can really transfer that. Not technically, but in terms of saying, “I don’t give a shit about all that—what does this sound like in my monitors? Do I like this?” I’d allow that to happen, and I wouldn’t question it any further. That’s how Models came about, I suppose.
I said, “I want to make a record that I really enjoy making.” It’s not like I’ve not done that before, but Flush Real Pharynx was such a big thing. It was really hard to do. I spoke to Steve [Goodman, AKA Kode9] at Hyperdub, and I said, “I’m making something, and it’s going to be really simple. I’m gonna do what I want; it’s a super emotional record.”
That seems really simple, but I’d lost my ability to do that. I’d lost the ability to judge, or believe, my own taste a little. There’s so many trends, sounds, and styles—things move so fast. And, all of a sudden, you’re like, “Where the f*ck have I gone?” The painting really helped with this. It helped me focus on what I wanted to do. Nobody’s gonna come up and say, “I think you should put some pink there.” It’s up to me; it starts and ends with my decisions. I approached Models in the same way.
You started this record in 2020, yeah?
Lee Gamble: I didn’t start the musical production then, but I did start to write the ideas for the record, and I started to correspond with Joseph, an AI programmer from America. I was aware of the possibilities that were available in AI. It’s moved a huge amount in three years. I was wary not to dive in too fast and get this rush of newness, write quickly and rely on the technology to hold the record up. We had back-and-forths where we were building vocal models and testing things out. I eventually told him, “I’m going to go away now; I need six months.” I needed to figure out how to make it my work rather than having it be the work of a piece of technology, which I have no interest in.
I worked with instrumentation, and I tried to figure out what would work with the voice and give it space. I was playing around with hooks. I wrote a couple of tracks without voice and added it in, but most of it was written around the vocals. I built it around earworms—phrases that I’d picked out from stuff that the network spat out.
Given I’m going down this wormhole, I’m aware that it’s problematic. I want to use this technology, but I want to be respectful of its problems and how it might sit in popular culture. So I wanted to take my time. I had more time than I’ve ever had to make a record. I read as much as I could, from as many different voices as I could, about how this would affect people: things like the racialized aspect of datasets, and the voice, too. It’s been mimicked a lot, and it’s been caricatured a lot, and it’s been taken and used by white culture, often in terms of American music.
I was conscious of these things, but it’s also such a tantalizing technology for an artist to not use. I’m still in a quandary with it. The fears around job replacement are totally valid; AI will definitely take jobs. It already has. I don’t think it’s going to take anyone’s singing away. I wanted to make sure this record didn’t feel like a technologized record. It’s about the idea that these aren’t real people, but it’s not suggesting the replacement of people. It’s about, “How much mood can I get out of this? How human can I make the technology?” How human does a singing voice feel when it’s disembodied? How emotional can a digital simulation feel? This is without ignoring what they are: simulations of people. I wanted to make sure I leaned into that, but I didn’t want it to be the thing that wrote the record.
I wanted to make sure I’m extremely conscious about it. The voices aren’t modeled after any particular person. It was, at first, but after a couple of years, I was like, “I can’t do this.” We built a system of four voices and kept feeding it into itself. Eventually, it got so far away from whatever it had been fed that it’s not anyone. The whole album, to me, feels like a trace of a pop record. I’d get phrases back and pick from them, and while I wouldn’t say I wrote these like pop songs, I allowed myself to drift into that space. This applies to the way it sounds, too—I wasn’t trying to make it sound crumbly, like Diversions, or super hi-res, like parts of Flush. With pop records that I love, even now, I’ll grab one track off one record, and I’ll hardly be able to play anything else. Everything else sounds terrible! I wanted to grab that energy: something like the aura of pop.
That’s what I was trying to think about. I wanted to make something that felt like you’re connecting to it in some way. The thing with these voices is that they don’t have a physical presence about them. I’m not trying to undermine the technology, of course. You feed them human stuff; all of AI is, fundamentally, human. But, in terms of their presence, they seem like they pop out of nowhere.
The first thing I did was build a model of my own voice, which is actually on the record, on A Million Pieces of You. I’m on there—well, it’s not me; it’s my deepfake. It’s taken a few years to wrangle with it as a cultural thing, which was as much the problem as making the record was. I wanted to make something that sounded like it was human, because, fundamentally, that’s what they are. They might come out of these incredible pieces of technology, but they’re human things. They’re a very weird entity. I don’t know what you’d call them. You have this idea of the uncanny valley, but it’s not that. It’s not like it’s trying to impersonate somebody; they seem to have their own character.
With this project, I kept returning to Autechre. You’re saying, “How can we take these ideas and put humanity back in?” I would say this and one of Autechre’s long-form live things are doing a similar thing.
Lee Gamble: When I hear Autechre’s current music—especially their live shows—it sounds like they’ve built a system that sounds like them. I think that’s a nice way of hearing their stuff. It’s a system that’s been honed over many years to sound that way and exist as that system. Models is not doing that in such a singular way, but the voice modeling aspect was definitely a similar kind of technological conversation. It doesn’t feel like a conversation when I’m playing a keyboard, although it is a piece of technology. I felt differently with this, because I wasn’t sure what I’d get back. It’s a matter of making a choice: do I like that or not? Do I put yellow on the canvas, or red, or something else?
That’s where the album emerged from. I’d run a plugin to find out what key those voices were in. I’d set everything around that key and write around them. The technology was giving me a lot of input, in a distant way. It’s not like I’m tweaking anything live: it’s a matter of taking stuff, and, an hour later, going “Wow, okay—you thought that? Interesting.” That’s where the magic was: saying, “Whoa! You reckon I should write something like that?” [laughs] There’s definitely that generative aspect to it. If we gave it reverb, it would turn the sounds into mush, and we’d have to start cleaning it up. I was literally mixing and mastering the sound of datasets.
I have a feeling that, with voice simulation technology, the endgame will be when you can’t tell the difference between AI voices and human voices. I don’t find that particularly interesting. The fact that they were scratchy and indecipherable and smudged, and had an Elizabeth Fraser, or “mumble” rap-esque non-language, where you don’t know what they’re saying but you’re getting emotion and narrative from it anyways—I was keen on exploring that. Some of the weird, fragmented, “mumble” rap stuff is huge, even though it’s really experimental. Once I got my head around that, I was happy to work with them in the state they’re in. It took tons of work to get them sounding like that, but, eventually, I was happy that they sounded like that rather than saying, “This sounds exactly like a person singing.”
That was only possible because the technology’s not great yet. In a few years, you might only have the really good models. I think you lose something in that, so I wanted to get it done now. In 10 years’ time, you might say, “Listen to how those voices sounded then! They were so bad!” The image things look completely different to how they did a year or two ago. I don’t know if the old ones are there anymore. It’s not like a piece of hardware that you can pick up again.
There’s an AI artist I found a few months ago who’s leaned into digital fragmentation and software errors. They took the bits that Midjourney doesn’t understand yet and said, “What do we do with the technology as it currently stands?” These aren’t flaws, necessarily.
Lee Gamble: That’s exactly what I did with Models. The crux, for me, was that detritus: this kind of childlike attempt to communicateThat kind of surplus value happens in technology all the time. But AI is so new, and it’s moving so fast. The first synthesized cello sound sounds terrible now. Now, if it’s mixed well, it’s hard to tell it’s fake. But that’s taken 50 years. AI’s moved at an incredible rate in the last three. The speed at which this technology evolved, in a few short years, is really scary. There’s no point in racing it, because it’s going to beat you. This stuff is owned by all the big corporations—they’ve got billions of pounds to throw at this.
With this record, I said, “I know damn well that if, in two years’ time, if I say this voice modeling system sounds like a human, it’ll be trash.” So, forget that—just make a record with the technology that’s in front of you, and lean into the limitations. I can’t compete with these companies; I’m just a guy who knows a brilliant programmer. We don’t have a team behind us. Forget them—they’re doing their thing, and I’ll lean into the problems that they’re trying to fix.
It’s the idea of making something that sounds like a moment.
Lee Gamble: That’s something I’ve been thinking about recently. You’ve got this idea about the hardcore continuum, for instance. It keeps shapeshifting and latching onto other things, and every now and again, you’ll get a burst, and you’ll get a new genre. I feel like that’s finished. I don’t think it’s not actually happening: I just think that since we’re so deep into social media and streaming, access to things has been leveled.
I can be on Spotify, and I could be in a playlist with Beyoncé. That wouldn’t have happened before. If I was 10 years old now, I don’t have this reference to date or time: here’s jazz, here’s prog, here’s krautrock, here’s techno. It’s just all there. In a sense, that’s really amazing. I’ve really enjoyed Lil Yachty’s new record this year, and it seems to lean into that: there are points where it sounds like Pink Floyd, and then it’ll sound super modern, and then it sounds like the ’70s, and then it sounds like old-school computer games, and then it sounds like old hip-hop. It sounds like now.
I don’t think younger kids give a shit—and I don’t think they should—if something was made in 1970 or 2023. Who cares? If someone’s playing a track in a club, and you’ve got 2,000 people dancing to it, who cares if it’s from 1994? That’s what I mean about this idea of a continuum. It’s all been piledrived into a wall and splattered. It comes with its issues; I get that. But I think that time splattering is a good thing.
The question about what is, or sounds, “modern” or “old” gets mushier over time. If you’ve never heard something before, it’s new. I heard someone on NTS the other day—a working-class person from London—playing Steve Reich. They said, “I’m going down this wormhole with Steve Reich.” That’s amazing. That stuff is historically of the “academy,” or the “middle classes,” or the “conservatoire,” but there’s also a clear history from Indian music to minimalism. It all gets constantly re-contextualised in the streaming era.
I went down a YouTube wormhole with 100 gecs’ stuff. I’m not a stan, but it crossed me in the same way. It’s a mixture of things that sounds exactly like now. It doesn’t matter if I like it aesthetically. As a cultural object, you’ve taken this on and just ran with it! You don’t give a f*ck. I don’t know how old they are, but they’re younger than me. I think younger people aren’t trying to make a point of that; that’s just how it is. [laughs] There’s no reason to have this linear time where everything gets discarded—”Oh, it’s old.” I always thought that was a bit pointless.
Also, to flip towards the future: people are always like, “This sounds like the future.” What does that mean? I’m not trying to be facetious; I honestly don’t know what that means.
I’m definitely guilty of using that terminology.
Lee Gamble: You’re a writer, so it makes sense—it’s a good semantic thing. People go “Ah! I get it.” But what do you think the person’s gonna think it sounds like?
I try to use it to say, “This person is going in a really exciting direction, and maybe this is a possible vision for the future.”
Lee Gamble: That Lil Yachty record sounds like now. It might sound old in a year. But, for now, it sounds like the present. It’s that cross-pollination—corecore shit on TikTok, and mashups. I keep hearing the word “Dada” again. It feels like vertical collage. I did a piece recently which was performed at Berlin Atonal; they did a Xenakis X100 thing. I’m a huge admirer of Xenakis‘ work. I wrote this in memory of him, so I thought, “What would he be doing now?” You’re damn well sure he’s not doing what he was doing then. He was doing stochastic computer synthesis methods in the ’90s when very few people were doing that in the world. If he was a young person in 2023, I doubt he’d be writing for solo timpani. He’s a techno person, in a sense. I doubt he’d want to becalled a techno guy, but f*ck it, that’s fine.
It’s that idea of possible futures: what kind of future are you offering?
Lee Gamble: There’s definitely some of that aspect to Models. Take the Flush Real Pharynx series I did. That was me trying to come to terms with the culture clash we’re talking about: multiple screens at the same time. It’s a kind of collage. That wasn’t the defining thing for the record, but I had that in my head. I wanted that forefront in my brain while I was writing it. That’s why I called it Flush Real Pharynx: it’s about flushing your voice away and being bombarded with stuff.
With Models, I said, “F*ck that.” That’s really bad for our mental health. I’m gonna bounce, and I’m gonna make something more emotionally driven, or soulful, rather than a TikTok kind of thing. I’m not sure how long people can take that, just as a flesh-and-bone kind of thing. As brains, you and I are not that different to how we were 2,000 years ago. It’s having an effect on people’s mental health. As a culture, we try to come to terms with it. That’s what I did with Flush, and I injured myself by doing that, so I wanted to soothe myself a bit and switch it all off.
I’m not qualified to explain this, but in terms of how things would affect my music: previously it was cities. If you grew up in Detroit, you might be into one thing, and if you grew up in New York, you might be into something else, and if you grew up in Birmingham, it might be another. Now, it’s the internet. If you’re into something, you can find it. There are so many positive things about it, but at this point in my life, it’s better for my mental health for me to switch it off when I’m going to make something. It’s taken me a few years to learn that, but it’s fine. Sometimes things take a while to work out. If I were to make another record, I’d want to hire a studio in a barn somewhere on a mountain for three months.
You started Models when AI was prehistoric compared to where it is now, and when it wasn’t really a public conversation. What’s it been like to watch that evolution in real time?
Lee Gamble: Neural networks are from the ’50s, so AI is not new. But data happened. They didn’t have enough data sets to make a good neural network in the ’50s. Even when we started doing the voice modeling stuff, you’d need a bunch of stuff. Eventually, it’d get down to 60 seconds of something that you could work with. That was with two people and a normal computer; it’s not with Meta or Google or whatever. It was obvious to me that it was going to accelerate at rate of knots.
It’s a little bit of brain, in a sense, that can learn how to do something. It’s a simulation technology. Anything you experience is because you have a brain, fundamentally. If the brain is allowing everything to be knowable, then a neural network can try any of that. So, of course it’s going to be everywhere. We’ve seen it first in image-making, because most online communities are built around images. I knew it was going to be quick, so I had to say, “We are where we are, and let’s roll with that version.”
When I started writing [Models], I bowed out of the AI conversation. I started looking at it again recently, because I needed a break from it all. I read the other day about someone who had been in a coma for 20 years, and they lost their voice. They came out of the coma, and they’ve used AI to train them to have a voice again, which is such a Models thing. I think it’ll become ubiquitous. Then it’ll level off, and that’s when the interesting things will happen.
I wanted this to sound like a record that you’d make when AI isn’t a “thing” anymore. It’s a bit like early Herbie Hancock records. Everybody wanted to know what synthesizer he was using. 20 years later, nobody’s asking; he just plays. I wanted to place Models there somehow. I realize this is coming close to making “futuristic” music. [laughs] Right now, I’m overwhelmed by the shininess and potential of this technology. In years’ time, people won’t think of it like that.
When you put something like this out, you get people acting as though you’re okay with all that happening. The problem, for me, is that AI is already in every part of culture. So saying, “I’m not going to use it because it’s bad”—it’s not like not using Amazon because they’re scumbags. But we can’t not use AI. Fundamentally, it’s built on our collective labor, art, and time. I don’t want that to sound like a cop-out; I’m no libertarian. But, at a certain point, you’re going to have to engage with it. When that happens, it’ll either be unknowingly or with a bit of knowledge. The only way for me to offset that was to say, “Educate yourself on it, and hear some voices that aren’t of a similar background.”
I’m not here to make an excuse as to why I’d use it. I’m not out there saying, “This is amazing; we don’t need singers anymore.” I’m an artist, and I wouldn’t even begin to think that. If I can make something that can resonate emotionally with people, like I hope this record can, then maybe that’s okay. If I’m able to make people think about how they can use the technology in their way outside of any corporate dystopias, then maybe that’s okay. I’m not expecting to be able to make a record and say something about AI in general. There are things that we should be aware of and use as artists, and there are other things we should be wary of and at least talk about.
For example, there’s the protection of people’s voices. If I fed the first model one person’s voice and said, “Okay, that’s it,” that wouldn’t have been enough. I had to process it until I couldn’t tell who it was anymore; I had to eliminate the training from it somewhat. It’s not my voice. I understand there’s a complex relationship there. I had to be, and will continue to be, aware of that. People that look like me have claimed ownership of other people’s cultures and work, and that has to be taken seriously. If this can open a conversation about that, then that’s the least you can do if you’re using the technology. You can at least acknowledge that there’s a history of people taking people’s voices, mimicking them, and selling them because they’re a different complexion to the people who invented the style. I don’t think that any of this sounds like any particular person. I hope that it doesn’t, because that’s something I worked really hard to avoid.
It would be so easy to do that: you’ve seen the AI deepfake voices. I think the voice and the image in particular are very hot things to start playing around with. The image of another person that isn’t you is a very sensitive thing to mess with. What else have you got to identify yourself with other than the way you feel, look, and sound? That’s the most intimate thing, and to go, “I could do that now, because I have this Google sheet I can click on?” We have to think of it more respectfully than that.
I’m sure some people will disagree with that, but I’ve not heard anything yet when I’ve spoken about the record. When I was making it, I thought, “Maybe people will be really against this.” But I hope that, when people hear the record, they hear that that’s not where it’s coming from. It doesn’t feel like it’s coming from a position of ownership to me. It’s a dialogue. It’s: “Here’s where we’re going, and I’m trying to make something beautiful and touching with it.” That’s all I could do. I didn’t want to make something “icy,” or “fresh,” or “modern.” “Alien” or “other-dimensional,” sure, but I wanted to make it as warm as I could. I’m not too saccharine, but there’s definitely a few moments on there where I’m like, “F*cking hell.” The inner Cure fan came out. [laughs] I hope that people don’t think of it as an ode to a dystopian technology.
It’s a reckoning.
Lee Gamble: Yeah. It’s here. Try and own it: make it your work. It’s just some paint. It’s just some acrylic in a tube. Until you do anything with it, it doesn’t mean anything. There’s this incredible number of images made by image generation systems. Hito Steyerl is really interesting on this topic in terms of images. She talks about the image in one of her pieces of writing, In Defense of the Poor Image. It’s a piece of writing about the digital image. It goes through all sorts of different processes; it’ll get compressed again and again, and the colors change, and it all falls apart. So you get these poor images even though we’re in these hi-tech times.
This writing goes into a defense of the poor image, and it explores what qualities can be found in it. In the piece of writing I did, I titled one of my chapters—where I talk about the ‘crappy’ nature of the voices I used—”In Defense of the Poor Voice.” I quote Steyerl there, of course. In her writing, she says, “The poor image embodies the afterlife of many former masterpieces of cinema and video art.” I read her talk about it very eloquently, and I said, “I get that.” If I just swap all of that for the voice rather than the .JPG, then it’s the same thing. I’m seeing the some of the same advantages in the lack of quality as she’d seen. That really resonated with me.
What’s next for you?
Lee Gamble: It’s mainly the live show. I’m working with Candela Capítan, a performance artist and choreographer. I looked at her work and thought it was amazing, in terms of how it relates to the body, repetition, alienation, and the gaze and technologized image. I wanted to think about how to present Models live, and I didn’t want to be onstage with a laptop. From day one, I wanted to bring a physical body back into space—but not mine. Humans gave me the information; I simulated and disembodied them; and, now, I ought to put them back in. A lot of the movement is informed by the ideas of cloning, mirrors, and how people replicate each others’ movements. We tried to get twins to perform in the live show.
It’ll premier at Unsound in Poland, and then in London. I wanted to develop the choreography around the idea of how things learn—sometimes they learn badly. Obviously, Candela has brought a huge amount to the work. I gave them a PDF of ideas, and they read it; I played the music to them, and they improvised. It was very moving: I’d never seen anyone do that to my work before. So, I thought, “I want to get on this.”
I’m DJing as regularly as I can, just banging it out and enjoying myself. I’m really enjoying painting. I’m hoping to do some installation work with these voices as well. I don’t know if it’s just me, but I feel like my life moves in ten-year cycles. I did therapy last year for the first time in my life, and that came up. It was really helpful. I was really glad that people pushed me to do it. I’m not against it, of course, but I come from a background where that wasn’t a thing. It took me a long time, but I needed it, and it was super helpful. One of the things that came up is the idea of ten-year cycles in my life. Diversions was 2012, and that was a moment where I did a bunch of other stuff. Now, it’s 2023, and there’s Models, and I’m painting again. Now, I’m ready to boot them glass ceilings as hard as I can for the next ten years.
Jungle came later, when I was a teenager. Techno, too: there was an amazing techno club in Birmingham called Que. That exposed me to that heavy Birmingham techno sound. Jungle was simultaneous. Sometimes you’d go and hear Regis, Jeff Mills, Surgeon, Claude Young—that third-wave Detroit generation sound, which I still adore. Then you’d go watch Fabio & Grooverider at The Institute. That was when I was quite young. That was, of course, the time where you might get into bands, but there was no way you were dragging me into an old smelly pub when I could go into a rave with lasers and fog and this incredible futuristic machine-music. It was a no-brainer where I was going to end up! [laughs]
I’d later branch off to jazz and things like that. I’d find who Photek sampled, and then I’d find the record: “This Pharoah Sanders record’s sick.” I’d go to the library and borrow jazz records. It all starts from that tape, really. That’s my rose-tinted version of events. That tape stuck out—it opened up a portal.
How’d you find your way towards pirate radio?
Lee Gamble: My dad had one turntable, and I nicked it and went upstairs. Then I got another: a belt-drive sort of thing. I played around with that, and I’d buy records. I was 14 or 15 [at this point]. I had some older cousins who were 6-10 years older than me, so they were already into acid house. I’m from a big Birmingham / Irish family, so there’d be a lot of people around, and people would just play tapes, so I was exposed to rave music pretty early. There’d be two people in the whole area who might have a set of turntables, so me and one of my cousins would go there and hang out, playing records and pooling our records together: I’d have four, and my cousin might have twelve. So you might be able to scratch a mixtape together.
Eventually, I got a turntable and a belt drive, and then another. Finally, you get a cassette deck, and you’ve got 20 records. All of a sudden, you can make a 60-minute cassette. I was buying records in Birmingham and Coventry. They’d eventually say, “Give us a tape,” and they’d give you a slot early in the night. I was still a teenager at this point. There was a pirate [station] in Birmingham. I didn’t play much, but I played a few times. They’d move around; they’d get shut down pretty quick. That was through my cousins; I mostly tagged along. I hopped in the car, brought my records, played, and went home.
In the late ‘90s, I got kind of bored of jungle and techno. It started to feel a bit stiff and less experimental. I started to dive into more experimental music. I studied art. I started listening to Stockhausen and the like every day. I spent a lot of time studying that and working on computer music. I did multi-channel stuff, and I did gallery-leaning things and more conceptual stuff: noise music, improvisations and things like that.
In the early 2000s, I was in London, and when grime and dubstep happened, I said, “Wow—what’s that?” That’s when it grabbed me again. That’s the short answer; it’s 10-12 years of your life! Once I dipped out of dance music, so much opened up: the history of electronic music, musique concrète, sound poetry and German electronische musik. Up until I made Diversions, I was in that space a lot. In university, I studied a sound and theory art degree. We’d still go raving sometimes, but I was burrowing deep into experimental music and the arts.
Who’s “we”?
Lee Gamble: I moved to London in 2000. It was the people I was at uni with, and people around in the scene. Not many of us had dance-music backgrounds; many of us were into dance music and something else, too. I threw a series of nights in London after I finished uni, because there were a bunch of us saying, “I really like musique concrète. But dance music is amazing, too.” At the time, they never seemed to be close together. So we put on a series of events in a small venue: we’d do a £3 cover, and we’d get people to do an experimental improv thing. We’d go on after, and we’d DJ, like, A Guy Called Gerald, Pan Sonic, and Jeff Mills, but we’d clash it on top of Merzbow. It was, like, “F*ck it—I see value in both of these things. I don’t see why they should feel separate.” We did about ten of these, and they got quite mental.
After the last few, we got banned from the venue. For one of these, we had someone playing computer music, and he did an unbearably loud set—I wouldn’t call it “drone,” because that sounds kind of relaxing. This was ear-piercing: totally hallucinogenic and alien Pretty much everyone left, and the people we were renting from said, “What’s he doing? You’ve got to tell him to stop.” I said, “I’m not telling him to stop.” It was this horrifically loud sine-tone. People behind the bar hid until it went away. When that stopped, we’d start playing, like, pounding techno or something. For £3, you can have quite a nice Tuesday night. [laughs]
When did you start painting?
Lee Gamble: I drew all the time when I was a kid. I started painting properly a couple of years ago. Before that, I never really had the space—I still don’t have enough space. I want to make bigger pieces! I lost a few people who were really close to me, and I had a hole to fill. And a painter friend of mine asked me if I painted, and I said, “Not really, but I drew all my life.” He told me he’d love to see what I’d paint. He very kindly gave me a canvas, and some paints, and told me to do it. I love it. I’ve been spending a lot of time on that in the past year.
After I’d finished Flush Real Pharynx for Hyperdub, I felt lost. Between the pandemic and losing these people, I said, “I’m gonna get grounded again.” Painting made me realize that I really like yellow! I didn’t know that. There’s things like that, where you say, “I like that color; I don’t like that color.” You can really transfer that. Not technically, but in terms of saying, “I don’t give a shit about all that—what does this sound like in my monitors? Do I like this?” I’d allow that to happen, and I wouldn’t question it any further. That’s how Models came about, I suppose.
I said, “I want to make a record that I really enjoy making.” It’s not like I’ve not done that before, but Flush Real Pharynx was such a big thing. It was really hard to do. I spoke to Steve [Goodman, AKA Kode9] at Hyperdub, and I said, “I’m making something, and it’s going to be really simple. I’m gonna do what I want; it’s a super emotional record.”
That seems really simple, but I’d lost my ability to do that. I’d lost the ability to judge, or believe, my own taste a little. There’s so many trends, sounds, and styles—things move so fast. And, all of a sudden, you’re like, “Where the f*ck have I gone?” The painting really helped with this. It helped me focus on what I wanted to do. Nobody’s gonna come up and say, “I think you should put some pink there.” It’s up to me; it starts and ends with my decisions. I approached Models in the same way.
You started this record in 2020, yeah?
Lee Gamble: I didn’t start the musical production then, but I did start to write the ideas for the record, and I started to correspond with Joseph, an AI programmer from America. I was aware of the possibilities that were available in AI. It’s moved a huge amount in three years. I was wary not to dive in too fast and get this rush of newness, write quickly and rely on the technology to hold the record up. We had back-and-forths where we were building vocal models and testing things out. I eventually told him, “I’m going to go away now; I need six months.” I needed to figure out how to make it my work rather than having it be the work of a piece of technology, which I have no interest in.
I worked with instrumentation, and I tried to figure out what would work with the voice and give it space. I was playing around with hooks. I wrote a couple of tracks without voice and added it in, but most of it was written around the vocals. I built it around earworms—phrases that I’d picked out from stuff that the network spat out.
Given I’m going down this wormhole, I’m aware that it’s problematic. I want to use this technology, but I want to be respectful of its problems and how it might sit in popular culture. So I wanted to take my time. I had more time than I’ve ever had to make a record. I read as much as I could, from as many different voices as I could, about how this would affect people: things like the racialized aspect of datasets, and the voice, too. It’s been mimicked a lot, and it’s been caricatured a lot, and it’s been taken and used by white culture, often in terms of American music.
I was conscious of these things, but it’s also such a tantalizing technology for an artist to not use. I’m still in a quandary with it. The fears around job replacement are totally valid; AI will definitely take jobs. It already has. I don’t think it’s going to take anyone’s singing away. I wanted to make sure this record didn’t feel like a technologized record. It’s about the idea that these aren’t real people, but it’s not suggesting the replacement of people. It’s about, “How much mood can I get out of this? How human can I make the technology?” How human does a singing voice feel when it’s disembodied? How emotional can a digital simulation feel? This is without ignoring what they are: simulations of people. I wanted to make sure I leaned into that, but I didn’t want it to be the thing that wrote the record.
I wanted to make sure I’m extremely conscious about it. The voices aren’t modeled after any particular person. It was, at first, but after a couple of years, I was like, “I can’t do this.” We built a system of four voices and kept feeding it into itself. Eventually, it got so far away from whatever it had been fed that it’s not anyone. The whole album, to me, feels like a trace of a pop record. I’d get phrases back and pick from them, and while I wouldn’t say I wrote these like pop songs, I allowed myself to drift into that space. This applies to the way it sounds, too—I wasn’t trying to make it sound crumbly, like Diversions, or super hi-res, like parts of Flush. With pop records that I love, even now, I’ll grab one track off one record, and I’ll hardly be able to play anything else. Everything else sounds terrible! I wanted to grab that energy: something like the aura of pop.
That’s what I was trying to think about. I wanted to make something that felt like you’re connecting to it in some way. The thing with these voices is that they don’t have a physical presence about them. I’m not trying to undermine the technology, of course. You feed them human stuff; all of AI is, fundamentally, human. But, in terms of their presence, they seem like they pop out of nowhere.
The first thing I did was build a model of my own voice, which is actually on the record, on A Million Pieces of You. I’m on there—well, it’s not me; it’s my deepfake. It’s taken a few years to wrangle with it as a cultural thing, which was as much the problem as making the record was. I wanted to make something that sounded like it was human, because, fundamentally, that’s what they are. They might come out of these incredible pieces of technology, but they’re human things. They’re a very weird entity. I don’t know what you’d call them. You have this idea of the uncanny valley, but it’s not that. It’s not like it’s trying to impersonate somebody; they seem to have their own character.
With this project, I kept returning to Autechre. You’re saying, “How can we take these ideas and put humanity back in?” I would say this and one of Autechre’s long-form live things are doing a similar thing.
Lee Gamble: When I hear Autechre’s current music—especially their live shows—it sounds like they’ve built a system that sounds like them. I think that’s a nice way of hearing their stuff. It’s a system that’s been honed over many years to sound that way and exist as that system. Models is not doing that in such a singular way, but the voice modeling aspect was definitely a similar kind of technological conversation. It doesn’t feel like a conversation when I’m playing a keyboard, although it is a piece of technology. I felt differently with this, because I wasn’t sure what I’d get back. It’s a matter of making a choice: do I like that or not? Do I put yellow on the canvas, or red, or something else?
That’s where the album emerged from. I’d run a plugin to find out what key those voices were in. I’d set everything around that key and write around them. The technology was giving me a lot of input, in a distant way. It’s not like I’m tweaking anything live: it’s a matter of taking stuff, and, an hour later, going “Wow, okay—you thought that? Interesting.” That’s where the magic was: saying, “Whoa! You reckon I should write something like that?” [laughs] There’s definitely that generative aspect to it. If we gave it reverb, it would turn the sounds into mush, and we’d have to start cleaning it up. I was literally mixing and mastering the sound of datasets.
I have a feeling that, with voice simulation technology, the endgame will be when you can’t tell the difference between AI voices and human voices. I don’t find that particularly interesting. The fact that they were scratchy and indecipherable and smudged, and had an Elizabeth Fraser, or “mumble” rap-esque non-language, where you don’t know what they’re saying but you’re getting emotion and narrative from it anyways—I was keen on exploring that. Some of the weird, fragmented, “mumble” rap stuff is huge, even though it’s really experimental. Once I got my head around that, I was happy to work with them in the state they’re in. It took tons of work to get them sounding like that, but, eventually, I was happy that they sounded like that rather than saying, “This sounds exactly like a person singing.”
That was only possible because the technology’s not great yet. In a few years, you might only have the really good models. I think you lose something in that, so I wanted to get it done now. In 10 years’ time, you might say, “Listen to how those voices sounded then! They were so bad!” The image things look completely different to how they did a year or two ago. I don’t know if the old ones are there anymore. It’s not like a piece of hardware that you can pick up again.
There’s an AI artist I found a few months ago who’s leaned into digital fragmentation and software errors. They took the bits that Midjourney doesn’t understand yet and said, “What do we do with the technology as it currently stands?” These aren’t flaws, necessarily.
Lee Gamble: That’s exactly what I did with Models. The crux, for me, was that detritus: this kind of childlike attempt to communicateThat kind of surplus value happens in technology all the time. But AI is so new, and it’s moving so fast. The first synthesized cello sound sounds terrible now. Now, if it’s mixed well, it’s hard to tell it’s fake. But that’s taken 50 years. AI’s moved at an incredible rate in the last three. The speed at which this technology evolved, in a few short years, is really scary. There’s no point in racing it, because it’s going to beat you. This stuff is owned by all the big corporations—they’ve got billions of pounds to throw at this.
With this record, I said, “I know damn well that if, in two years’ time, if I say this voice modeling system sounds like a human, it’ll be trash.” So, forget that—just make a record with the technology that’s in front of you, and lean into the limitations. I can’t compete with these companies; I’m just a guy who knows a brilliant programmer. We don’t have a team behind us. Forget them—they’re doing their thing, and I’ll lean into the problems that they’re trying to fix.
It’s the idea of making something that sounds like a moment.
Lee Gamble: That’s something I’ve been thinking about recently. You’ve got this idea about the hardcore continuum, for instance. It keeps shapeshifting and latching onto other things, and every now and again, you’ll get a burst, and you’ll get a new genre. I feel like that’s finished. I don’t think it’s not actually happening: I just think that since we’re so deep into social media and streaming, access to things has been leveled.
I can be on Spotify, and I could be in a playlist with Beyoncé. That wouldn’t have happened before. If I was 10 years old now, I don’t have this reference to date or time: here’s jazz, here’s prog, here’s krautrock, here’s techno. It’s just all there. In a sense, that’s really amazing. I’ve really enjoyed Lil Yachty’s new record this year, and it seems to lean into that: there are points where it sounds like Pink Floyd, and then it’ll sound super modern, and then it sounds like the ’70s, and then it sounds like old-school computer games, and then it sounds like old hip-hop. It sounds like now.
I don’t think younger kids give a shit—and I don’t think they should—if something was made in 1970 or 2023. Who cares? If someone’s playing a track in a club, and you’ve got 2,000 people dancing to it, who cares if it’s from 1994? That’s what I mean about this idea of a continuum. It’s all been piledrived into a wall and splattered. It comes with its issues; I get that. But I think that time splattering is a good thing.
The question about what is, or sounds, “modern” or “old” gets mushier over time. If you’ve never heard something before, it’s new. I heard someone on NTS the other day—a working-class person from London—playing Steve Reich. They said, “I’m going down this wormhole with Steve Reich.” That’s amazing. That stuff is historically of the “academy,” or the “middle classes,” or the “conservatoire,” but there’s also a clear history from Indian music to minimalism. It all gets constantly re-contextualised in the streaming era.
I went down a YouTube wormhole with 100 gecs’ stuff. I’m not a stan, but it crossed me in the same way. It’s a mixture of things that sounds exactly like now. It doesn’t matter if I like it aesthetically. As a cultural object, you’ve taken this on and just ran with it! You don’t give a f*ck. I don’t know how old they are, but they’re younger than me. I think younger people aren’t trying to make a point of that; that’s just how it is. [laughs] There’s no reason to have this linear time where everything gets discarded—”Oh, it’s old.” I always thought that was a bit pointless.
Also, to flip towards the future: people are always like, “This sounds like the future.” What does that mean? I’m not trying to be facetious; I honestly don’t know what that means.
I’m definitely guilty of using that terminology.
Lee Gamble: You’re a writer, so it makes sense—it’s a good semantic thing. People go “Ah! I get it.” But what do you think the person’s gonna think it sounds like?
I try to use it to say, “This person is going in a really exciting direction, and maybe this is a possible vision for the future.”
Lee Gamble: That Lil Yachty record sounds like now. It might sound old in a year. But, for now, it sounds like the present. It’s that cross-pollination—corecore shit on TikTok, and mashups. I keep hearing the word “Dada” again. It feels like vertical collage. I did a piece recently which was performed at Berlin Atonal; they did a Xenakis X100 thing. I’m a huge admirer of Xenakis‘ work. I wrote this in memory of him, so I thought, “What would he be doing now?” You’re damn well sure he’s not doing what he was doing then. He was doing stochastic computer synthesis methods in the ’90s when very few people were doing that in the world. If he was a young person in 2023, I doubt he’d be writing for solo timpani. He’s a techno person, in a sense. I doubt he’d want to becalled a techno guy, but f*ck it, that’s fine.
It’s that idea of possible futures: what kind of future are you offering?
Lee Gamble: There’s definitely some of that aspect to Models. Take the Flush Real Pharynx series I did. That was me trying to come to terms with the culture clash we’re talking about: multiple screens at the same time. It’s a kind of collage. That wasn’t the defining thing for the record, but I had that in my head. I wanted that forefront in my brain while I was writing it. That’s why I called it Flush Real Pharynx: it’s about flushing your voice away and being bombarded with stuff.
With Models, I said, “F*ck that.” That’s really bad for our mental health. I’m gonna bounce, and I’m gonna make something more emotionally driven, or soulful, rather than a TikTok kind of thing. I’m not sure how long people can take that, just as a flesh-and-bone kind of thing. As brains, you and I are not that different to how we were 2,000 years ago. It’s having an effect on people’s mental health. As a culture, we try to come to terms with it. That’s what I did with Flush, and I injured myself by doing that, so I wanted to soothe myself a bit and switch it all off.
I’m not qualified to explain this, but in terms of how things would affect my music: previously it was cities. If you grew up in Detroit, you might be into one thing, and if you grew up in New York, you might be into something else, and if you grew up in Birmingham, it might be another. Now, it’s the internet. If you’re into something, you can find it. There are so many positive things about it, but at this point in my life, it’s better for my mental health for me to switch it off when I’m going to make something. It’s taken me a few years to learn that, but it’s fine. Sometimes things take a while to work out. If I were to make another record, I’d want to hire a studio in a barn somewhere on a mountain for three months.
You started Models when AI was prehistoric compared to where it is now, and when it wasn’t really a public conversation. What’s it been like to watch that evolution in real time?
Lee Gamble: Neural networks are from the ’50s, so AI is not new. But data happened. They didn’t have enough data sets to make a good neural network in the ’50s. Even when we started doing the voice modeling stuff, you’d need a bunch of stuff. Eventually, it’d get down to 60 seconds of something that you could work with. That was with two people and a normal computer; it’s not with Meta or Google or whatever. It was obvious to me that it was going to accelerate at rate of knots.
It’s a little bit of brain, in a sense, that can learn how to do something. It’s a simulation technology. Anything you experience is because you have a brain, fundamentally. If the brain is allowing everything to be knowable, then a neural network can try any of that. So, of course it’s going to be everywhere. We’ve seen it first in image-making, because most online communities are built around images. I knew it was going to be quick, so I had to say, “We are where we are, and let’s roll with that version.”
When I started writing [Models], I bowed out of the AI conversation. I started looking at it again recently, because I needed a break from it all. I read the other day about someone who had been in a coma for 20 years, and they lost their voice. They came out of the coma, and they’ve used AI to train them to have a voice again, which is such a Models thing. I think it’ll become ubiquitous. Then it’ll level off, and that’s when the interesting things will happen.
I wanted this to sound like a record that you’d make when AI isn’t a “thing” anymore. It’s a bit like early Herbie Hancock records. Everybody wanted to know what synthesizer he was using. 20 years later, nobody’s asking; he just plays. I wanted to place Models there somehow. I realize this is coming close to making “futuristic” music. [laughs] Right now, I’m overwhelmed by the shininess and potential of this technology. In years’ time, people won’t think of it like that.
When you put something like this out, you get people acting as though you’re okay with all that happening. The problem, for me, is that AI is already in every part of culture. So saying, “I’m not going to use it because it’s bad”—it’s not like not using Amazon because they’re scumbags. But we can’t not use AI. Fundamentally, it’s built on our collective labor, art, and time. I don’t want that to sound like a cop-out; I’m no libertarian. But, at a certain point, you’re going to have to engage with it. When that happens, it’ll either be unknowingly or with a bit of knowledge. The only way for me to offset that was to say, “Educate yourself on it, and hear some voices that aren’t of a similar background.”
I’m not here to make an excuse as to why I’d use it. I’m not out there saying, “This is amazing; we don’t need singers anymore.” I’m an artist, and I wouldn’t even begin to think that. If I can make something that can resonate emotionally with people, like I hope this record can, then maybe that’s okay. If I’m able to make people think about how they can use the technology in their way outside of any corporate dystopias, then maybe that’s okay. I’m not expecting to be able to make a record and say something about AI in general. There are things that we should be aware of and use as artists, and there are other things we should be wary of and at least talk about.
For example, there’s the protection of people’s voices. If I fed the first model one person’s voice and said, “Okay, that’s it,” that wouldn’t have been enough. I had to process it until I couldn’t tell who it was anymore; I had to eliminate the training from it somewhat. It’s not my voice. I understand there’s a complex relationship there. I had to be, and will continue to be, aware of that. People that look like me have claimed ownership of other people’s cultures and work, and that has to be taken seriously. If this can open a conversation about that, then that’s the least you can do if you’re using the technology. You can at least acknowledge that there’s a history of people taking people’s voices, mimicking them, and selling them because they’re a different complexion to the people who invented the style. I don’t think that any of this sounds like any particular person. I hope that it doesn’t, because that’s something I worked really hard to avoid.
It would be so easy to do that: you’ve seen the AI deepfake voices. I think the voice and the image in particular are very hot things to start playing around with. The image of another person that isn’t you is a very sensitive thing to mess with. What else have you got to identify yourself with other than the way you feel, look, and sound? That’s the most intimate thing, and to go, “I could do that now, because I have this Google sheet I can click on?” We have to think of it more respectfully than that.
I’m sure some people will disagree with that, but I’ve not heard anything yet when I’ve spoken about the record. When I was making it, I thought, “Maybe people will be really against this.” But I hope that, when people hear the record, they hear that that’s not where it’s coming from. It doesn’t feel like it’s coming from a position of ownership to me. It’s a dialogue. It’s: “Here’s where we’re going, and I’m trying to make something beautiful and touching with it.” That’s all I could do. I didn’t want to make something “icy,” or “fresh,” or “modern.” “Alien” or “other-dimensional,” sure, but I wanted to make it as warm as I could. I’m not too saccharine, but there’s definitely a few moments on there where I’m like, “F*cking hell.” The inner Cure fan came out. [laughs] I hope that people don’t think of it as an ode to a dystopian technology.
It’s a reckoning.
Lee Gamble: Yeah. It’s here. Try and own it: make it your work. It’s just some paint. It’s just some acrylic in a tube. Until you do anything with it, it doesn’t mean anything. There’s this incredible number of images made by image generation systems. Hito Steyerl is really interesting on this topic in terms of images. She talks about the image in one of her pieces of writing, In Defense of the Poor Image. It’s a piece of writing about the digital image. It goes through all sorts of different processes; it’ll get compressed again and again, and the colors change, and it all falls apart. So you get these poor images even though we’re in these hi-tech times.
This writing goes into a defense of the poor image, and it explores what qualities can be found in it. In the piece of writing I did, I titled one of my chapters—where I talk about the ‘crappy’ nature of the voices I used—”In Defense of the Poor Voice.” I quote Steyerl there, of course. In her writing, she says, “The poor image embodies the afterlife of many former masterpieces of cinema and video art.” I read her talk about it very eloquently, and I said, “I get that.” If I just swap all of that for the voice rather than the .JPG, then it’s the same thing. I’m seeing the some of the same advantages in the lack of quality as she’d seen. That really resonated with me.
What’s next for you?
Lee Gamble: It’s mainly the live show. I’m working with Candela Capítan, a performance artist and choreographer. I looked at her work and thought it was amazing, in terms of how it relates to the body, repetition, alienation, and the gaze and technologized image. I wanted to think about how to present Models live, and I didn’t want to be onstage with a laptop. From day one, I wanted to bring a physical body back into space—but not mine. Humans gave me the information; I simulated and disembodied them; and, now, I ought to put them back in. A lot of the movement is informed by the ideas of cloning, mirrors, and how people replicate each others’ movements. We tried to get twins to perform in the live show.
It’ll premier at Unsound in Poland, and then in London. I wanted to develop the choreography around the idea of how things learn—sometimes they learn badly. Obviously, Candela has brought a huge amount to the work. I gave them a PDF of ideas, and they read it; I played the music to them, and they improvised. It was very moving: I’d never seen anyone do that to my work before. So, I thought, “I want to get on this.”
I’m DJing as regularly as I can, just banging it out and enjoying myself. I’m really enjoying painting. I’m hoping to do some installation work with these voices as well. I don’t know if it’s just me, but I feel like my life moves in ten-year cycles. I did therapy last year for the first time in my life, and that came up. It was really helpful. I was really glad that people pushed me to do it. I’m not against it, of course, but I come from a background where that wasn’t a thing. It took me a long time, but I needed it, and it was super helpful. One of the things that came up is the idea of ten-year cycles in my life. Diversions was 2012, and that was a moment where I did a bunch of other stuff. Now, it’s 2023, and there’s Models, and I’m painting again. Now, I’m ready to boot them glass ceilings as hard as I can for the next ten years.
Eventually, I got a turntable and a belt drive, and then another. Finally, you get a cassette deck, and you’ve got 20 records. All of a sudden, you can make a 60-minute cassette. I was buying records in Birmingham and Coventry. They’d eventually say, “Give us a tape,” and they’d give you a slot early in the night. I was still a teenager at this point. There was a pirate [station] in Birmingham. I didn’t play much, but I played a few times. They’d move around; they’d get shut down pretty quick. That was through my cousins; I mostly tagged along. I hopped in the car, brought my records, played, and went home.
In the late ‘90s, I got kind of bored of jungle and techno. It started to feel a bit stiff and less experimental. I started to dive into more experimental music. I studied art. I started listening to Stockhausen and the like every day. I spent a lot of time studying that and working on computer music. I did multi-channel stuff, and I did gallery-leaning things and more conceptual stuff: noise music, improvisations and things like that.
In the early 2000s, I was in London, and when grime and dubstep happened, I said, “Wow—what’s that?” That’s when it grabbed me again. That’s the short answer; it’s 10-12 years of your life! Once I dipped out of dance music, so much opened up: the history of electronic music, musique concrète, sound poetry and German electronische musik. Up until I made Diversions, I was in that space a lot. In university, I studied a sound and theory art degree. We’d still go raving sometimes, but I was burrowing deep into experimental music and the arts.
Who’s “we”?
Lee Gamble: I moved to London in 2000. It was the people I was at uni with, and people around in the scene. Not many of us had dance-music backgrounds; many of us were into dance music and something else, too. I threw a series of nights in London after I finished uni, because there were a bunch of us saying, “I really like musique concrète. But dance music is amazing, too.” At the time, they never seemed to be close together. So we put on a series of events in a small venue: we’d do a £3 cover, and we’d get people to do an experimental improv thing. We’d go on after, and we’d DJ, like, A Guy Called Gerald, Pan Sonic, and Jeff Mills, but we’d clash it on top of Merzbow. It was, like, “F*ck it—I see value in both of these things. I don’t see why they should feel separate.” We did about ten of these, and they got quite mental.
After the last few, we got banned from the venue. For one of these, we had someone playing computer music, and he did an unbearably loud set—I wouldn’t call it “drone,” because that sounds kind of relaxing. This was ear-piercing: totally hallucinogenic and alien Pretty much everyone left, and the people we were renting from said, “What’s he doing? You’ve got to tell him to stop.” I said, “I’m not telling him to stop.” It was this horrifically loud sine-tone. People behind the bar hid until it went away. When that stopped, we’d start playing, like, pounding techno or something. For £3, you can have quite a nice Tuesday night. [laughs]
When did you start painting?
Lee Gamble: I drew all the time when I was a kid. I started painting properly a couple of years ago. Before that, I never really had the space—I still don’t have enough space. I want to make bigger pieces! I lost a few people who were really close to me, and I had a hole to fill. And a painter friend of mine asked me if I painted, and I said, “Not really, but I drew all my life.” He told me he’d love to see what I’d paint. He very kindly gave me a canvas, and some paints, and told me to do it. I love it. I’ve been spending a lot of time on that in the past year.
After I’d finished Flush Real Pharynx for Hyperdub, I felt lost. Between the pandemic and losing these people, I said, “I’m gonna get grounded again.” Painting made me realize that I really like yellow! I didn’t know that. There’s things like that, where you say, “I like that color; I don’t like that color.” You can really transfer that. Not technically, but in terms of saying, “I don’t give a shit about all that—what does this sound like in my monitors? Do I like this?” I’d allow that to happen, and I wouldn’t question it any further. That’s how Models came about, I suppose.
I said, “I want to make a record that I really enjoy making.” It’s not like I’ve not done that before, but Flush Real Pharynx was such a big thing. It was really hard to do. I spoke to Steve [Goodman, AKA Kode9] at Hyperdub, and I said, “I’m making something, and it’s going to be really simple. I’m gonna do what I want; it’s a super emotional record.”
That seems really simple, but I’d lost my ability to do that. I’d lost the ability to judge, or believe, my own taste a little. There’s so many trends, sounds, and styles—things move so fast. And, all of a sudden, you’re like, “Where the f*ck have I gone?” The painting really helped with this. It helped me focus on what I wanted to do. Nobody’s gonna come up and say, “I think you should put some pink there.” It’s up to me; it starts and ends with my decisions. I approached Models in the same way.
You started this record in 2020, yeah?
Lee Gamble: I didn’t start the musical production then, but I did start to write the ideas for the record, and I started to correspond with Joseph, an AI programmer from America. I was aware of the possibilities that were available in AI. It’s moved a huge amount in three years. I was wary not to dive in too fast and get this rush of newness, write quickly and rely on the technology to hold the record up. We had back-and-forths where we were building vocal models and testing things out. I eventually told him, “I’m going to go away now; I need six months.” I needed to figure out how to make it my work rather than having it be the work of a piece of technology, which I have no interest in.
I worked with instrumentation, and I tried to figure out what would work with the voice and give it space. I was playing around with hooks. I wrote a couple of tracks without voice and added it in, but most of it was written around the vocals. I built it around earworms—phrases that I’d picked out from stuff that the network spat out.
Given I’m going down this wormhole, I’m aware that it’s problematic. I want to use this technology, but I want to be respectful of its problems and how it might sit in popular culture. So I wanted to take my time. I had more time than I’ve ever had to make a record. I read as much as I could, from as many different voices as I could, about how this would affect people: things like the racialized aspect of datasets, and the voice, too. It’s been mimicked a lot, and it’s been caricatured a lot, and it’s been taken and used by white culture, often in terms of American music.
I was conscious of these things, but it’s also such a tantalizing technology for an artist to not use. I’m still in a quandary with it. The fears around job replacement are totally valid; AI will definitely take jobs. It already has. I don’t think it’s going to take anyone’s singing away. I wanted to make sure this record didn’t feel like a technologized record. It’s about the idea that these aren’t real people, but it’s not suggesting the replacement of people. It’s about, “How much mood can I get out of this? How human can I make the technology?” How human does a singing voice feel when it’s disembodied? How emotional can a digital simulation feel? This is without ignoring what they are: simulations of people. I wanted to make sure I leaned into that, but I didn’t want it to be the thing that wrote the record.
I wanted to make sure I’m extremely conscious about it. The voices aren’t modeled after any particular person. It was, at first, but after a couple of years, I was like, “I can’t do this.” We built a system of four voices and kept feeding it into itself. Eventually, it got so far away from whatever it had been fed that it’s not anyone. The whole album, to me, feels like a trace of a pop record. I’d get phrases back and pick from them, and while I wouldn’t say I wrote these like pop songs, I allowed myself to drift into that space. This applies to the way it sounds, too—I wasn’t trying to make it sound crumbly, like Diversions, or super hi-res, like parts of Flush. With pop records that I love, even now, I’ll grab one track off one record, and I’ll hardly be able to play anything else. Everything else sounds terrible! I wanted to grab that energy: something like the aura of pop.
That’s what I was trying to think about. I wanted to make something that felt like you’re connecting to it in some way. The thing with these voices is that they don’t have a physical presence about them. I’m not trying to undermine the technology, of course. You feed them human stuff; all of AI is, fundamentally, human. But, in terms of their presence, they seem like they pop out of nowhere.
The first thing I did was build a model of my own voice, which is actually on the record, on A Million Pieces of You. I’m on there—well, it’s not me; it’s my deepfake. It’s taken a few years to wrangle with it as a cultural thing, which was as much the problem as making the record was. I wanted to make something that sounded like it was human, because, fundamentally, that’s what they are. They might come out of these incredible pieces of technology, but they’re human things. They’re a very weird entity. I don’t know what you’d call them. You have this idea of the uncanny valley, but it’s not that. It’s not like it’s trying to impersonate somebody; they seem to have their own character.
With this project, I kept returning to Autechre. You’re saying, “How can we take these ideas and put humanity back in?” I would say this and one of Autechre’s long-form live things are doing a similar thing.
Lee Gamble: When I hear Autechre’s current music—especially their live shows—it sounds like they’ve built a system that sounds like them. I think that’s a nice way of hearing their stuff. It’s a system that’s been honed over many years to sound that way and exist as that system. Models is not doing that in such a singular way, but the voice modeling aspect was definitely a similar kind of technological conversation. It doesn’t feel like a conversation when I’m playing a keyboard, although it is a piece of technology. I felt differently with this, because I wasn’t sure what I’d get back. It’s a matter of making a choice: do I like that or not? Do I put yellow on the canvas, or red, or something else?
That’s where the album emerged from. I’d run a plugin to find out what key those voices were in. I’d set everything around that key and write around them. The technology was giving me a lot of input, in a distant way. It’s not like I’m tweaking anything live: it’s a matter of taking stuff, and, an hour later, going “Wow, okay—you thought that? Interesting.” That’s where the magic was: saying, “Whoa! You reckon I should write something like that?” [laughs] There’s definitely that generative aspect to it. If we gave it reverb, it would turn the sounds into mush, and we’d have to start cleaning it up. I was literally mixing and mastering the sound of datasets.
I have a feeling that, with voice simulation technology, the endgame will be when you can’t tell the difference between AI voices and human voices. I don’t find that particularly interesting. The fact that they were scratchy and indecipherable and smudged, and had an Elizabeth Fraser, or “mumble” rap-esque non-language, where you don’t know what they’re saying but you’re getting emotion and narrative from it anyways—I was keen on exploring that. Some of the weird, fragmented, “mumble” rap stuff is huge, even though it’s really experimental. Once I got my head around that, I was happy to work with them in the state they’re in. It took tons of work to get them sounding like that, but, eventually, I was happy that they sounded like that rather than saying, “This sounds exactly like a person singing.”
That was only possible because the technology’s not great yet. In a few years, you might only have the really good models. I think you lose something in that, so I wanted to get it done now. In 10 years’ time, you might say, “Listen to how those voices sounded then! They were so bad!” The image things look completely different to how they did a year or two ago. I don’t know if the old ones are there anymore. It’s not like a piece of hardware that you can pick up again.
There’s an AI artist I found a few months ago who’s leaned into digital fragmentation and software errors. They took the bits that Midjourney doesn’t understand yet and said, “What do we do with the technology as it currently stands?” These aren’t flaws, necessarily.
Lee Gamble: That’s exactly what I did with Models. The crux, for me, was that detritus: this kind of childlike attempt to communicateThat kind of surplus value happens in technology all the time. But AI is so new, and it’s moving so fast. The first synthesized cello sound sounds terrible now. Now, if it’s mixed well, it’s hard to tell it’s fake. But that’s taken 50 years. AI’s moved at an incredible rate in the last three. The speed at which this technology evolved, in a few short years, is really scary. There’s no point in racing it, because it’s going to beat you. This stuff is owned by all the big corporations—they’ve got billions of pounds to throw at this.
With this record, I said, “I know damn well that if, in two years’ time, if I say this voice modeling system sounds like a human, it’ll be trash.” So, forget that—just make a record with the technology that’s in front of you, and lean into the limitations. I can’t compete with these companies; I’m just a guy who knows a brilliant programmer. We don’t have a team behind us. Forget them—they’re doing their thing, and I’ll lean into the problems that they’re trying to fix.
It’s the idea of making something that sounds like a moment.
Lee Gamble: That’s something I’ve been thinking about recently. You’ve got this idea about the hardcore continuum, for instance. It keeps shapeshifting and latching onto other things, and every now and again, you’ll get a burst, and you’ll get a new genre. I feel like that’s finished. I don’t think it’s not actually happening: I just think that since we’re so deep into social media and streaming, access to things has been leveled.
I can be on Spotify, and I could be in a playlist with Beyoncé. That wouldn’t have happened before. If I was 10 years old now, I don’t have this reference to date or time: here’s jazz, here’s prog, here’s krautrock, here’s techno. It’s just all there. In a sense, that’s really amazing. I’ve really enjoyed Lil Yachty’s new record this year, and it seems to lean into that: there are points where it sounds like Pink Floyd, and then it’ll sound super modern, and then it sounds like the ’70s, and then it sounds like old-school computer games, and then it sounds like old hip-hop. It sounds like now.
I don’t think younger kids give a shit—and I don’t think they should—if something was made in 1970 or 2023. Who cares? If someone’s playing a track in a club, and you’ve got 2,000 people dancing to it, who cares if it’s from 1994? That’s what I mean about this idea of a continuum. It’s all been piledrived into a wall and splattered. It comes with its issues; I get that. But I think that time splattering is a good thing.
The question about what is, or sounds, “modern” or “old” gets mushier over time. If you’ve never heard something before, it’s new. I heard someone on NTS the other day—a working-class person from London—playing Steve Reich. They said, “I’m going down this wormhole with Steve Reich.” That’s amazing. That stuff is historically of the “academy,” or the “middle classes,” or the “conservatoire,” but there’s also a clear history from Indian music to minimalism. It all gets constantly re-contextualised in the streaming era.
I went down a YouTube wormhole with 100 gecs’ stuff. I’m not a stan, but it crossed me in the same way. It’s a mixture of things that sounds exactly like now. It doesn’t matter if I like it aesthetically. As a cultural object, you’ve taken this on and just ran with it! You don’t give a f*ck. I don’t know how old they are, but they’re younger than me. I think younger people aren’t trying to make a point of that; that’s just how it is. [laughs] There’s no reason to have this linear time where everything gets discarded—”Oh, it’s old.” I always thought that was a bit pointless.
Also, to flip towards the future: people are always like, “This sounds like the future.” What does that mean? I’m not trying to be facetious; I honestly don’t know what that means.
I’m definitely guilty of using that terminology.
Lee Gamble: You’re a writer, so it makes sense—it’s a good semantic thing. People go “Ah! I get it.” But what do you think the person’s gonna think it sounds like?
I try to use it to say, “This person is going in a really exciting direction, and maybe this is a possible vision for the future.”
Lee Gamble: That Lil Yachty record sounds like now. It might sound old in a year. But, for now, it sounds like the present. It’s that cross-pollination—corecore shit on TikTok, and mashups. I keep hearing the word “Dada” again. It feels like vertical collage. I did a piece recently which was performed at Berlin Atonal; they did a Xenakis X100 thing. I’m a huge admirer of Xenakis‘ work. I wrote this in memory of him, so I thought, “What would he be doing now?” You’re damn well sure he’s not doing what he was doing then. He was doing stochastic computer synthesis methods in the ’90s when very few people were doing that in the world. If he was a young person in 2023, I doubt he’d be writing for solo timpani. He’s a techno person, in a sense. I doubt he’d want to becalled a techno guy, but f*ck it, that’s fine.
It’s that idea of possible futures: what kind of future are you offering?
Lee Gamble: There’s definitely some of that aspect to Models. Take the Flush Real Pharynx series I did. That was me trying to come to terms with the culture clash we’re talking about: multiple screens at the same time. It’s a kind of collage. That wasn’t the defining thing for the record, but I had that in my head. I wanted that forefront in my brain while I was writing it. That’s why I called it Flush Real Pharynx: it’s about flushing your voice away and being bombarded with stuff.
With Models, I said, “F*ck that.” That’s really bad for our mental health. I’m gonna bounce, and I’m gonna make something more emotionally driven, or soulful, rather than a TikTok kind of thing. I’m not sure how long people can take that, just as a flesh-and-bone kind of thing. As brains, you and I are not that different to how we were 2,000 years ago. It’s having an effect on people’s mental health. As a culture, we try to come to terms with it. That’s what I did with Flush, and I injured myself by doing that, so I wanted to soothe myself a bit and switch it all off.
I’m not qualified to explain this, but in terms of how things would affect my music: previously it was cities. If you grew up in Detroit, you might be into one thing, and if you grew up in New York, you might be into something else, and if you grew up in Birmingham, it might be another. Now, it’s the internet. If you’re into something, you can find it. There are so many positive things about it, but at this point in my life, it’s better for my mental health for me to switch it off when I’m going to make something. It’s taken me a few years to learn that, but it’s fine. Sometimes things take a while to work out. If I were to make another record, I’d want to hire a studio in a barn somewhere on a mountain for three months.
You started Models when AI was prehistoric compared to where it is now, and when it wasn’t really a public conversation. What’s it been like to watch that evolution in real time?
Lee Gamble: Neural networks are from the ’50s, so AI is not new. But data happened. They didn’t have enough data sets to make a good neural network in the ’50s. Even when we started doing the voice modeling stuff, you’d need a bunch of stuff. Eventually, it’d get down to 60 seconds of something that you could work with. That was with two people and a normal computer; it’s not with Meta or Google or whatever. It was obvious to me that it was going to accelerate at rate of knots.
It’s a little bit of brain, in a sense, that can learn how to do something. It’s a simulation technology. Anything you experience is because you have a brain, fundamentally. If the brain is allowing everything to be knowable, then a neural network can try any of that. So, of course it’s going to be everywhere. We’ve seen it first in image-making, because most online communities are built around images. I knew it was going to be quick, so I had to say, “We are where we are, and let’s roll with that version.”
When I started writing [Models], I bowed out of the AI conversation. I started looking at it again recently, because I needed a break from it all. I read the other day about someone who had been in a coma for 20 years, and they lost their voice. They came out of the coma, and they’ve used AI to train them to have a voice again, which is such a Models thing. I think it’ll become ubiquitous. Then it’ll level off, and that’s when the interesting things will happen.
I wanted this to sound like a record that you’d make when AI isn’t a “thing” anymore. It’s a bit like early Herbie Hancock records. Everybody wanted to know what synthesizer he was using. 20 years later, nobody’s asking; he just plays. I wanted to place Models there somehow. I realize this is coming close to making “futuristic” music. [laughs] Right now, I’m overwhelmed by the shininess and potential of this technology. In years’ time, people won’t think of it like that.
When you put something like this out, you get people acting as though you’re okay with all that happening. The problem, for me, is that AI is already in every part of culture. So saying, “I’m not going to use it because it’s bad”—it’s not like not using Amazon because they’re scumbags. But we can’t not use AI. Fundamentally, it’s built on our collective labor, art, and time. I don’t want that to sound like a cop-out; I’m no libertarian. But, at a certain point, you’re going to have to engage with it. When that happens, it’ll either be unknowingly or with a bit of knowledge. The only way for me to offset that was to say, “Educate yourself on it, and hear some voices that aren’t of a similar background.”
I’m not here to make an excuse as to why I’d use it. I’m not out there saying, “This is amazing; we don’t need singers anymore.” I’m an artist, and I wouldn’t even begin to think that. If I can make something that can resonate emotionally with people, like I hope this record can, then maybe that’s okay. If I’m able to make people think about how they can use the technology in their way outside of any corporate dystopias, then maybe that’s okay. I’m not expecting to be able to make a record and say something about AI in general. There are things that we should be aware of and use as artists, and there are other things we should be wary of and at least talk about.
For example, there’s the protection of people’s voices. If I fed the first model one person’s voice and said, “Okay, that’s it,” that wouldn’t have been enough. I had to process it until I couldn’t tell who it was anymore; I had to eliminate the training from it somewhat. It’s not my voice. I understand there’s a complex relationship there. I had to be, and will continue to be, aware of that. People that look like me have claimed ownership of other people’s cultures and work, and that has to be taken seriously. If this can open a conversation about that, then that’s the least you can do if you’re using the technology. You can at least acknowledge that there’s a history of people taking people’s voices, mimicking them, and selling them because they’re a different complexion to the people who invented the style. I don’t think that any of this sounds like any particular person. I hope that it doesn’t, because that’s something I worked really hard to avoid.
It would be so easy to do that: you’ve seen the AI deepfake voices. I think the voice and the image in particular are very hot things to start playing around with. The image of another person that isn’t you is a very sensitive thing to mess with. What else have you got to identify yourself with other than the way you feel, look, and sound? That’s the most intimate thing, and to go, “I could do that now, because I have this Google sheet I can click on?” We have to think of it more respectfully than that.
I’m sure some people will disagree with that, but I’ve not heard anything yet when I’ve spoken about the record. When I was making it, I thought, “Maybe people will be really against this.” But I hope that, when people hear the record, they hear that that’s not where it’s coming from. It doesn’t feel like it’s coming from a position of ownership to me. It’s a dialogue. It’s: “Here’s where we’re going, and I’m trying to make something beautiful and touching with it.” That’s all I could do. I didn’t want to make something “icy,” or “fresh,” or “modern.” “Alien” or “other-dimensional,” sure, but I wanted to make it as warm as I could. I’m not too saccharine, but there’s definitely a few moments on there where I’m like, “F*cking hell.” The inner Cure fan came out. [laughs] I hope that people don’t think of it as an ode to a dystopian technology.
It’s a reckoning.
Lee Gamble: Yeah. It’s here. Try and own it: make it your work. It’s just some paint. It’s just some acrylic in a tube. Until you do anything with it, it doesn’t mean anything. There’s this incredible number of images made by image generation systems. Hito Steyerl is really interesting on this topic in terms of images. She talks about the image in one of her pieces of writing, In Defense of the Poor Image. It’s a piece of writing about the digital image. It goes through all sorts of different processes; it’ll get compressed again and again, and the colors change, and it all falls apart. So you get these poor images even though we’re in these hi-tech times.
This writing goes into a defense of the poor image, and it explores what qualities can be found in it. In the piece of writing I did, I titled one of my chapters—where I talk about the ‘crappy’ nature of the voices I used—”In Defense of the Poor Voice.” I quote Steyerl there, of course. In her writing, she says, “The poor image embodies the afterlife of many former masterpieces of cinema and video art.” I read her talk about it very eloquently, and I said, “I get that.” If I just swap all of that for the voice rather than the .JPG, then it’s the same thing. I’m seeing the some of the same advantages in the lack of quality as she’d seen. That really resonated with me.
What’s next for you?
Lee Gamble: It’s mainly the live show. I’m working with Candela Capítan, a performance artist and choreographer. I looked at her work and thought it was amazing, in terms of how it relates to the body, repetition, alienation, and the gaze and technologized image. I wanted to think about how to present Models live, and I didn’t want to be onstage with a laptop. From day one, I wanted to bring a physical body back into space—but not mine. Humans gave me the information; I simulated and disembodied them; and, now, I ought to put them back in. A lot of the movement is informed by the ideas of cloning, mirrors, and how people replicate each others’ movements. We tried to get twins to perform in the live show.
It’ll premier at Unsound in Poland, and then in London. I wanted to develop the choreography around the idea of how things learn—sometimes they learn badly. Obviously, Candela has brought a huge amount to the work. I gave them a PDF of ideas, and they read it; I played the music to them, and they improvised. It was very moving: I’d never seen anyone do that to my work before. So, I thought, “I want to get on this.”
I’m DJing as regularly as I can, just banging it out and enjoying myself. I’m really enjoying painting. I’m hoping to do some installation work with these voices as well. I don’t know if it’s just me, but I feel like my life moves in ten-year cycles. I did therapy last year for the first time in my life, and that came up. It was really helpful. I was really glad that people pushed me to do it. I’m not against it, of course, but I come from a background where that wasn’t a thing. It took me a long time, but I needed it, and it was super helpful. One of the things that came up is the idea of ten-year cycles in my life. Diversions was 2012, and that was a moment where I did a bunch of other stuff. Now, it’s 2023, and there’s Models, and I’m painting again. Now, I’m ready to boot them glass ceilings as hard as I can for the next ten years.
After the last few, we got banned from the venue. For one of these, we had someone playing computer music, and he did an unbearably loud set—I wouldn’t call it “drone,” because that sounds kind of relaxing. This was ear-piercing: totally hallucinogenic and alien Pretty much everyone left, and the people we were renting from said, “What’s he doing? You’ve got to tell him to stop.” I said, “I’m not telling him to stop.” It was this horrifically loud sine-tone. People behind the bar hid until it went away. When that stopped, we’d start playing, like, pounding techno or something. For £3, you can have quite a nice Tuesday night. [laughs]
When did you start painting?
Lee Gamble: I drew all the time when I was a kid. I started painting properly a couple of years ago. Before that, I never really had the space—I still don’t have enough space. I want to make bigger pieces! I lost a few people who were really close to me, and I had a hole to fill. And a painter friend of mine asked me if I painted, and I said, “Not really, but I drew all my life.” He told me he’d love to see what I’d paint. He very kindly gave me a canvas, and some paints, and told me to do it. I love it. I’ve been spending a lot of time on that in the past year.
After I’d finished Flush Real Pharynx for Hyperdub, I felt lost. Between the pandemic and losing these people, I said, “I’m gonna get grounded again.” Painting made me realize that I really like yellow! I didn’t know that. There’s things like that, where you say, “I like that color; I don’t like that color.” You can really transfer that. Not technically, but in terms of saying, “I don’t give a shit about all that—what does this sound like in my monitors? Do I like this?” I’d allow that to happen, and I wouldn’t question it any further. That’s how Models came about, I suppose.
I said, “I want to make a record that I really enjoy making.” It’s not like I’ve not done that before, but Flush Real Pharynx was such a big thing. It was really hard to do. I spoke to Steve [Goodman, AKA Kode9] at Hyperdub, and I said, “I’m making something, and it’s going to be really simple. I’m gonna do what I want; it’s a super emotional record.”
That seems really simple, but I’d lost my ability to do that. I’d lost the ability to judge, or believe, my own taste a little. There’s so many trends, sounds, and styles—things move so fast. And, all of a sudden, you’re like, “Where the f*ck have I gone?” The painting really helped with this. It helped me focus on what I wanted to do. Nobody’s gonna come up and say, “I think you should put some pink there.” It’s up to me; it starts and ends with my decisions. I approached Models in the same way.
You started this record in 2020, yeah?
Lee Gamble: I didn’t start the musical production then, but I did start to write the ideas for the record, and I started to correspond with Joseph, an AI programmer from America. I was aware of the possibilities that were available in AI. It’s moved a huge amount in three years. I was wary not to dive in too fast and get this rush of newness, write quickly and rely on the technology to hold the record up. We had back-and-forths where we were building vocal models and testing things out. I eventually told him, “I’m going to go away now; I need six months.” I needed to figure out how to make it my work rather than having it be the work of a piece of technology, which I have no interest in.
I worked with instrumentation, and I tried to figure out what would work with the voice and give it space. I was playing around with hooks. I wrote a couple of tracks without voice and added it in, but most of it was written around the vocals. I built it around earworms—phrases that I’d picked out from stuff that the network spat out.
Given I’m going down this wormhole, I’m aware that it’s problematic. I want to use this technology, but I want to be respectful of its problems and how it might sit in popular culture. So I wanted to take my time. I had more time than I’ve ever had to make a record. I read as much as I could, from as many different voices as I could, about how this would affect people: things like the racialized aspect of datasets, and the voice, too. It’s been mimicked a lot, and it’s been caricatured a lot, and it’s been taken and used by white culture, often in terms of American music.
I was conscious of these things, but it’s also such a tantalizing technology for an artist to not use. I’m still in a quandary with it. The fears around job replacement are totally valid; AI will definitely take jobs. It already has. I don’t think it’s going to take anyone’s singing away. I wanted to make sure this record didn’t feel like a technologized record. It’s about the idea that these aren’t real people, but it’s not suggesting the replacement of people. It’s about, “How much mood can I get out of this? How human can I make the technology?” How human does a singing voice feel when it’s disembodied? How emotional can a digital simulation feel? This is without ignoring what they are: simulations of people. I wanted to make sure I leaned into that, but I didn’t want it to be the thing that wrote the record.
I wanted to make sure I’m extremely conscious about it. The voices aren’t modeled after any particular person. It was, at first, but after a couple of years, I was like, “I can’t do this.” We built a system of four voices and kept feeding it into itself. Eventually, it got so far away from whatever it had been fed that it’s not anyone. The whole album, to me, feels like a trace of a pop record. I’d get phrases back and pick from them, and while I wouldn’t say I wrote these like pop songs, I allowed myself to drift into that space. This applies to the way it sounds, too—I wasn’t trying to make it sound crumbly, like Diversions, or super hi-res, like parts of Flush. With pop records that I love, even now, I’ll grab one track off one record, and I’ll hardly be able to play anything else. Everything else sounds terrible! I wanted to grab that energy: something like the aura of pop.
That’s what I was trying to think about. I wanted to make something that felt like you’re connecting to it in some way. The thing with these voices is that they don’t have a physical presence about them. I’m not trying to undermine the technology, of course. You feed them human stuff; all of AI is, fundamentally, human. But, in terms of their presence, they seem like they pop out of nowhere.
The first thing I did was build a model of my own voice, which is actually on the record, on A Million Pieces of You. I’m on there—well, it’s not me; it’s my deepfake. It’s taken a few years to wrangle with it as a cultural thing, which was as much the problem as making the record was. I wanted to make something that sounded like it was human, because, fundamentally, that’s what they are. They might come out of these incredible pieces of technology, but they’re human things. They’re a very weird entity. I don’t know what you’d call them. You have this idea of the uncanny valley, but it’s not that. It’s not like it’s trying to impersonate somebody; they seem to have their own character.
With this project, I kept returning to Autechre. You’re saying, “How can we take these ideas and put humanity back in?” I would say this and one of Autechre’s long-form live things are doing a similar thing.
Lee Gamble: When I hear Autechre’s current music—especially their live shows—it sounds like they’ve built a system that sounds like them. I think that’s a nice way of hearing their stuff. It’s a system that’s been honed over many years to sound that way and exist as that system. Models is not doing that in such a singular way, but the voice modeling aspect was definitely a similar kind of technological conversation. It doesn’t feel like a conversation when I’m playing a keyboard, although it is a piece of technology. I felt differently with this, because I wasn’t sure what I’d get back. It’s a matter of making a choice: do I like that or not? Do I put yellow on the canvas, or red, or something else?
That’s where the album emerged from. I’d run a plugin to find out what key those voices were in. I’d set everything around that key and write around them. The technology was giving me a lot of input, in a distant way. It’s not like I’m tweaking anything live: it’s a matter of taking stuff, and, an hour later, going “Wow, okay—you thought that? Interesting.” That’s where the magic was: saying, “Whoa! You reckon I should write something like that?” [laughs] There’s definitely that generative aspect to it. If we gave it reverb, it would turn the sounds into mush, and we’d have to start cleaning it up. I was literally mixing and mastering the sound of datasets.
I have a feeling that, with voice simulation technology, the endgame will be when you can’t tell the difference between AI voices and human voices. I don’t find that particularly interesting. The fact that they were scratchy and indecipherable and smudged, and had an Elizabeth Fraser, or “mumble” rap-esque non-language, where you don’t know what they’re saying but you’re getting emotion and narrative from it anyways—I was keen on exploring that. Some of the weird, fragmented, “mumble” rap stuff is huge, even though it’s really experimental. Once I got my head around that, I was happy to work with them in the state they’re in. It took tons of work to get them sounding like that, but, eventually, I was happy that they sounded like that rather than saying, “This sounds exactly like a person singing.”
That was only possible because the technology’s not great yet. In a few years, you might only have the really good models. I think you lose something in that, so I wanted to get it done now. In 10 years’ time, you might say, “Listen to how those voices sounded then! They were so bad!” The image things look completely different to how they did a year or two ago. I don’t know if the old ones are there anymore. It’s not like a piece of hardware that you can pick up again.
There’s an AI artist I found a few months ago who’s leaned into digital fragmentation and software errors. They took the bits that Midjourney doesn’t understand yet and said, “What do we do with the technology as it currently stands?” These aren’t flaws, necessarily.
Lee Gamble: That’s exactly what I did with Models. The crux, for me, was that detritus: this kind of childlike attempt to communicateThat kind of surplus value happens in technology all the time. But AI is so new, and it’s moving so fast. The first synthesized cello sound sounds terrible now. Now, if it’s mixed well, it’s hard to tell it’s fake. But that’s taken 50 years. AI’s moved at an incredible rate in the last three. The speed at which this technology evolved, in a few short years, is really scary. There’s no point in racing it, because it’s going to beat you. This stuff is owned by all the big corporations—they’ve got billions of pounds to throw at this.
With this record, I said, “I know damn well that if, in two years’ time, if I say this voice modeling system sounds like a human, it’ll be trash.” So, forget that—just make a record with the technology that’s in front of you, and lean into the limitations. I can’t compete with these companies; I’m just a guy who knows a brilliant programmer. We don’t have a team behind us. Forget them—they’re doing their thing, and I’ll lean into the problems that they’re trying to fix.
It’s the idea of making something that sounds like a moment.
Lee Gamble: That’s something I’ve been thinking about recently. You’ve got this idea about the hardcore continuum, for instance. It keeps shapeshifting and latching onto other things, and every now and again, you’ll get a burst, and you’ll get a new genre. I feel like that’s finished. I don’t think it’s not actually happening: I just think that since we’re so deep into social media and streaming, access to things has been leveled.
I can be on Spotify, and I could be in a playlist with Beyoncé. That wouldn’t have happened before. If I was 10 years old now, I don’t have this reference to date or time: here’s jazz, here’s prog, here’s krautrock, here’s techno. It’s just all there. In a sense, that’s really amazing. I’ve really enjoyed Lil Yachty’s new record this year, and it seems to lean into that: there are points where it sounds like Pink Floyd, and then it’ll sound super modern, and then it sounds like the ’70s, and then it sounds like old-school computer games, and then it sounds like old hip-hop. It sounds like now.
I don’t think younger kids give a shit—and I don’t think they should—if something was made in 1970 or 2023. Who cares? If someone’s playing a track in a club, and you’ve got 2,000 people dancing to it, who cares if it’s from 1994? That’s what I mean about this idea of a continuum. It’s all been piledrived into a wall and splattered. It comes with its issues; I get that. But I think that time splattering is a good thing.
The question about what is, or sounds, “modern” or “old” gets mushier over time. If you’ve never heard something before, it’s new. I heard someone on NTS the other day—a working-class person from London—playing Steve Reich. They said, “I’m going down this wormhole with Steve Reich.” That’s amazing. That stuff is historically of the “academy,” or the “middle classes,” or the “conservatoire,” but there’s also a clear history from Indian music to minimalism. It all gets constantly re-contextualised in the streaming era.
I went down a YouTube wormhole with 100 gecs’ stuff. I’m not a stan, but it crossed me in the same way. It’s a mixture of things that sounds exactly like now. It doesn’t matter if I like it aesthetically. As a cultural object, you’ve taken this on and just ran with it! You don’t give a f*ck. I don’t know how old they are, but they’re younger than me. I think younger people aren’t trying to make a point of that; that’s just how it is. [laughs] There’s no reason to have this linear time where everything gets discarded—”Oh, it’s old.” I always thought that was a bit pointless.
Also, to flip towards the future: people are always like, “This sounds like the future.” What does that mean? I’m not trying to be facetious; I honestly don’t know what that means.
I’m definitely guilty of using that terminology.
Lee Gamble: You’re a writer, so it makes sense—it’s a good semantic thing. People go “Ah! I get it.” But what do you think the person’s gonna think it sounds like?
I try to use it to say, “This person is going in a really exciting direction, and maybe this is a possible vision for the future.”
Lee Gamble: That Lil Yachty record sounds like now. It might sound old in a year. But, for now, it sounds like the present. It’s that cross-pollination—corecore shit on TikTok, and mashups. I keep hearing the word “Dada” again. It feels like vertical collage. I did a piece recently which was performed at Berlin Atonal; they did a Xenakis X100 thing. I’m a huge admirer of Xenakis‘ work. I wrote this in memory of him, so I thought, “What would he be doing now?” You’re damn well sure he’s not doing what he was doing then. He was doing stochastic computer synthesis methods in the ’90s when very few people were doing that in the world. If he was a young person in 2023, I doubt he’d be writing for solo timpani. He’s a techno person, in a sense. I doubt he’d want to becalled a techno guy, but f*ck it, that’s fine.
It’s that idea of possible futures: what kind of future are you offering?
Lee Gamble: There’s definitely some of that aspect to Models. Take the Flush Real Pharynx series I did. That was me trying to come to terms with the culture clash we’re talking about: multiple screens at the same time. It’s a kind of collage. That wasn’t the defining thing for the record, but I had that in my head. I wanted that forefront in my brain while I was writing it. That’s why I called it Flush Real Pharynx: it’s about flushing your voice away and being bombarded with stuff.
With Models, I said, “F*ck that.” That’s really bad for our mental health. I’m gonna bounce, and I’m gonna make something more emotionally driven, or soulful, rather than a TikTok kind of thing. I’m not sure how long people can take that, just as a flesh-and-bone kind of thing. As brains, you and I are not that different to how we were 2,000 years ago. It’s having an effect on people’s mental health. As a culture, we try to come to terms with it. That’s what I did with Flush, and I injured myself by doing that, so I wanted to soothe myself a bit and switch it all off.
I’m not qualified to explain this, but in terms of how things would affect my music: previously it was cities. If you grew up in Detroit, you might be into one thing, and if you grew up in New York, you might be into something else, and if you grew up in Birmingham, it might be another. Now, it’s the internet. If you’re into something, you can find it. There are so many positive things about it, but at this point in my life, it’s better for my mental health for me to switch it off when I’m going to make something. It’s taken me a few years to learn that, but it’s fine. Sometimes things take a while to work out. If I were to make another record, I’d want to hire a studio in a barn somewhere on a mountain for three months.
You started Models when AI was prehistoric compared to where it is now, and when it wasn’t really a public conversation. What’s it been like to watch that evolution in real time?
Lee Gamble: Neural networks are from the ’50s, so AI is not new. But data happened. They didn’t have enough data sets to make a good neural network in the ’50s. Even when we started doing the voice modeling stuff, you’d need a bunch of stuff. Eventually, it’d get down to 60 seconds of something that you could work with. That was with two people and a normal computer; it’s not with Meta or Google or whatever. It was obvious to me that it was going to accelerate at rate of knots.
It’s a little bit of brain, in a sense, that can learn how to do something. It’s a simulation technology. Anything you experience is because you have a brain, fundamentally. If the brain is allowing everything to be knowable, then a neural network can try any of that. So, of course it’s going to be everywhere. We’ve seen it first in image-making, because most online communities are built around images. I knew it was going to be quick, so I had to say, “We are where we are, and let’s roll with that version.”
When I started writing [Models], I bowed out of the AI conversation. I started looking at it again recently, because I needed a break from it all. I read the other day about someone who had been in a coma for 20 years, and they lost their voice. They came out of the coma, and they’ve used AI to train them to have a voice again, which is such a Models thing. I think it’ll become ubiquitous. Then it’ll level off, and that’s when the interesting things will happen.
I wanted this to sound like a record that you’d make when AI isn’t a “thing” anymore. It’s a bit like early Herbie Hancock records. Everybody wanted to know what synthesizer he was using. 20 years later, nobody’s asking; he just plays. I wanted to place Models there somehow. I realize this is coming close to making “futuristic” music. [laughs] Right now, I’m overwhelmed by the shininess and potential of this technology. In years’ time, people won’t think of it like that.
When you put something like this out, you get people acting as though you’re okay with all that happening. The problem, for me, is that AI is already in every part of culture. So saying, “I’m not going to use it because it’s bad”—it’s not like not using Amazon because they’re scumbags. But we can’t not use AI. Fundamentally, it’s built on our collective labor, art, and time. I don’t want that to sound like a cop-out; I’m no libertarian. But, at a certain point, you’re going to have to engage with it. When that happens, it’ll either be unknowingly or with a bit of knowledge. The only way for me to offset that was to say, “Educate yourself on it, and hear some voices that aren’t of a similar background.”
I’m not here to make an excuse as to why I’d use it. I’m not out there saying, “This is amazing; we don’t need singers anymore.” I’m an artist, and I wouldn’t even begin to think that. If I can make something that can resonate emotionally with people, like I hope this record can, then maybe that’s okay. If I’m able to make people think about how they can use the technology in their way outside of any corporate dystopias, then maybe that’s okay. I’m not expecting to be able to make a record and say something about AI in general. There are things that we should be aware of and use as artists, and there are other things we should be wary of and at least talk about.
For example, there’s the protection of people’s voices. If I fed the first model one person’s voice and said, “Okay, that’s it,” that wouldn’t have been enough. I had to process it until I couldn’t tell who it was anymore; I had to eliminate the training from it somewhat. It’s not my voice. I understand there’s a complex relationship there. I had to be, and will continue to be, aware of that. People that look like me have claimed ownership of other people’s cultures and work, and that has to be taken seriously. If this can open a conversation about that, then that’s the least you can do if you’re using the technology. You can at least acknowledge that there’s a history of people taking people’s voices, mimicking them, and selling them because they’re a different complexion to the people who invented the style. I don’t think that any of this sounds like any particular person. I hope that it doesn’t, because that’s something I worked really hard to avoid.
It would be so easy to do that: you’ve seen the AI deepfake voices. I think the voice and the image in particular are very hot things to start playing around with. The image of another person that isn’t you is a very sensitive thing to mess with. What else have you got to identify yourself with other than the way you feel, look, and sound? That’s the most intimate thing, and to go, “I could do that now, because I have this Google sheet I can click on?” We have to think of it more respectfully than that.
I’m sure some people will disagree with that, but I’ve not heard anything yet when I’ve spoken about the record. When I was making it, I thought, “Maybe people will be really against this.” But I hope that, when people hear the record, they hear that that’s not where it’s coming from. It doesn’t feel like it’s coming from a position of ownership to me. It’s a dialogue. It’s: “Here’s where we’re going, and I’m trying to make something beautiful and touching with it.” That’s all I could do. I didn’t want to make something “icy,” or “fresh,” or “modern.” “Alien” or “other-dimensional,” sure, but I wanted to make it as warm as I could. I’m not too saccharine, but there’s definitely a few moments on there where I’m like, “F*cking hell.” The inner Cure fan came out. [laughs] I hope that people don’t think of it as an ode to a dystopian technology.
It’s a reckoning.
Lee Gamble: Yeah. It’s here. Try and own it: make it your work. It’s just some paint. It’s just some acrylic in a tube. Until you do anything with it, it doesn’t mean anything. There’s this incredible number of images made by image generation systems. Hito Steyerl is really interesting on this topic in terms of images. She talks about the image in one of her pieces of writing, In Defense of the Poor Image. It’s a piece of writing about the digital image. It goes through all sorts of different processes; it’ll get compressed again and again, and the colors change, and it all falls apart. So you get these poor images even though we’re in these hi-tech times.
This writing goes into a defense of the poor image, and it explores what qualities can be found in it. In the piece of writing I did, I titled one of my chapters—where I talk about the ‘crappy’ nature of the voices I used—”In Defense of the Poor Voice.” I quote Steyerl there, of course. In her writing, she says, “The poor image embodies the afterlife of many former masterpieces of cinema and video art.” I read her talk about it very eloquently, and I said, “I get that.” If I just swap all of that for the voice rather than the .JPG, then it’s the same thing. I’m seeing the some of the same advantages in the lack of quality as she’d seen. That really resonated with me.
What’s next for you?
Lee Gamble: It’s mainly the live show. I’m working with Candela Capítan, a performance artist and choreographer. I looked at her work and thought it was amazing, in terms of how it relates to the body, repetition, alienation, and the gaze and technologized image. I wanted to think about how to present Models live, and I didn’t want to be onstage with a laptop. From day one, I wanted to bring a physical body back into space—but not mine. Humans gave me the information; I simulated and disembodied them; and, now, I ought to put them back in. A lot of the movement is informed by the ideas of cloning, mirrors, and how people replicate each others’ movements. We tried to get twins to perform in the live show.
It’ll premier at Unsound in Poland, and then in London. I wanted to develop the choreography around the idea of how things learn—sometimes they learn badly. Obviously, Candela has brought a huge amount to the work. I gave them a PDF of ideas, and they read it; I played the music to them, and they improvised. It was very moving: I’d never seen anyone do that to my work before. So, I thought, “I want to get on this.”
I’m DJing as regularly as I can, just banging it out and enjoying myself. I’m really enjoying painting. I’m hoping to do some installation work with these voices as well. I don’t know if it’s just me, but I feel like my life moves in ten-year cycles. I did therapy last year for the first time in my life, and that came up. It was really helpful. I was really glad that people pushed me to do it. I’m not against it, of course, but I come from a background where that wasn’t a thing. It took me a long time, but I needed it, and it was super helpful. One of the things that came up is the idea of ten-year cycles in my life. Diversions was 2012, and that was a moment where I did a bunch of other stuff. Now, it’s 2023, and there’s Models, and I’m painting again. Now, I’m ready to boot them glass ceilings as hard as I can for the next ten years.
After I’d finished Flush Real Pharynx for Hyperdub, I felt lost. Between the pandemic and losing these people, I said, “I’m gonna get grounded again.” Painting made me realize that I really like yellow! I didn’t know that. There’s things like that, where you say, “I like that color; I don’t like that color.” You can really transfer that. Not technically, but in terms of saying, “I don’t give a shit about all that—what does this sound like in my monitors? Do I like this?” I’d allow that to happen, and I wouldn’t question it any further. That’s how Models came about, I suppose.
I said, “I want to make a record that I really enjoy making.” It’s not like I’ve not done that before, but Flush Real Pharynx was such a big thing. It was really hard to do. I spoke to Steve [Goodman, AKA Kode9] at Hyperdub, and I said, “I’m making something, and it’s going to be really simple. I’m gonna do what I want; it’s a super emotional record.”
That seems really simple, but I’d lost my ability to do that. I’d lost the ability to judge, or believe, my own taste a little. There’s so many trends, sounds, and styles—things move so fast. And, all of a sudden, you’re like, “Where the f*ck have I gone?” The painting really helped with this. It helped me focus on what I wanted to do. Nobody’s gonna come up and say, “I think you should put some pink there.” It’s up to me; it starts and ends with my decisions. I approached Models in the same way.
You started this record in 2020, yeah?
Lee Gamble: I didn’t start the musical production then, but I did start to write the ideas for the record, and I started to correspond with Joseph, an AI programmer from America. I was aware of the possibilities that were available in AI. It’s moved a huge amount in three years. I was wary not to dive in too fast and get this rush of newness, write quickly and rely on the technology to hold the record up. We had back-and-forths where we were building vocal models and testing things out. I eventually told him, “I’m going to go away now; I need six months.” I needed to figure out how to make it my work rather than having it be the work of a piece of technology, which I have no interest in.
I worked with instrumentation, and I tried to figure out what would work with the voice and give it space. I was playing around with hooks. I wrote a couple of tracks without voice and added it in, but most of it was written around the vocals. I built it around earworms—phrases that I’d picked out from stuff that the network spat out.
Given I’m going down this wormhole, I’m aware that it’s problematic. I want to use this technology, but I want to be respectful of its problems and how it might sit in popular culture. So I wanted to take my time. I had more time than I’ve ever had to make a record. I read as much as I could, from as many different voices as I could, about how this would affect people: things like the racialized aspect of datasets, and the voice, too. It’s been mimicked a lot, and it’s been caricatured a lot, and it’s been taken and used by white culture, often in terms of American music.
I was conscious of these things, but it’s also such a tantalizing technology for an artist to not use. I’m still in a quandary with it. The fears around job replacement are totally valid; AI will definitely take jobs. It already has. I don’t think it’s going to take anyone’s singing away. I wanted to make sure this record didn’t feel like a technologized record. It’s about the idea that these aren’t real people, but it’s not suggesting the replacement of people. It’s about, “How much mood can I get out of this? How human can I make the technology?” How human does a singing voice feel when it’s disembodied? How emotional can a digital simulation feel? This is without ignoring what they are: simulations of people. I wanted to make sure I leaned into that, but I didn’t want it to be the thing that wrote the record.
I wanted to make sure I’m extremely conscious about it. The voices aren’t modeled after any particular person. It was, at first, but after a couple of years, I was like, “I can’t do this.” We built a system of four voices and kept feeding it into itself. Eventually, it got so far away from whatever it had been fed that it’s not anyone. The whole album, to me, feels like a trace of a pop record. I’d get phrases back and pick from them, and while I wouldn’t say I wrote these like pop songs, I allowed myself to drift into that space. This applies to the way it sounds, too—I wasn’t trying to make it sound crumbly, like Diversions, or super hi-res, like parts of Flush. With pop records that I love, even now, I’ll grab one track off one record, and I’ll hardly be able to play anything else. Everything else sounds terrible! I wanted to grab that energy: something like the aura of pop.
That’s what I was trying to think about. I wanted to make something that felt like you’re connecting to it in some way. The thing with these voices is that they don’t have a physical presence about them. I’m not trying to undermine the technology, of course. You feed them human stuff; all of AI is, fundamentally, human. But, in terms of their presence, they seem like they pop out of nowhere.
The first thing I did was build a model of my own voice, which is actually on the record, on A Million Pieces of You. I’m on there—well, it’s not me; it’s my deepfake. It’s taken a few years to wrangle with it as a cultural thing, which was as much the problem as making the record was. I wanted to make something that sounded like it was human, because, fundamentally, that’s what they are. They might come out of these incredible pieces of technology, but they’re human things. They’re a very weird entity. I don’t know what you’d call them. You have this idea of the uncanny valley, but it’s not that. It’s not like it’s trying to impersonate somebody; they seem to have their own character.
With this project, I kept returning to Autechre. You’re saying, “How can we take these ideas and put humanity back in?” I would say this and one of Autechre’s long-form live things are doing a similar thing.
Lee Gamble: When I hear Autechre’s current music—especially their live shows—it sounds like they’ve built a system that sounds like them. I think that’s a nice way of hearing their stuff. It’s a system that’s been honed over many years to sound that way and exist as that system. Models is not doing that in such a singular way, but the voice modeling aspect was definitely a similar kind of technological conversation. It doesn’t feel like a conversation when I’m playing a keyboard, although it is a piece of technology. I felt differently with this, because I wasn’t sure what I’d get back. It’s a matter of making a choice: do I like that or not? Do I put yellow on the canvas, or red, or something else?
That’s where the album emerged from. I’d run a plugin to find out what key those voices were in. I’d set everything around that key and write around them. The technology was giving me a lot of input, in a distant way. It’s not like I’m tweaking anything live: it’s a matter of taking stuff, and, an hour later, going “Wow, okay—you thought that? Interesting.” That’s where the magic was: saying, “Whoa! You reckon I should write something like that?” [laughs] There’s definitely that generative aspect to it. If we gave it reverb, it would turn the sounds into mush, and we’d have to start cleaning it up. I was literally mixing and mastering the sound of datasets.
I have a feeling that, with voice simulation technology, the endgame will be when you can’t tell the difference between AI voices and human voices. I don’t find that particularly interesting. The fact that they were scratchy and indecipherable and smudged, and had an Elizabeth Fraser, or “mumble” rap-esque non-language, where you don’t know what they’re saying but you’re getting emotion and narrative from it anyways—I was keen on exploring that. Some of the weird, fragmented, “mumble” rap stuff is huge, even though it’s really experimental. Once I got my head around that, I was happy to work with them in the state they’re in. It took tons of work to get them sounding like that, but, eventually, I was happy that they sounded like that rather than saying, “This sounds exactly like a person singing.”
That was only possible because the technology’s not great yet. In a few years, you might only have the really good models. I think you lose something in that, so I wanted to get it done now. In 10 years’ time, you might say, “Listen to how those voices sounded then! They were so bad!” The image things look completely different to how they did a year or two ago. I don’t know if the old ones are there anymore. It’s not like a piece of hardware that you can pick up again.
There’s an AI artist I found a few months ago who’s leaned into digital fragmentation and software errors. They took the bits that Midjourney doesn’t understand yet and said, “What do we do with the technology as it currently stands?” These aren’t flaws, necessarily.
Lee Gamble: That’s exactly what I did with Models. The crux, for me, was that detritus: this kind of childlike attempt to communicateThat kind of surplus value happens in technology all the time. But AI is so new, and it’s moving so fast. The first synthesized cello sound sounds terrible now. Now, if it’s mixed well, it’s hard to tell it’s fake. But that’s taken 50 years. AI’s moved at an incredible rate in the last three. The speed at which this technology evolved, in a few short years, is really scary. There’s no point in racing it, because it’s going to beat you. This stuff is owned by all the big corporations—they’ve got billions of pounds to throw at this.
With this record, I said, “I know damn well that if, in two years’ time, if I say this voice modeling system sounds like a human, it’ll be trash.” So, forget that—just make a record with the technology that’s in front of you, and lean into the limitations. I can’t compete with these companies; I’m just a guy who knows a brilliant programmer. We don’t have a team behind us. Forget them—they’re doing their thing, and I’ll lean into the problems that they’re trying to fix.
It’s the idea of making something that sounds like a moment.
Lee Gamble: That’s something I’ve been thinking about recently. You’ve got this idea about the hardcore continuum, for instance. It keeps shapeshifting and latching onto other things, and every now and again, you’ll get a burst, and you’ll get a new genre. I feel like that’s finished. I don’t think it’s not actually happening: I just think that since we’re so deep into social media and streaming, access to things has been leveled.
I can be on Spotify, and I could be in a playlist with Beyoncé. That wouldn’t have happened before. If I was 10 years old now, I don’t have this reference to date or time: here’s jazz, here’s prog, here’s krautrock, here’s techno. It’s just all there. In a sense, that’s really amazing. I’ve really enjoyed Lil Yachty’s new record this year, and it seems to lean into that: there are points where it sounds like Pink Floyd, and then it’ll sound super modern, and then it sounds like the ’70s, and then it sounds like old-school computer games, and then it sounds like old hip-hop. It sounds like now.
I don’t think younger kids give a shit—and I don’t think they should—if something was made in 1970 or 2023. Who cares? If someone’s playing a track in a club, and you’ve got 2,000 people dancing to it, who cares if it’s from 1994? That’s what I mean about this idea of a continuum. It’s all been piledrived into a wall and splattered. It comes with its issues; I get that. But I think that time splattering is a good thing.
The question about what is, or sounds, “modern” or “old” gets mushier over time. If you’ve never heard something before, it’s new. I heard someone on NTS the other day—a working-class person from London—playing Steve Reich. They said, “I’m going down this wormhole with Steve Reich.” That’s amazing. That stuff is historically of the “academy,” or the “middle classes,” or the “conservatoire,” but there’s also a clear history from Indian music to minimalism. It all gets constantly re-contextualised in the streaming era.
I went down a YouTube wormhole with 100 gecs’ stuff. I’m not a stan, but it crossed me in the same way. It’s a mixture of things that sounds exactly like now. It doesn’t matter if I like it aesthetically. As a cultural object, you’ve taken this on and just ran with it! You don’t give a f*ck. I don’t know how old they are, but they’re younger than me. I think younger people aren’t trying to make a point of that; that’s just how it is. [laughs] There’s no reason to have this linear time where everything gets discarded—”Oh, it’s old.” I always thought that was a bit pointless.
Also, to flip towards the future: people are always like, “This sounds like the future.” What does that mean? I’m not trying to be facetious; I honestly don’t know what that means.
I’m definitely guilty of using that terminology.
Lee Gamble: You’re a writer, so it makes sense—it’s a good semantic thing. People go “Ah! I get it.” But what do you think the person’s gonna think it sounds like?
I try to use it to say, “This person is going in a really exciting direction, and maybe this is a possible vision for the future.”
Lee Gamble: That Lil Yachty record sounds like now. It might sound old in a year. But, for now, it sounds like the present. It’s that cross-pollination—corecore shit on TikTok, and mashups. I keep hearing the word “Dada” again. It feels like vertical collage. I did a piece recently which was performed at Berlin Atonal; they did a Xenakis X100 thing. I’m a huge admirer of Xenakis‘ work. I wrote this in memory of him, so I thought, “What would he be doing now?” You’re damn well sure he’s not doing what he was doing then. He was doing stochastic computer synthesis methods in the ’90s when very few people were doing that in the world. If he was a young person in 2023, I doubt he’d be writing for solo timpani. He’s a techno person, in a sense. I doubt he’d want to becalled a techno guy, but f*ck it, that’s fine.
It’s that idea of possible futures: what kind of future are you offering?
Lee Gamble: There’s definitely some of that aspect to Models. Take the Flush Real Pharynx series I did. That was me trying to come to terms with the culture clash we’re talking about: multiple screens at the same time. It’s a kind of collage. That wasn’t the defining thing for the record, but I had that in my head. I wanted that forefront in my brain while I was writing it. That’s why I called it Flush Real Pharynx: it’s about flushing your voice away and being bombarded with stuff.
With Models, I said, “F*ck that.” That’s really bad for our mental health. I’m gonna bounce, and I’m gonna make something more emotionally driven, or soulful, rather than a TikTok kind of thing. I’m not sure how long people can take that, just as a flesh-and-bone kind of thing. As brains, you and I are not that different to how we were 2,000 years ago. It’s having an effect on people’s mental health. As a culture, we try to come to terms with it. That’s what I did with Flush, and I injured myself by doing that, so I wanted to soothe myself a bit and switch it all off.
I’m not qualified to explain this, but in terms of how things would affect my music: previously it was cities. If you grew up in Detroit, you might be into one thing, and if you grew up in New York, you might be into something else, and if you grew up in Birmingham, it might be another. Now, it’s the internet. If you’re into something, you can find it. There are so many positive things about it, but at this point in my life, it’s better for my mental health for me to switch it off when I’m going to make something. It’s taken me a few years to learn that, but it’s fine. Sometimes things take a while to work out. If I were to make another record, I’d want to hire a studio in a barn somewhere on a mountain for three months.
You started Models when AI was prehistoric compared to where it is now, and when it wasn’t really a public conversation. What’s it been like to watch that evolution in real time?
Lee Gamble: Neural networks are from the ’50s, so AI is not new. But data happened. They didn’t have enough data sets to make a good neural network in the ’50s. Even when we started doing the voice modeling stuff, you’d need a bunch of stuff. Eventually, it’d get down to 60 seconds of something that you could work with. That was with two people and a normal computer; it’s not with Meta or Google or whatever. It was obvious to me that it was going to accelerate at rate of knots.
It’s a little bit of brain, in a sense, that can learn how to do something. It’s a simulation technology. Anything you experience is because you have a brain, fundamentally. If the brain is allowing everything to be knowable, then a neural network can try any of that. So, of course it’s going to be everywhere. We’ve seen it first in image-making, because most online communities are built around images. I knew it was going to be quick, so I had to say, “We are where we are, and let’s roll with that version.”
When I started writing [Models], I bowed out of the AI conversation. I started looking at it again recently, because I needed a break from it all. I read the other day about someone who had been in a coma for 20 years, and they lost their voice. They came out of the coma, and they’ve used AI to train them to have a voice again, which is such a Models thing. I think it’ll become ubiquitous. Then it’ll level off, and that’s when the interesting things will happen.
I wanted this to sound like a record that you’d make when AI isn’t a “thing” anymore. It’s a bit like early Herbie Hancock records. Everybody wanted to know what synthesizer he was using. 20 years later, nobody’s asking; he just plays. I wanted to place Models there somehow. I realize this is coming close to making “futuristic” music. [laughs] Right now, I’m overwhelmed by the shininess and potential of this technology. In years’ time, people won’t think of it like that.
When you put something like this out, you get people acting as though you’re okay with all that happening. The problem, for me, is that AI is already in every part of culture. So saying, “I’m not going to use it because it’s bad”—it’s not like not using Amazon because they’re scumbags. But we can’t not use AI. Fundamentally, it’s built on our collective labor, art, and time. I don’t want that to sound like a cop-out; I’m no libertarian. But, at a certain point, you’re going to have to engage with it. When that happens, it’ll either be unknowingly or with a bit of knowledge. The only way for me to offset that was to say, “Educate yourself on it, and hear some voices that aren’t of a similar background.”
I’m not here to make an excuse as to why I’d use it. I’m not out there saying, “This is amazing; we don’t need singers anymore.” I’m an artist, and I wouldn’t even begin to think that. If I can make something that can resonate emotionally with people, like I hope this record can, then maybe that’s okay. If I’m able to make people think about how they can use the technology in their way outside of any corporate dystopias, then maybe that’s okay. I’m not expecting to be able to make a record and say something about AI in general. There are things that we should be aware of and use as artists, and there are other things we should be wary of and at least talk about.
For example, there’s the protection of people’s voices. If I fed the first model one person’s voice and said, “Okay, that’s it,” that wouldn’t have been enough. I had to process it until I couldn’t tell who it was anymore; I had to eliminate the training from it somewhat. It’s not my voice. I understand there’s a complex relationship there. I had to be, and will continue to be, aware of that. People that look like me have claimed ownership of other people’s cultures and work, and that has to be taken seriously. If this can open a conversation about that, then that’s the least you can do if you’re using the technology. You can at least acknowledge that there’s a history of people taking people’s voices, mimicking them, and selling them because they’re a different complexion to the people who invented the style. I don’t think that any of this sounds like any particular person. I hope that it doesn’t, because that’s something I worked really hard to avoid.
It would be so easy to do that: you’ve seen the AI deepfake voices. I think the voice and the image in particular are very hot things to start playing around with. The image of another person that isn’t you is a very sensitive thing to mess with. What else have you got to identify yourself with other than the way you feel, look, and sound? That’s the most intimate thing, and to go, “I could do that now, because I have this Google sheet I can click on?” We have to think of it more respectfully than that.
I’m sure some people will disagree with that, but I’ve not heard anything yet when I’ve spoken about the record. When I was making it, I thought, “Maybe people will be really against this.” But I hope that, when people hear the record, they hear that that’s not where it’s coming from. It doesn’t feel like it’s coming from a position of ownership to me. It’s a dialogue. It’s: “Here’s where we’re going, and I’m trying to make something beautiful and touching with it.” That’s all I could do. I didn’t want to make something “icy,” or “fresh,” or “modern.” “Alien” or “other-dimensional,” sure, but I wanted to make it as warm as I could. I’m not too saccharine, but there’s definitely a few moments on there where I’m like, “F*cking hell.” The inner Cure fan came out. [laughs] I hope that people don’t think of it as an ode to a dystopian technology.
It’s a reckoning.
Lee Gamble: Yeah. It’s here. Try and own it: make it your work. It’s just some paint. It’s just some acrylic in a tube. Until you do anything with it, it doesn’t mean anything. There’s this incredible number of images made by image generation systems. Hito Steyerl is really interesting on this topic in terms of images. She talks about the image in one of her pieces of writing, In Defense of the Poor Image. It’s a piece of writing about the digital image. It goes through all sorts of different processes; it’ll get compressed again and again, and the colors change, and it all falls apart. So you get these poor images even though we’re in these hi-tech times.
This writing goes into a defense of the poor image, and it explores what qualities can be found in it. In the piece of writing I did, I titled one of my chapters—where I talk about the ‘crappy’ nature of the voices I used—”In Defense of the Poor Voice.” I quote Steyerl there, of course. In her writing, she says, “The poor image embodies the afterlife of many former masterpieces of cinema and video art.” I read her talk about it very eloquently, and I said, “I get that.” If I just swap all of that for the voice rather than the .JPG, then it’s the same thing. I’m seeing the some of the same advantages in the lack of quality as she’d seen. That really resonated with me.
What’s next for you?
Lee Gamble: It’s mainly the live show. I’m working with Candela Capítan, a performance artist and choreographer. I looked at her work and thought it was amazing, in terms of how it relates to the body, repetition, alienation, and the gaze and technologized image. I wanted to think about how to present Models live, and I didn’t want to be onstage with a laptop. From day one, I wanted to bring a physical body back into space—but not mine. Humans gave me the information; I simulated and disembodied them; and, now, I ought to put them back in. A lot of the movement is informed by the ideas of cloning, mirrors, and how people replicate each others’ movements. We tried to get twins to perform in the live show.
It’ll premier at Unsound in Poland, and then in London. I wanted to develop the choreography around the idea of how things learn—sometimes they learn badly. Obviously, Candela has brought a huge amount to the work. I gave them a PDF of ideas, and they read it; I played the music to them, and they improvised. It was very moving: I’d never seen anyone do that to my work before. So, I thought, “I want to get on this.”
I’m DJing as regularly as I can, just banging it out and enjoying myself. I’m really enjoying painting. I’m hoping to do some installation work with these voices as well. I don’t know if it’s just me, but I feel like my life moves in ten-year cycles. I did therapy last year for the first time in my life, and that came up. It was really helpful. I was really glad that people pushed me to do it. I’m not against it, of course, but I come from a background where that wasn’t a thing. It took me a long time, but I needed it, and it was super helpful. One of the things that came up is the idea of ten-year cycles in my life. Diversions was 2012, and that was a moment where I did a bunch of other stuff. Now, it’s 2023, and there’s Models, and I’m painting again. Now, I’m ready to boot them glass ceilings as hard as I can for the next ten years.
I worked with instrumentation, and I tried to figure out what would work with the voice and give it space. I was playing around with hooks. I wrote a couple of tracks without voice and added it in, but most of it was written around the vocals. I built it around earworms—phrases that I’d picked out from stuff that the network spat out.
Given I’m going down this wormhole, I’m aware that it’s problematic. I want to use this technology, but I want to be respectful of its problems and how it might sit in popular culture. So I wanted to take my time. I had more time than I’ve ever had to make a record. I read as much as I could, from as many different voices as I could, about how this would affect people: things like the racialized aspect of datasets, and the voice, too. It’s been mimicked a lot, and it’s been caricatured a lot, and it’s been taken and used by white culture, often in terms of American music.
I was conscious of these things, but it’s also such a tantalizing technology for an artist to not use. I’m still in a quandary with it. The fears around job replacement are totally valid; AI will definitely take jobs. It already has. I don’t think it’s going to take anyone’s singing away. I wanted to make sure this record didn’t feel like a technologized record. It’s about the idea that these aren’t real people, but it’s not suggesting the replacement of people. It’s about, “How much mood can I get out of this? How human can I make the technology?” How human does a singing voice feel when it’s disembodied? How emotional can a digital simulation feel? This is without ignoring what they are: simulations of people. I wanted to make sure I leaned into that, but I didn’t want it to be the thing that wrote the record.
I wanted to make sure I’m extremely conscious about it. The voices aren’t modeled after any particular person. It was, at first, but after a couple of years, I was like, “I can’t do this.” We built a system of four voices and kept feeding it into itself. Eventually, it got so far away from whatever it had been fed that it’s not anyone. The whole album, to me, feels like a trace of a pop record. I’d get phrases back and pick from them, and while I wouldn’t say I wrote these like pop songs, I allowed myself to drift into that space. This applies to the way it sounds, too—I wasn’t trying to make it sound crumbly, like Diversions, or super hi-res, like parts of Flush. With pop records that I love, even now, I’ll grab one track off one record, and I’ll hardly be able to play anything else. Everything else sounds terrible! I wanted to grab that energy: something like the aura of pop.
That’s what I was trying to think about. I wanted to make something that felt like you’re connecting to it in some way. The thing with these voices is that they don’t have a physical presence about them. I’m not trying to undermine the technology, of course. You feed them human stuff; all of AI is, fundamentally, human. But, in terms of their presence, they seem like they pop out of nowhere.
The first thing I did was build a model of my own voice, which is actually on the record, on A Million Pieces of You. I’m on there—well, it’s not me; it’s my deepfake. It’s taken a few years to wrangle with it as a cultural thing, which was as much the problem as making the record was. I wanted to make something that sounded like it was human, because, fundamentally, that’s what they are. They might come out of these incredible pieces of technology, but they’re human things. They’re a very weird entity. I don’t know what you’d call them. You have this idea of the uncanny valley, but it’s not that. It’s not like it’s trying to impersonate somebody; they seem to have their own character.
With this project, I kept returning to Autechre. You’re saying, “How can we take these ideas and put humanity back in?” I would say this and one of Autechre’s long-form live things are doing a similar thing.
Lee Gamble: When I hear Autechre’s current music—especially their live shows—it sounds like they’ve built a system that sounds like them. I think that’s a nice way of hearing their stuff. It’s a system that’s been honed over many years to sound that way and exist as that system. Models is not doing that in such a singular way, but the voice modeling aspect was definitely a similar kind of technological conversation. It doesn’t feel like a conversation when I’m playing a keyboard, although it is a piece of technology. I felt differently with this, because I wasn’t sure what I’d get back. It’s a matter of making a choice: do I like that or not? Do I put yellow on the canvas, or red, or something else?
That’s where the album emerged from. I’d run a plugin to find out what key those voices were in. I’d set everything around that key and write around them. The technology was giving me a lot of input, in a distant way. It’s not like I’m tweaking anything live: it’s a matter of taking stuff, and, an hour later, going “Wow, okay—you thought that? Interesting.” That’s where the magic was: saying, “Whoa! You reckon I should write something like that?” [laughs] There’s definitely that generative aspect to it. If we gave it reverb, it would turn the sounds into mush, and we’d have to start cleaning it up. I was literally mixing and mastering the sound of datasets.
I have a feeling that, with voice simulation technology, the endgame will be when you can’t tell the difference between AI voices and human voices. I don’t find that particularly interesting. The fact that they were scratchy and indecipherable and smudged, and had an Elizabeth Fraser, or “mumble” rap-esque non-language, where you don’t know what they’re saying but you’re getting emotion and narrative from it anyways—I was keen on exploring that. Some of the weird, fragmented, “mumble” rap stuff is huge, even though it’s really experimental. Once I got my head around that, I was happy to work with them in the state they’re in. It took tons of work to get them sounding like that, but, eventually, I was happy that they sounded like that rather than saying, “This sounds exactly like a person singing.”
That was only possible because the technology’s not great yet. In a few years, you might only have the really good models. I think you lose something in that, so I wanted to get it done now. In 10 years’ time, you might say, “Listen to how those voices sounded then! They were so bad!” The image things look completely different to how they did a year or two ago. I don’t know if the old ones are there anymore. It’s not like a piece of hardware that you can pick up again.
There’s an AI artist I found a few months ago who’s leaned into digital fragmentation and software errors. They took the bits that Midjourney doesn’t understand yet and said, “What do we do with the technology as it currently stands?” These aren’t flaws, necessarily.
Lee Gamble: That’s exactly what I did with Models. The crux, for me, was that detritus: this kind of childlike attempt to communicateThat kind of surplus value happens in technology all the time. But AI is so new, and it’s moving so fast. The first synthesized cello sound sounds terrible now. Now, if it’s mixed well, it’s hard to tell it’s fake. But that’s taken 50 years. AI’s moved at an incredible rate in the last three. The speed at which this technology evolved, in a few short years, is really scary. There’s no point in racing it, because it’s going to beat you. This stuff is owned by all the big corporations—they’ve got billions of pounds to throw at this.
With this record, I said, “I know damn well that if, in two years’ time, if I say this voice modeling system sounds like a human, it’ll be trash.” So, forget that—just make a record with the technology that’s in front of you, and lean into the limitations. I can’t compete with these companies; I’m just a guy who knows a brilliant programmer. We don’t have a team behind us. Forget them—they’re doing their thing, and I’ll lean into the problems that they’re trying to fix.
It’s the idea of making something that sounds like a moment.
Lee Gamble: That’s something I’ve been thinking about recently. You’ve got this idea about the hardcore continuum, for instance. It keeps shapeshifting and latching onto other things, and every now and again, you’ll get a burst, and you’ll get a new genre. I feel like that’s finished. I don’t think it’s not actually happening: I just think that since we’re so deep into social media and streaming, access to things has been leveled.
I can be on Spotify, and I could be in a playlist with Beyoncé. That wouldn’t have happened before. If I was 10 years old now, I don’t have this reference to date or time: here’s jazz, here’s prog, here’s krautrock, here’s techno. It’s just all there. In a sense, that’s really amazing. I’ve really enjoyed Lil Yachty’s new record this year, and it seems to lean into that: there are points where it sounds like Pink Floyd, and then it’ll sound super modern, and then it sounds like the ’70s, and then it sounds like old-school computer games, and then it sounds like old hip-hop. It sounds like now.
I don’t think younger kids give a shit—and I don’t think they should—if something was made in 1970 or 2023. Who cares? If someone’s playing a track in a club, and you’ve got 2,000 people dancing to it, who cares if it’s from 1994? That’s what I mean about this idea of a continuum. It’s all been piledrived into a wall and splattered. It comes with its issues; I get that. But I think that time splattering is a good thing.
The question about what is, or sounds, “modern” or “old” gets mushier over time. If you’ve never heard something before, it’s new. I heard someone on NTS the other day—a working-class person from London—playing Steve Reich. They said, “I’m going down this wormhole with Steve Reich.” That’s amazing. That stuff is historically of the “academy,” or the “middle classes,” or the “conservatoire,” but there’s also a clear history from Indian music to minimalism. It all gets constantly re-contextualised in the streaming era.
I went down a YouTube wormhole with 100 gecs’ stuff. I’m not a stan, but it crossed me in the same way. It’s a mixture of things that sounds exactly like now. It doesn’t matter if I like it aesthetically. As a cultural object, you’ve taken this on and just ran with it! You don’t give a f*ck. I don’t know how old they are, but they’re younger than me. I think younger people aren’t trying to make a point of that; that’s just how it is. [laughs] There’s no reason to have this linear time where everything gets discarded—”Oh, it’s old.” I always thought that was a bit pointless.
Also, to flip towards the future: people are always like, “This sounds like the future.” What does that mean? I’m not trying to be facetious; I honestly don’t know what that means.
I’m definitely guilty of using that terminology.
Lee Gamble: You’re a writer, so it makes sense—it’s a good semantic thing. People go “Ah! I get it.” But what do you think the person’s gonna think it sounds like?
I try to use it to say, “This person is going in a really exciting direction, and maybe this is a possible vision for the future.”
Lee Gamble: That Lil Yachty record sounds like now. It might sound old in a year. But, for now, it sounds like the present. It’s that cross-pollination—corecore shit on TikTok, and mashups. I keep hearing the word “Dada” again. It feels like vertical collage. I did a piece recently which was performed at Berlin Atonal; they did a Xenakis X100 thing. I’m a huge admirer of Xenakis‘ work. I wrote this in memory of him, so I thought, “What would he be doing now?” You’re damn well sure he’s not doing what he was doing then. He was doing stochastic computer synthesis methods in the ’90s when very few people were doing that in the world. If he was a young person in 2023, I doubt he’d be writing for solo timpani. He’s a techno person, in a sense. I doubt he’d want to becalled a techno guy, but f*ck it, that’s fine.
It’s that idea of possible futures: what kind of future are you offering?
Lee Gamble: There’s definitely some of that aspect to Models. Take the Flush Real Pharynx series I did. That was me trying to come to terms with the culture clash we’re talking about: multiple screens at the same time. It’s a kind of collage. That wasn’t the defining thing for the record, but I had that in my head. I wanted that forefront in my brain while I was writing it. That’s why I called it Flush Real Pharynx: it’s about flushing your voice away and being bombarded with stuff.
With Models, I said, “F*ck that.” That’s really bad for our mental health. I’m gonna bounce, and I’m gonna make something more emotionally driven, or soulful, rather than a TikTok kind of thing. I’m not sure how long people can take that, just as a flesh-and-bone kind of thing. As brains, you and I are not that different to how we were 2,000 years ago. It’s having an effect on people’s mental health. As a culture, we try to come to terms with it. That’s what I did with Flush, and I injured myself by doing that, so I wanted to soothe myself a bit and switch it all off.
I’m not qualified to explain this, but in terms of how things would affect my music: previously it was cities. If you grew up in Detroit, you might be into one thing, and if you grew up in New York, you might be into something else, and if you grew up in Birmingham, it might be another. Now, it’s the internet. If you’re into something, you can find it. There are so many positive things about it, but at this point in my life, it’s better for my mental health for me to switch it off when I’m going to make something. It’s taken me a few years to learn that, but it’s fine. Sometimes things take a while to work out. If I were to make another record, I’d want to hire a studio in a barn somewhere on a mountain for three months.
You started Models when AI was prehistoric compared to where it is now, and when it wasn’t really a public conversation. What’s it been like to watch that evolution in real time?
Lee Gamble: Neural networks are from the ’50s, so AI is not new. But data happened. They didn’t have enough data sets to make a good neural network in the ’50s. Even when we started doing the voice modeling stuff, you’d need a bunch of stuff. Eventually, it’d get down to 60 seconds of something that you could work with. That was with two people and a normal computer; it’s not with Meta or Google or whatever. It was obvious to me that it was going to accelerate at rate of knots.
It’s a little bit of brain, in a sense, that can learn how to do something. It’s a simulation technology. Anything you experience is because you have a brain, fundamentally. If the brain is allowing everything to be knowable, then a neural network can try any of that. So, of course it’s going to be everywhere. We’ve seen it first in image-making, because most online communities are built around images. I knew it was going to be quick, so I had to say, “We are where we are, and let’s roll with that version.”
When I started writing [Models], I bowed out of the AI conversation. I started looking at it again recently, because I needed a break from it all. I read the other day about someone who had been in a coma for 20 years, and they lost their voice. They came out of the coma, and they’ve used AI to train them to have a voice again, which is such a Models thing. I think it’ll become ubiquitous. Then it’ll level off, and that’s when the interesting things will happen.
I wanted this to sound like a record that you’d make when AI isn’t a “thing” anymore. It’s a bit like early Herbie Hancock records. Everybody wanted to know what synthesizer he was using. 20 years later, nobody’s asking; he just plays. I wanted to place Models there somehow. I realize this is coming close to making “futuristic” music. [laughs] Right now, I’m overwhelmed by the shininess and potential of this technology. In years’ time, people won’t think of it like that.
When you put something like this out, you get people acting as though you’re okay with all that happening. The problem, for me, is that AI is already in every part of culture. So saying, “I’m not going to use it because it’s bad”—it’s not like not using Amazon because they’re scumbags. But we can’t not use AI. Fundamentally, it’s built on our collective labor, art, and time. I don’t want that to sound like a cop-out; I’m no libertarian. But, at a certain point, you’re going to have to engage with it. When that happens, it’ll either be unknowingly or with a bit of knowledge. The only way for me to offset that was to say, “Educate yourself on it, and hear some voices that aren’t of a similar background.”
I’m not here to make an excuse as to why I’d use it. I’m not out there saying, “This is amazing; we don’t need singers anymore.” I’m an artist, and I wouldn’t even begin to think that. If I can make something that can resonate emotionally with people, like I hope this record can, then maybe that’s okay. If I’m able to make people think about how they can use the technology in their way outside of any corporate dystopias, then maybe that’s okay. I’m not expecting to be able to make a record and say something about AI in general. There are things that we should be aware of and use as artists, and there are other things we should be wary of and at least talk about.
For example, there’s the protection of people’s voices. If I fed the first model one person’s voice and said, “Okay, that’s it,” that wouldn’t have been enough. I had to process it until I couldn’t tell who it was anymore; I had to eliminate the training from it somewhat. It’s not my voice. I understand there’s a complex relationship there. I had to be, and will continue to be, aware of that. People that look like me have claimed ownership of other people’s cultures and work, and that has to be taken seriously. If this can open a conversation about that, then that’s the least you can do if you’re using the technology. You can at least acknowledge that there’s a history of people taking people’s voices, mimicking them, and selling them because they’re a different complexion to the people who invented the style. I don’t think that any of this sounds like any particular person. I hope that it doesn’t, because that’s something I worked really hard to avoid.
It would be so easy to do that: you’ve seen the AI deepfake voices. I think the voice and the image in particular are very hot things to start playing around with. The image of another person that isn’t you is a very sensitive thing to mess with. What else have you got to identify yourself with other than the way you feel, look, and sound? That’s the most intimate thing, and to go, “I could do that now, because I have this Google sheet I can click on?” We have to think of it more respectfully than that.
I’m sure some people will disagree with that, but I’ve not heard anything yet when I’ve spoken about the record. When I was making it, I thought, “Maybe people will be really against this.” But I hope that, when people hear the record, they hear that that’s not where it’s coming from. It doesn’t feel like it’s coming from a position of ownership to me. It’s a dialogue. It’s: “Here’s where we’re going, and I’m trying to make something beautiful and touching with it.” That’s all I could do. I didn’t want to make something “icy,” or “fresh,” or “modern.” “Alien” or “other-dimensional,” sure, but I wanted to make it as warm as I could. I’m not too saccharine, but there’s definitely a few moments on there where I’m like, “F*cking hell.” The inner Cure fan came out. [laughs] I hope that people don’t think of it as an ode to a dystopian technology.
It’s a reckoning.
Lee Gamble: Yeah. It’s here. Try and own it: make it your work. It’s just some paint. It’s just some acrylic in a tube. Until you do anything with it, it doesn’t mean anything. There’s this incredible number of images made by image generation systems. Hito Steyerl is really interesting on this topic in terms of images. She talks about the image in one of her pieces of writing, In Defense of the Poor Image. It’s a piece of writing about the digital image. It goes through all sorts of different processes; it’ll get compressed again and again, and the colors change, and it all falls apart. So you get these poor images even though we’re in these hi-tech times.
This writing goes into a defense of the poor image, and it explores what qualities can be found in it. In the piece of writing I did, I titled one of my chapters—where I talk about the ‘crappy’ nature of the voices I used—”In Defense of the Poor Voice.” I quote Steyerl there, of course. In her writing, she says, “The poor image embodies the afterlife of many former masterpieces of cinema and video art.” I read her talk about it very eloquently, and I said, “I get that.” If I just swap all of that for the voice rather than the .JPG, then it’s the same thing. I’m seeing the some of the same advantages in the lack of quality as she’d seen. That really resonated with me.
What’s next for you?
Lee Gamble: It’s mainly the live show. I’m working with Candela Capítan, a performance artist and choreographer. I looked at her work and thought it was amazing, in terms of how it relates to the body, repetition, alienation, and the gaze and technologized image. I wanted to think about how to present Models live, and I didn’t want to be onstage with a laptop. From day one, I wanted to bring a physical body back into space—but not mine. Humans gave me the information; I simulated and disembodied them; and, now, I ought to put them back in. A lot of the movement is informed by the ideas of cloning, mirrors, and how people replicate each others’ movements. We tried to get twins to perform in the live show.
It’ll premier at Unsound in Poland, and then in London. I wanted to develop the choreography around the idea of how things learn—sometimes they learn badly. Obviously, Candela has brought a huge amount to the work. I gave them a PDF of ideas, and they read it; I played the music to them, and they improvised. It was very moving: I’d never seen anyone do that to my work before. So, I thought, “I want to get on this.”
I’m DJing as regularly as I can, just banging it out and enjoying myself. I’m really enjoying painting. I’m hoping to do some installation work with these voices as well. I don’t know if it’s just me, but I feel like my life moves in ten-year cycles. I did therapy last year for the first time in my life, and that came up. It was really helpful. I was really glad that people pushed me to do it. I’m not against it, of course, but I come from a background where that wasn’t a thing. It took me a long time, but I needed it, and it was super helpful. One of the things that came up is the idea of ten-year cycles in my life. Diversions was 2012, and that was a moment where I did a bunch of other stuff. Now, it’s 2023, and there’s Models, and I’m painting again. Now, I’m ready to boot them glass ceilings as hard as I can for the next ten years.
That’s where the album emerged from. I’d run a plugin to find out what key those voices were in. I’d set everything around that key and write around them. The technology was giving me a lot of input, in a distant way. It’s not like I’m tweaking anything live: it’s a matter of taking stuff, and, an hour later, going “Wow, okay—you thought that? Interesting.” That’s where the magic was: saying, “Whoa! You reckon I should write something like that?” [laughs] There’s definitely that generative aspect to it. If we gave it reverb, it would turn the sounds into mush, and we’d have to start cleaning it up. I was literally mixing and mastering the sound of datasets.
I have a feeling that, with voice simulation technology, the endgame will be when you can’t tell the difference between AI voices and human voices. I don’t find that particularly interesting. The fact that they were scratchy and indecipherable and smudged, and had an Elizabeth Fraser, or “mumble” rap-esque non-language, where you don’t know what they’re saying but you’re getting emotion and narrative from it anyways—I was keen on exploring that. Some of the weird, fragmented, “mumble” rap stuff is huge, even though it’s really experimental. Once I got my head around that, I was happy to work with them in the state they’re in. It took tons of work to get them sounding like that, but, eventually, I was happy that they sounded like that rather than saying, “This sounds exactly like a person singing.”
That was only possible because the technology’s not great yet. In a few years, you might only have the really good models. I think you lose something in that, so I wanted to get it done now. In 10 years’ time, you might say, “Listen to how those voices sounded then! They were so bad!” The image things look completely different to how they did a year or two ago. I don’t know if the old ones are there anymore. It’s not like a piece of hardware that you can pick up again.
There’s an AI artist I found a few months ago who’s leaned into digital fragmentation and software errors. They took the bits that Midjourney doesn’t understand yet and said, “What do we do with the technology as it currently stands?” These aren’t flaws, necessarily.
Lee Gamble: That’s exactly what I did with Models. The crux, for me, was that detritus: this kind of childlike attempt to communicateThat kind of surplus value happens in technology all the time. But AI is so new, and it’s moving so fast. The first synthesized cello sound sounds terrible now. Now, if it’s mixed well, it’s hard to tell it’s fake. But that’s taken 50 years. AI’s moved at an incredible rate in the last three. The speed at which this technology evolved, in a few short years, is really scary. There’s no point in racing it, because it’s going to beat you. This stuff is owned by all the big corporations—they’ve got billions of pounds to throw at this.
With this record, I said, “I know damn well that if, in two years’ time, if I say this voice modeling system sounds like a human, it’ll be trash.” So, forget that—just make a record with the technology that’s in front of you, and lean into the limitations. I can’t compete with these companies; I’m just a guy who knows a brilliant programmer. We don’t have a team behind us. Forget them—they’re doing their thing, and I’ll lean into the problems that they’re trying to fix.
It’s the idea of making something that sounds like a moment.
Lee Gamble: That’s something I’ve been thinking about recently. You’ve got this idea about the hardcore continuum, for instance. It keeps shapeshifting and latching onto other things, and every now and again, you’ll get a burst, and you’ll get a new genre. I feel like that’s finished. I don’t think it’s not actually happening: I just think that since we’re so deep into social media and streaming, access to things has been leveled.
I can be on Spotify, and I could be in a playlist with Beyoncé. That wouldn’t have happened before. If I was 10 years old now, I don’t have this reference to date or time: here’s jazz, here’s prog, here’s krautrock, here’s techno. It’s just all there. In a sense, that’s really amazing. I’ve really enjoyed Lil Yachty’s new record this year, and it seems to lean into that: there are points where it sounds like Pink Floyd, and then it’ll sound super modern, and then it sounds like the ’70s, and then it sounds like old-school computer games, and then it sounds like old hip-hop. It sounds like now.
I don’t think younger kids give a shit—and I don’t think they should—if something was made in 1970 or 2023. Who cares? If someone’s playing a track in a club, and you’ve got 2,000 people dancing to it, who cares if it’s from 1994? That’s what I mean about this idea of a continuum. It’s all been piledrived into a wall and splattered. It comes with its issues; I get that. But I think that time splattering is a good thing.
The question about what is, or sounds, “modern” or “old” gets mushier over time. If you’ve never heard something before, it’s new. I heard someone on NTS the other day—a working-class person from London—playing Steve Reich. They said, “I’m going down this wormhole with Steve Reich.” That’s amazing. That stuff is historically of the “academy,” or the “middle classes,” or the “conservatoire,” but there’s also a clear history from Indian music to minimalism. It all gets constantly re-contextualised in the streaming era.
I went down a YouTube wormhole with 100 gecs’ stuff. I’m not a stan, but it crossed me in the same way. It’s a mixture of things that sounds exactly like now. It doesn’t matter if I like it aesthetically. As a cultural object, you’ve taken this on and just ran with it! You don’t give a f*ck. I don’t know how old they are, but they’re younger than me. I think younger people aren’t trying to make a point of that; that’s just how it is. [laughs] There’s no reason to have this linear time where everything gets discarded—”Oh, it’s old.” I always thought that was a bit pointless.
Also, to flip towards the future: people are always like, “This sounds like the future.” What does that mean? I’m not trying to be facetious; I honestly don’t know what that means.
I’m definitely guilty of using that terminology.
Lee Gamble: You’re a writer, so it makes sense—it’s a good semantic thing. People go “Ah! I get it.” But what do you think the person’s gonna think it sounds like?
I try to use it to say, “This person is going in a really exciting direction, and maybe this is a possible vision for the future.”
Lee Gamble: That Lil Yachty record sounds like now. It might sound old in a year. But, for now, it sounds like the present. It’s that cross-pollination—corecore shit on TikTok, and mashups. I keep hearing the word “Dada” again. It feels like vertical collage. I did a piece recently which was performed at Berlin Atonal; they did a Xenakis X100 thing. I’m a huge admirer of Xenakis‘ work. I wrote this in memory of him, so I thought, “What would he be doing now?” You’re damn well sure he’s not doing what he was doing then. He was doing stochastic computer synthesis methods in the ’90s when very few people were doing that in the world. If he was a young person in 2023, I doubt he’d be writing for solo timpani. He’s a techno person, in a sense. I doubt he’d want to becalled a techno guy, but f*ck it, that’s fine.
It’s that idea of possible futures: what kind of future are you offering?
Lee Gamble: There’s definitely some of that aspect to Models. Take the Flush Real Pharynx series I did. That was me trying to come to terms with the culture clash we’re talking about: multiple screens at the same time. It’s a kind of collage. That wasn’t the defining thing for the record, but I had that in my head. I wanted that forefront in my brain while I was writing it. That’s why I called it Flush Real Pharynx: it’s about flushing your voice away and being bombarded with stuff.
With Models, I said, “F*ck that.” That’s really bad for our mental health. I’m gonna bounce, and I’m gonna make something more emotionally driven, or soulful, rather than a TikTok kind of thing. I’m not sure how long people can take that, just as a flesh-and-bone kind of thing. As brains, you and I are not that different to how we were 2,000 years ago. It’s having an effect on people’s mental health. As a culture, we try to come to terms with it. That’s what I did with Flush, and I injured myself by doing that, so I wanted to soothe myself a bit and switch it all off.
I’m not qualified to explain this, but in terms of how things would affect my music: previously it was cities. If you grew up in Detroit, you might be into one thing, and if you grew up in New York, you might be into something else, and if you grew up in Birmingham, it might be another. Now, it’s the internet. If you’re into something, you can find it. There are so many positive things about it, but at this point in my life, it’s better for my mental health for me to switch it off when I’m going to make something. It’s taken me a few years to learn that, but it’s fine. Sometimes things take a while to work out. If I were to make another record, I’d want to hire a studio in a barn somewhere on a mountain for three months.
You started Models when AI was prehistoric compared to where it is now, and when it wasn’t really a public conversation. What’s it been like to watch that evolution in real time?
Lee Gamble: Neural networks are from the ’50s, so AI is not new. But data happened. They didn’t have enough data sets to make a good neural network in the ’50s. Even when we started doing the voice modeling stuff, you’d need a bunch of stuff. Eventually, it’d get down to 60 seconds of something that you could work with. That was with two people and a normal computer; it’s not with Meta or Google or whatever. It was obvious to me that it was going to accelerate at rate of knots.
It’s a little bit of brain, in a sense, that can learn how to do something. It’s a simulation technology. Anything you experience is because you have a brain, fundamentally. If the brain is allowing everything to be knowable, then a neural network can try any of that. So, of course it’s going to be everywhere. We’ve seen it first in image-making, because most online communities are built around images. I knew it was going to be quick, so I had to say, “We are where we are, and let’s roll with that version.”
When I started writing [Models], I bowed out of the AI conversation. I started looking at it again recently, because I needed a break from it all. I read the other day about someone who had been in a coma for 20 years, and they lost their voice. They came out of the coma, and they’ve used AI to train them to have a voice again, which is such a Models thing. I think it’ll become ubiquitous. Then it’ll level off, and that’s when the interesting things will happen.
I wanted this to sound like a record that you’d make when AI isn’t a “thing” anymore. It’s a bit like early Herbie Hancock records. Everybody wanted to know what synthesizer he was using. 20 years later, nobody’s asking; he just plays. I wanted to place Models there somehow. I realize this is coming close to making “futuristic” music. [laughs] Right now, I’m overwhelmed by the shininess and potential of this technology. In years’ time, people won’t think of it like that.
When you put something like this out, you get people acting as though you’re okay with all that happening. The problem, for me, is that AI is already in every part of culture. So saying, “I’m not going to use it because it’s bad”—it’s not like not using Amazon because they’re scumbags. But we can’t not use AI. Fundamentally, it’s built on our collective labor, art, and time. I don’t want that to sound like a cop-out; I’m no libertarian. But, at a certain point, you’re going to have to engage with it. When that happens, it’ll either be unknowingly or with a bit of knowledge. The only way for me to offset that was to say, “Educate yourself on it, and hear some voices that aren’t of a similar background.”
I’m not here to make an excuse as to why I’d use it. I’m not out there saying, “This is amazing; we don’t need singers anymore.” I’m an artist, and I wouldn’t even begin to think that. If I can make something that can resonate emotionally with people, like I hope this record can, then maybe that’s okay. If I’m able to make people think about how they can use the technology in their way outside of any corporate dystopias, then maybe that’s okay. I’m not expecting to be able to make a record and say something about AI in general. There are things that we should be aware of and use as artists, and there are other things we should be wary of and at least talk about.
For example, there’s the protection of people’s voices. If I fed the first model one person’s voice and said, “Okay, that’s it,” that wouldn’t have been enough. I had to process it until I couldn’t tell who it was anymore; I had to eliminate the training from it somewhat. It’s not my voice. I understand there’s a complex relationship there. I had to be, and will continue to be, aware of that. People that look like me have claimed ownership of other people’s cultures and work, and that has to be taken seriously. If this can open a conversation about that, then that’s the least you can do if you’re using the technology. You can at least acknowledge that there’s a history of people taking people’s voices, mimicking them, and selling them because they’re a different complexion to the people who invented the style. I don’t think that any of this sounds like any particular person. I hope that it doesn’t, because that’s something I worked really hard to avoid.
It would be so easy to do that: you’ve seen the AI deepfake voices. I think the voice and the image in particular are very hot things to start playing around with. The image of another person that isn’t you is a very sensitive thing to mess with. What else have you got to identify yourself with other than the way you feel, look, and sound? That’s the most intimate thing, and to go, “I could do that now, because I have this Google sheet I can click on?” We have to think of it more respectfully than that.
I’m sure some people will disagree with that, but I’ve not heard anything yet when I’ve spoken about the record. When I was making it, I thought, “Maybe people will be really against this.” But I hope that, when people hear the record, they hear that that’s not where it’s coming from. It doesn’t feel like it’s coming from a position of ownership to me. It’s a dialogue. It’s: “Here’s where we’re going, and I’m trying to make something beautiful and touching with it.” That’s all I could do. I didn’t want to make something “icy,” or “fresh,” or “modern.” “Alien” or “other-dimensional,” sure, but I wanted to make it as warm as I could. I’m not too saccharine, but there’s definitely a few moments on there where I’m like, “F*cking hell.” The inner Cure fan came out. [laughs] I hope that people don’t think of it as an ode to a dystopian technology.
It’s a reckoning.
Lee Gamble: Yeah. It’s here. Try and own it: make it your work. It’s just some paint. It’s just some acrylic in a tube. Until you do anything with it, it doesn’t mean anything. There’s this incredible number of images made by image generation systems. Hito Steyerl is really interesting on this topic in terms of images. She talks about the image in one of her pieces of writing, In Defense of the Poor Image. It’s a piece of writing about the digital image. It goes through all sorts of different processes; it’ll get compressed again and again, and the colors change, and it all falls apart. So you get these poor images even though we’re in these hi-tech times.
This writing goes into a defense of the poor image, and it explores what qualities can be found in it. In the piece of writing I did, I titled one of my chapters—where I talk about the ‘crappy’ nature of the voices I used—”In Defense of the Poor Voice.” I quote Steyerl there, of course. In her writing, she says, “The poor image embodies the afterlife of many former masterpieces of cinema and video art.” I read her talk about it very eloquently, and I said, “I get that.” If I just swap all of that for the voice rather than the .JPG, then it’s the same thing. I’m seeing the some of the same advantages in the lack of quality as she’d seen. That really resonated with me.
What’s next for you?
Lee Gamble: It’s mainly the live show. I’m working with Candela Capítan, a performance artist and choreographer. I looked at her work and thought it was amazing, in terms of how it relates to the body, repetition, alienation, and the gaze and technologized image. I wanted to think about how to present Models live, and I didn’t want to be onstage with a laptop. From day one, I wanted to bring a physical body back into space—but not mine. Humans gave me the information; I simulated and disembodied them; and, now, I ought to put them back in. A lot of the movement is informed by the ideas of cloning, mirrors, and how people replicate each others’ movements. We tried to get twins to perform in the live show.
It’ll premier at Unsound in Poland, and then in London. I wanted to develop the choreography around the idea of how things learn—sometimes they learn badly. Obviously, Candela has brought a huge amount to the work. I gave them a PDF of ideas, and they read it; I played the music to them, and they improvised. It was very moving: I’d never seen anyone do that to my work before. So, I thought, “I want to get on this.”
I’m DJing as regularly as I can, just banging it out and enjoying myself. I’m really enjoying painting. I’m hoping to do some installation work with these voices as well. I don’t know if it’s just me, but I feel like my life moves in ten-year cycles. I did therapy last year for the first time in my life, and that came up. It was really helpful. I was really glad that people pushed me to do it. I’m not against it, of course, but I come from a background where that wasn’t a thing. It took me a long time, but I needed it, and it was super helpful. One of the things that came up is the idea of ten-year cycles in my life. Diversions was 2012, and that was a moment where I did a bunch of other stuff. Now, it’s 2023, and there’s Models, and I’m painting again. Now, I’m ready to boot them glass ceilings as hard as I can for the next ten years.
With this record, I said, “I know damn well that if, in two years’ time, if I say this voice modeling system sounds like a human, it’ll be trash.” So, forget that—just make a record with the technology that’s in front of you, and lean into the limitations. I can’t compete with these companies; I’m just a guy who knows a brilliant programmer. We don’t have a team behind us. Forget them—they’re doing their thing, and I’ll lean into the problems that they’re trying to fix.
It’s the idea of making something that sounds like a moment.
Lee Gamble: That’s something I’ve been thinking about recently. You’ve got this idea about the hardcore continuum, for instance. It keeps shapeshifting and latching onto other things, and every now and again, you’ll get a burst, and you’ll get a new genre. I feel like that’s finished. I don’t think it’s not actually happening: I just think that since we’re so deep into social media and streaming, access to things has been leveled.
I can be on Spotify, and I could be in a playlist with Beyoncé. That wouldn’t have happened before. If I was 10 years old now, I don’t have this reference to date or time: here’s jazz, here’s prog, here’s krautrock, here’s techno. It’s just all there. In a sense, that’s really amazing. I’ve really enjoyed Lil Yachty’s new record this year, and it seems to lean into that: there are points where it sounds like Pink Floyd, and then it’ll sound super modern, and then it sounds like the ’70s, and then it sounds like old-school computer games, and then it sounds like old hip-hop. It sounds like now.
I don’t think younger kids give a shit—and I don’t think they should—if something was made in 1970 or 2023. Who cares? If someone’s playing a track in a club, and you’ve got 2,000 people dancing to it, who cares if it’s from 1994? That’s what I mean about this idea of a continuum. It’s all been piledrived into a wall and splattered. It comes with its issues; I get that. But I think that time splattering is a good thing.
The question about what is, or sounds, “modern” or “old” gets mushier over time. If you’ve never heard something before, it’s new. I heard someone on NTS the other day—a working-class person from London—playing Steve Reich. They said, “I’m going down this wormhole with Steve Reich.” That’s amazing. That stuff is historically of the “academy,” or the “middle classes,” or the “conservatoire,” but there’s also a clear history from Indian music to minimalism. It all gets constantly re-contextualised in the streaming era.
I went down a YouTube wormhole with 100 gecs’ stuff. I’m not a stan, but it crossed me in the same way. It’s a mixture of things that sounds exactly like now. It doesn’t matter if I like it aesthetically. As a cultural object, you’ve taken this on and just ran with it! You don’t give a f*ck. I don’t know how old they are, but they’re younger than me. I think younger people aren’t trying to make a point of that; that’s just how it is. [laughs] There’s no reason to have this linear time where everything gets discarded—”Oh, it’s old.” I always thought that was a bit pointless.
Also, to flip towards the future: people are always like, “This sounds like the future.” What does that mean? I’m not trying to be facetious; I honestly don’t know what that means.
I’m definitely guilty of using that terminology.
Lee Gamble: You’re a writer, so it makes sense—it’s a good semantic thing. People go “Ah! I get it.” But what do you think the person’s gonna think it sounds like?
I try to use it to say, “This person is going in a really exciting direction, and maybe this is a possible vision for the future.”
Lee Gamble: That Lil Yachty record sounds like now. It might sound old in a year. But, for now, it sounds like the present. It’s that cross-pollination—corecore shit on TikTok, and mashups. I keep hearing the word “Dada” again. It feels like vertical collage. I did a piece recently which was performed at Berlin Atonal; they did a Xenakis X100 thing. I’m a huge admirer of Xenakis‘ work. I wrote this in memory of him, so I thought, “What would he be doing now?” You’re damn well sure he’s not doing what he was doing then. He was doing stochastic computer synthesis methods in the ’90s when very few people were doing that in the world. If he was a young person in 2023, I doubt he’d be writing for solo timpani. He’s a techno person, in a sense. I doubt he’d want to becalled a techno guy, but f*ck it, that’s fine.
It’s that idea of possible futures: what kind of future are you offering?
Lee Gamble: There’s definitely some of that aspect to Models. Take the Flush Real Pharynx series I did. That was me trying to come to terms with the culture clash we’re talking about: multiple screens at the same time. It’s a kind of collage. That wasn’t the defining thing for the record, but I had that in my head. I wanted that forefront in my brain while I was writing it. That’s why I called it Flush Real Pharynx: it’s about flushing your voice away and being bombarded with stuff.
With Models, I said, “F*ck that.” That’s really bad for our mental health. I’m gonna bounce, and I’m gonna make something more emotionally driven, or soulful, rather than a TikTok kind of thing. I’m not sure how long people can take that, just as a flesh-and-bone kind of thing. As brains, you and I are not that different to how we were 2,000 years ago. It’s having an effect on people’s mental health. As a culture, we try to come to terms with it. That’s what I did with Flush, and I injured myself by doing that, so I wanted to soothe myself a bit and switch it all off.
I’m not qualified to explain this, but in terms of how things would affect my music: previously it was cities. If you grew up in Detroit, you might be into one thing, and if you grew up in New York, you might be into something else, and if you grew up in Birmingham, it might be another. Now, it’s the internet. If you’re into something, you can find it. There are so many positive things about it, but at this point in my life, it’s better for my mental health for me to switch it off when I’m going to make something. It’s taken me a few years to learn that, but it’s fine. Sometimes things take a while to work out. If I were to make another record, I’d want to hire a studio in a barn somewhere on a mountain for three months.
You started Models when AI was prehistoric compared to where it is now, and when it wasn’t really a public conversation. What’s it been like to watch that evolution in real time?
Lee Gamble: Neural networks are from the ’50s, so AI is not new. But data happened. They didn’t have enough data sets to make a good neural network in the ’50s. Even when we started doing the voice modeling stuff, you’d need a bunch of stuff. Eventually, it’d get down to 60 seconds of something that you could work with. That was with two people and a normal computer; it’s not with Meta or Google or whatever. It was obvious to me that it was going to accelerate at rate of knots.
It’s a little bit of brain, in a sense, that can learn how to do something. It’s a simulation technology. Anything you experience is because you have a brain, fundamentally. If the brain is allowing everything to be knowable, then a neural network can try any of that. So, of course it’s going to be everywhere. We’ve seen it first in image-making, because most online communities are built around images. I knew it was going to be quick, so I had to say, “We are where we are, and let’s roll with that version.”
When I started writing [Models], I bowed out of the AI conversation. I started looking at it again recently, because I needed a break from it all. I read the other day about someone who had been in a coma for 20 years, and they lost their voice. They came out of the coma, and they’ve used AI to train them to have a voice again, which is such a Models thing. I think it’ll become ubiquitous. Then it’ll level off, and that’s when the interesting things will happen.
I wanted this to sound like a record that you’d make when AI isn’t a “thing” anymore. It’s a bit like early Herbie Hancock records. Everybody wanted to know what synthesizer he was using. 20 years later, nobody’s asking; he just plays. I wanted to place Models there somehow. I realize this is coming close to making “futuristic” music. [laughs] Right now, I’m overwhelmed by the shininess and potential of this technology. In years’ time, people won’t think of it like that.
When you put something like this out, you get people acting as though you’re okay with all that happening. The problem, for me, is that AI is already in every part of culture. So saying, “I’m not going to use it because it’s bad”—it’s not like not using Amazon because they’re scumbags. But we can’t not use AI. Fundamentally, it’s built on our collective labor, art, and time. I don’t want that to sound like a cop-out; I’m no libertarian. But, at a certain point, you’re going to have to engage with it. When that happens, it’ll either be unknowingly or with a bit of knowledge. The only way for me to offset that was to say, “Educate yourself on it, and hear some voices that aren’t of a similar background.”
I’m not here to make an excuse as to why I’d use it. I’m not out there saying, “This is amazing; we don’t need singers anymore.” I’m an artist, and I wouldn’t even begin to think that. If I can make something that can resonate emotionally with people, like I hope this record can, then maybe that’s okay. If I’m able to make people think about how they can use the technology in their way outside of any corporate dystopias, then maybe that’s okay. I’m not expecting to be able to make a record and say something about AI in general. There are things that we should be aware of and use as artists, and there are other things we should be wary of and at least talk about.
For example, there’s the protection of people’s voices. If I fed the first model one person’s voice and said, “Okay, that’s it,” that wouldn’t have been enough. I had to process it until I couldn’t tell who it was anymore; I had to eliminate the training from it somewhat. It’s not my voice. I understand there’s a complex relationship there. I had to be, and will continue to be, aware of that. People that look like me have claimed ownership of other people’s cultures and work, and that has to be taken seriously. If this can open a conversation about that, then that’s the least you can do if you’re using the technology. You can at least acknowledge that there’s a history of people taking people’s voices, mimicking them, and selling them because they’re a different complexion to the people who invented the style. I don’t think that any of this sounds like any particular person. I hope that it doesn’t, because that’s something I worked really hard to avoid.
It would be so easy to do that: you’ve seen the AI deepfake voices. I think the voice and the image in particular are very hot things to start playing around with. The image of another person that isn’t you is a very sensitive thing to mess with. What else have you got to identify yourself with other than the way you feel, look, and sound? That’s the most intimate thing, and to go, “I could do that now, because I have this Google sheet I can click on?” We have to think of it more respectfully than that.
I’m sure some people will disagree with that, but I’ve not heard anything yet when I’ve spoken about the record. When I was making it, I thought, “Maybe people will be really against this.” But I hope that, when people hear the record, they hear that that’s not where it’s coming from. It doesn’t feel like it’s coming from a position of ownership to me. It’s a dialogue. It’s: “Here’s where we’re going, and I’m trying to make something beautiful and touching with it.” That’s all I could do. I didn’t want to make something “icy,” or “fresh,” or “modern.” “Alien” or “other-dimensional,” sure, but I wanted to make it as warm as I could. I’m not too saccharine, but there’s definitely a few moments on there where I’m like, “F*cking hell.” The inner Cure fan came out. [laughs] I hope that people don’t think of it as an ode to a dystopian technology.
It’s a reckoning.
Lee Gamble: Yeah. It’s here. Try and own it: make it your work. It’s just some paint. It’s just some acrylic in a tube. Until you do anything with it, it doesn’t mean anything. There’s this incredible number of images made by image generation systems. Hito Steyerl is really interesting on this topic in terms of images. She talks about the image in one of her pieces of writing, In Defense of the Poor Image. It’s a piece of writing about the digital image. It goes through all sorts of different processes; it’ll get compressed again and again, and the colors change, and it all falls apart. So you get these poor images even though we’re in these hi-tech times.
This writing goes into a defense of the poor image, and it explores what qualities can be found in it. In the piece of writing I did, I titled one of my chapters—where I talk about the ‘crappy’ nature of the voices I used—”In Defense of the Poor Voice.” I quote Steyerl there, of course. In her writing, she says, “The poor image embodies the afterlife of many former masterpieces of cinema and video art.” I read her talk about it very eloquently, and I said, “I get that.” If I just swap all of that for the voice rather than the .JPG, then it’s the same thing. I’m seeing the some of the same advantages in the lack of quality as she’d seen. That really resonated with me.
What’s next for you?
Lee Gamble: It’s mainly the live show. I’m working with Candela Capítan, a performance artist and choreographer. I looked at her work and thought it was amazing, in terms of how it relates to the body, repetition, alienation, and the gaze and technologized image. I wanted to think about how to present Models live, and I didn’t want to be onstage with a laptop. From day one, I wanted to bring a physical body back into space—but not mine. Humans gave me the information; I simulated and disembodied them; and, now, I ought to put them back in. A lot of the movement is informed by the ideas of cloning, mirrors, and how people replicate each others’ movements. We tried to get twins to perform in the live show.
It’ll premier at Unsound in Poland, and then in London. I wanted to develop the choreography around the idea of how things learn—sometimes they learn badly. Obviously, Candela has brought a huge amount to the work. I gave them a PDF of ideas, and they read it; I played the music to them, and they improvised. It was very moving: I’d never seen anyone do that to my work before. So, I thought, “I want to get on this.”
I’m DJing as regularly as I can, just banging it out and enjoying myself. I’m really enjoying painting. I’m hoping to do some installation work with these voices as well. I don’t know if it’s just me, but I feel like my life moves in ten-year cycles. I did therapy last year for the first time in my life, and that came up. It was really helpful. I was really glad that people pushed me to do it. I’m not against it, of course, but I come from a background where that wasn’t a thing. It took me a long time, but I needed it, and it was super helpful. One of the things that came up is the idea of ten-year cycles in my life. Diversions was 2012, and that was a moment where I did a bunch of other stuff. Now, it’s 2023, and there’s Models, and I’m painting again. Now, I’m ready to boot them glass ceilings as hard as I can for the next ten years.
I can be on Spotify, and I could be in a playlist with Beyoncé. That wouldn’t have happened before. If I was 10 years old now, I don’t have this reference to date or time: here’s jazz, here’s prog, here’s krautrock, here’s techno. It’s just all there. In a sense, that’s really amazing. I’ve really enjoyed Lil Yachty’s new record this year, and it seems to lean into that: there are points where it sounds like Pink Floyd, and then it’ll sound super modern, and then it sounds like the ’70s, and then it sounds like old-school computer games, and then it sounds like old hip-hop. It sounds like now.
I don’t think younger kids give a shit—and I don’t think they should—if something was made in 1970 or 2023. Who cares? If someone’s playing a track in a club, and you’ve got 2,000 people dancing to it, who cares if it’s from 1994? That’s what I mean about this idea of a continuum. It’s all been piledrived into a wall and splattered. It comes with its issues; I get that. But I think that time splattering is a good thing.
The question about what is, or sounds, “modern” or “old” gets mushier over time. If you’ve never heard something before, it’s new. I heard someone on NTS the other day—a working-class person from London—playing Steve Reich. They said, “I’m going down this wormhole with Steve Reich.” That’s amazing. That stuff is historically of the “academy,” or the “middle classes,” or the “conservatoire,” but there’s also a clear history from Indian music to minimalism. It all gets constantly re-contextualised in the streaming era.
I went down a YouTube wormhole with 100 gecs’ stuff. I’m not a stan, but it crossed me in the same way. It’s a mixture of things that sounds exactly like now. It doesn’t matter if I like it aesthetically. As a cultural object, you’ve taken this on and just ran with it! You don’t give a f*ck. I don’t know how old they are, but they’re younger than me. I think younger people aren’t trying to make a point of that; that’s just how it is. [laughs] There’s no reason to have this linear time where everything gets discarded—”Oh, it’s old.” I always thought that was a bit pointless.
Also, to flip towards the future: people are always like, “This sounds like the future.” What does that mean? I’m not trying to be facetious; I honestly don’t know what that means.
I’m definitely guilty of using that terminology.
Lee Gamble: You’re a writer, so it makes sense—it’s a good semantic thing. People go “Ah! I get it.” But what do you think the person’s gonna think it sounds like?
I try to use it to say, “This person is going in a really exciting direction, and maybe this is a possible vision for the future.”
Lee Gamble: That Lil Yachty record sounds like now. It might sound old in a year. But, for now, it sounds like the present. It’s that cross-pollination—corecore shit on TikTok, and mashups. I keep hearing the word “Dada” again. It feels like vertical collage. I did a piece recently which was performed at Berlin Atonal; they did a Xenakis X100 thing. I’m a huge admirer of Xenakis‘ work. I wrote this in memory of him, so I thought, “What would he be doing now?” You’re damn well sure he’s not doing what he was doing then. He was doing stochastic computer synthesis methods in the ’90s when very few people were doing that in the world. If he was a young person in 2023, I doubt he’d be writing for solo timpani. He’s a techno person, in a sense. I doubt he’d want to becalled a techno guy, but f*ck it, that’s fine.
It’s that idea of possible futures: what kind of future are you offering?
Lee Gamble: There’s definitely some of that aspect to Models. Take the Flush Real Pharynx series I did. That was me trying to come to terms with the culture clash we’re talking about: multiple screens at the same time. It’s a kind of collage. That wasn’t the defining thing for the record, but I had that in my head. I wanted that forefront in my brain while I was writing it. That’s why I called it Flush Real Pharynx: it’s about flushing your voice away and being bombarded with stuff.
With Models, I said, “F*ck that.” That’s really bad for our mental health. I’m gonna bounce, and I’m gonna make something more emotionally driven, or soulful, rather than a TikTok kind of thing. I’m not sure how long people can take that, just as a flesh-and-bone kind of thing. As brains, you and I are not that different to how we were 2,000 years ago. It’s having an effect on people’s mental health. As a culture, we try to come to terms with it. That’s what I did with Flush, and I injured myself by doing that, so I wanted to soothe myself a bit and switch it all off.
I’m not qualified to explain this, but in terms of how things would affect my music: previously it was cities. If you grew up in Detroit, you might be into one thing, and if you grew up in New York, you might be into something else, and if you grew up in Birmingham, it might be another. Now, it’s the internet. If you’re into something, you can find it. There are so many positive things about it, but at this point in my life, it’s better for my mental health for me to switch it off when I’m going to make something. It’s taken me a few years to learn that, but it’s fine. Sometimes things take a while to work out. If I were to make another record, I’d want to hire a studio in a barn somewhere on a mountain for three months.
You started Models when AI was prehistoric compared to where it is now, and when it wasn’t really a public conversation. What’s it been like to watch that evolution in real time?
Lee Gamble: Neural networks are from the ’50s, so AI is not new. But data happened. They didn’t have enough data sets to make a good neural network in the ’50s. Even when we started doing the voice modeling stuff, you’d need a bunch of stuff. Eventually, it’d get down to 60 seconds of something that you could work with. That was with two people and a normal computer; it’s not with Meta or Google or whatever. It was obvious to me that it was going to accelerate at rate of knots.
It’s a little bit of brain, in a sense, that can learn how to do something. It’s a simulation technology. Anything you experience is because you have a brain, fundamentally. If the brain is allowing everything to be knowable, then a neural network can try any of that. So, of course it’s going to be everywhere. We’ve seen it first in image-making, because most online communities are built around images. I knew it was going to be quick, so I had to say, “We are where we are, and let’s roll with that version.”
When I started writing [Models], I bowed out of the AI conversation. I started looking at it again recently, because I needed a break from it all. I read the other day about someone who had been in a coma for 20 years, and they lost their voice. They came out of the coma, and they’ve used AI to train them to have a voice again, which is such a Models thing. I think it’ll become ubiquitous. Then it’ll level off, and that’s when the interesting things will happen.
I wanted this to sound like a record that you’d make when AI isn’t a “thing” anymore. It’s a bit like early Herbie Hancock records. Everybody wanted to know what synthesizer he was using. 20 years later, nobody’s asking; he just plays. I wanted to place Models there somehow. I realize this is coming close to making “futuristic” music. [laughs] Right now, I’m overwhelmed by the shininess and potential of this technology. In years’ time, people won’t think of it like that.
When you put something like this out, you get people acting as though you’re okay with all that happening. The problem, for me, is that AI is already in every part of culture. So saying, “I’m not going to use it because it’s bad”—it’s not like not using Amazon because they’re scumbags. But we can’t not use AI. Fundamentally, it’s built on our collective labor, art, and time. I don’t want that to sound like a cop-out; I’m no libertarian. But, at a certain point, you’re going to have to engage with it. When that happens, it’ll either be unknowingly or with a bit of knowledge. The only way for me to offset that was to say, “Educate yourself on it, and hear some voices that aren’t of a similar background.”
I’m not here to make an excuse as to why I’d use it. I’m not out there saying, “This is amazing; we don’t need singers anymore.” I’m an artist, and I wouldn’t even begin to think that. If I can make something that can resonate emotionally with people, like I hope this record can, then maybe that’s okay. If I’m able to make people think about how they can use the technology in their way outside of any corporate dystopias, then maybe that’s okay. I’m not expecting to be able to make a record and say something about AI in general. There are things that we should be aware of and use as artists, and there are other things we should be wary of and at least talk about.
For example, there’s the protection of people’s voices. If I fed the first model one person’s voice and said, “Okay, that’s it,” that wouldn’t have been enough. I had to process it until I couldn’t tell who it was anymore; I had to eliminate the training from it somewhat. It’s not my voice. I understand there’s a complex relationship there. I had to be, and will continue to be, aware of that. People that look like me have claimed ownership of other people’s cultures and work, and that has to be taken seriously. If this can open a conversation about that, then that’s the least you can do if you’re using the technology. You can at least acknowledge that there’s a history of people taking people’s voices, mimicking them, and selling them because they’re a different complexion to the people who invented the style. I don’t think that any of this sounds like any particular person. I hope that it doesn’t, because that’s something I worked really hard to avoid.
It would be so easy to do that: you’ve seen the AI deepfake voices. I think the voice and the image in particular are very hot things to start playing around with. The image of another person that isn’t you is a very sensitive thing to mess with. What else have you got to identify yourself with other than the way you feel, look, and sound? That’s the most intimate thing, and to go, “I could do that now, because I have this Google sheet I can click on?” We have to think of it more respectfully than that.
I’m sure some people will disagree with that, but I’ve not heard anything yet when I’ve spoken about the record. When I was making it, I thought, “Maybe people will be really against this.” But I hope that, when people hear the record, they hear that that’s not where it’s coming from. It doesn’t feel like it’s coming from a position of ownership to me. It’s a dialogue. It’s: “Here’s where we’re going, and I’m trying to make something beautiful and touching with it.” That’s all I could do. I didn’t want to make something “icy,” or “fresh,” or “modern.” “Alien” or “other-dimensional,” sure, but I wanted to make it as warm as I could. I’m not too saccharine, but there’s definitely a few moments on there where I’m like, “F*cking hell.” The inner Cure fan came out. [laughs] I hope that people don’t think of it as an ode to a dystopian technology.
It’s a reckoning.
Lee Gamble: Yeah. It’s here. Try and own it: make it your work. It’s just some paint. It’s just some acrylic in a tube. Until you do anything with it, it doesn’t mean anything. There’s this incredible number of images made by image generation systems. Hito Steyerl is really interesting on this topic in terms of images. She talks about the image in one of her pieces of writing, In Defense of the Poor Image. It’s a piece of writing about the digital image. It goes through all sorts of different processes; it’ll get compressed again and again, and the colors change, and it all falls apart. So you get these poor images even though we’re in these hi-tech times.
This writing goes into a defense of the poor image, and it explores what qualities can be found in it. In the piece of writing I did, I titled one of my chapters—where I talk about the ‘crappy’ nature of the voices I used—”In Defense of the Poor Voice.” I quote Steyerl there, of course. In her writing, she says, “The poor image embodies the afterlife of many former masterpieces of cinema and video art.” I read her talk about it very eloquently, and I said, “I get that.” If I just swap all of that for the voice rather than the .JPG, then it’s the same thing. I’m seeing the some of the same advantages in the lack of quality as she’d seen. That really resonated with me.
What’s next for you?
Lee Gamble: It’s mainly the live show. I’m working with Candela Capítan, a performance artist and choreographer. I looked at her work and thought it was amazing, in terms of how it relates to the body, repetition, alienation, and the gaze and technologized image. I wanted to think about how to present Models live, and I didn’t want to be onstage with a laptop. From day one, I wanted to bring a physical body back into space—but not mine. Humans gave me the information; I simulated and disembodied them; and, now, I ought to put them back in. A lot of the movement is informed by the ideas of cloning, mirrors, and how people replicate each others’ movements. We tried to get twins to perform in the live show.
It’ll premier at Unsound in Poland, and then in London. I wanted to develop the choreography around the idea of how things learn—sometimes they learn badly. Obviously, Candela has brought a huge amount to the work. I gave them a PDF of ideas, and they read it; I played the music to them, and they improvised. It was very moving: I’d never seen anyone do that to my work before. So, I thought, “I want to get on this.”
I’m DJing as regularly as I can, just banging it out and enjoying myself. I’m really enjoying painting. I’m hoping to do some installation work with these voices as well. I don’t know if it’s just me, but I feel like my life moves in ten-year cycles. I did therapy last year for the first time in my life, and that came up. It was really helpful. I was really glad that people pushed me to do it. I’m not against it, of course, but I come from a background where that wasn’t a thing. It took me a long time, but I needed it, and it was super helpful. One of the things that came up is the idea of ten-year cycles in my life. Diversions was 2012, and that was a moment where I did a bunch of other stuff. Now, it’s 2023, and there’s Models, and I’m painting again. Now, I’m ready to boot them glass ceilings as hard as I can for the next ten years.
I try to use it to say, “This person is going in a really exciting direction, and maybe this is a possible vision for the future.”
Lee Gamble: That Lil Yachty record sounds like now. It might sound old in a year. But, for now, it sounds like the present. It’s that cross-pollination—corecore shit on TikTok, and mashups. I keep hearing the word “Dada” again. It feels like vertical collage. I did a piece recently which was performed at Berlin Atonal; they did a Xenakis X100 thing. I’m a huge admirer of Xenakis‘ work. I wrote this in memory of him, so I thought, “What would he be doing now?” You’re damn well sure he’s not doing what he was doing then. He was doing stochastic computer synthesis methods in the ’90s when very few people were doing that in the world. If he was a young person in 2023, I doubt he’d be writing for solo timpani. He’s a techno person, in a sense. I doubt he’d want to becalled a techno guy, but f*ck it, that’s fine.
It’s that idea of possible futures: what kind of future are you offering?
Lee Gamble: There’s definitely some of that aspect to Models. Take the Flush Real Pharynx series I did. That was me trying to come to terms with the culture clash we’re talking about: multiple screens at the same time. It’s a kind of collage. That wasn’t the defining thing for the record, but I had that in my head. I wanted that forefront in my brain while I was writing it. That’s why I called it Flush Real Pharynx: it’s about flushing your voice away and being bombarded with stuff.
With Models, I said, “F*ck that.” That’s really bad for our mental health. I’m gonna bounce, and I’m gonna make something more emotionally driven, or soulful, rather than a TikTok kind of thing. I’m not sure how long people can take that, just as a flesh-and-bone kind of thing. As brains, you and I are not that different to how we were 2,000 years ago. It’s having an effect on people’s mental health. As a culture, we try to come to terms with it. That’s what I did with Flush, and I injured myself by doing that, so I wanted to soothe myself a bit and switch it all off.
I’m not qualified to explain this, but in terms of how things would affect my music: previously it was cities. If you grew up in Detroit, you might be into one thing, and if you grew up in New York, you might be into something else, and if you grew up in Birmingham, it might be another. Now, it’s the internet. If you’re into something, you can find it. There are so many positive things about it, but at this point in my life, it’s better for my mental health for me to switch it off when I’m going to make something. It’s taken me a few years to learn that, but it’s fine. Sometimes things take a while to work out. If I were to make another record, I’d want to hire a studio in a barn somewhere on a mountain for three months.
You started Models when AI was prehistoric compared to where it is now, and when it wasn’t really a public conversation. What’s it been like to watch that evolution in real time?
Lee Gamble: Neural networks are from the ’50s, so AI is not new. But data happened. They didn’t have enough data sets to make a good neural network in the ’50s. Even when we started doing the voice modeling stuff, you’d need a bunch of stuff. Eventually, it’d get down to 60 seconds of something that you could work with. That was with two people and a normal computer; it’s not with Meta or Google or whatever. It was obvious to me that it was going to accelerate at rate of knots.
It’s a little bit of brain, in a sense, that can learn how to do something. It’s a simulation technology. Anything you experience is because you have a brain, fundamentally. If the brain is allowing everything to be knowable, then a neural network can try any of that. So, of course it’s going to be everywhere. We’ve seen it first in image-making, because most online communities are built around images. I knew it was going to be quick, so I had to say, “We are where we are, and let’s roll with that version.”
When I started writing [Models], I bowed out of the AI conversation. I started looking at it again recently, because I needed a break from it all. I read the other day about someone who had been in a coma for 20 years, and they lost their voice. They came out of the coma, and they’ve used AI to train them to have a voice again, which is such a Models thing. I think it’ll become ubiquitous. Then it’ll level off, and that’s when the interesting things will happen.
I wanted this to sound like a record that you’d make when AI isn’t a “thing” anymore. It’s a bit like early Herbie Hancock records. Everybody wanted to know what synthesizer he was using. 20 years later, nobody’s asking; he just plays. I wanted to place Models there somehow. I realize this is coming close to making “futuristic” music. [laughs] Right now, I’m overwhelmed by the shininess and potential of this technology. In years’ time, people won’t think of it like that.
When you put something like this out, you get people acting as though you’re okay with all that happening. The problem, for me, is that AI is already in every part of culture. So saying, “I’m not going to use it because it’s bad”—it’s not like not using Amazon because they’re scumbags. But we can’t not use AI. Fundamentally, it’s built on our collective labor, art, and time. I don’t want that to sound like a cop-out; I’m no libertarian. But, at a certain point, you’re going to have to engage with it. When that happens, it’ll either be unknowingly or with a bit of knowledge. The only way for me to offset that was to say, “Educate yourself on it, and hear some voices that aren’t of a similar background.”
I’m not here to make an excuse as to why I’d use it. I’m not out there saying, “This is amazing; we don’t need singers anymore.” I’m an artist, and I wouldn’t even begin to think that. If I can make something that can resonate emotionally with people, like I hope this record can, then maybe that’s okay. If I’m able to make people think about how they can use the technology in their way outside of any corporate dystopias, then maybe that’s okay. I’m not expecting to be able to make a record and say something about AI in general. There are things that we should be aware of and use as artists, and there are other things we should be wary of and at least talk about.
For example, there’s the protection of people’s voices. If I fed the first model one person’s voice and said, “Okay, that’s it,” that wouldn’t have been enough. I had to process it until I couldn’t tell who it was anymore; I had to eliminate the training from it somewhat. It’s not my voice. I understand there’s a complex relationship there. I had to be, and will continue to be, aware of that. People that look like me have claimed ownership of other people’s cultures and work, and that has to be taken seriously. If this can open a conversation about that, then that’s the least you can do if you’re using the technology. You can at least acknowledge that there’s a history of people taking people’s voices, mimicking them, and selling them because they’re a different complexion to the people who invented the style. I don’t think that any of this sounds like any particular person. I hope that it doesn’t, because that’s something I worked really hard to avoid.
It would be so easy to do that: you’ve seen the AI deepfake voices. I think the voice and the image in particular are very hot things to start playing around with. The image of another person that isn’t you is a very sensitive thing to mess with. What else have you got to identify yourself with other than the way you feel, look, and sound? That’s the most intimate thing, and to go, “I could do that now, because I have this Google sheet I can click on?” We have to think of it more respectfully than that.
I’m sure some people will disagree with that, but I’ve not heard anything yet when I’ve spoken about the record. When I was making it, I thought, “Maybe people will be really against this.” But I hope that, when people hear the record, they hear that that’s not where it’s coming from. It doesn’t feel like it’s coming from a position of ownership to me. It’s a dialogue. It’s: “Here’s where we’re going, and I’m trying to make something beautiful and touching with it.” That’s all I could do. I didn’t want to make something “icy,” or “fresh,” or “modern.” “Alien” or “other-dimensional,” sure, but I wanted to make it as warm as I could. I’m not too saccharine, but there’s definitely a few moments on there where I’m like, “F*cking hell.” The inner Cure fan came out. [laughs] I hope that people don’t think of it as an ode to a dystopian technology.
It’s a reckoning.
Lee Gamble: Yeah. It’s here. Try and own it: make it your work. It’s just some paint. It’s just some acrylic in a tube. Until you do anything with it, it doesn’t mean anything. There’s this incredible number of images made by image generation systems. Hito Steyerl is really interesting on this topic in terms of images. She talks about the image in one of her pieces of writing, In Defense of the Poor Image. It’s a piece of writing about the digital image. It goes through all sorts of different processes; it’ll get compressed again and again, and the colors change, and it all falls apart. So you get these poor images even though we’re in these hi-tech times.
This writing goes into a defense of the poor image, and it explores what qualities can be found in it. In the piece of writing I did, I titled one of my chapters—where I talk about the ‘crappy’ nature of the voices I used—”In Defense of the Poor Voice.” I quote Steyerl there, of course. In her writing, she says, “The poor image embodies the afterlife of many former masterpieces of cinema and video art.” I read her talk about it very eloquently, and I said, “I get that.” If I just swap all of that for the voice rather than the .JPG, then it’s the same thing. I’m seeing the some of the same advantages in the lack of quality as she’d seen. That really resonated with me.
What’s next for you?
Lee Gamble: It’s mainly the live show. I’m working with Candela Capítan, a performance artist and choreographer. I looked at her work and thought it was amazing, in terms of how it relates to the body, repetition, alienation, and the gaze and technologized image. I wanted to think about how to present Models live, and I didn’t want to be onstage with a laptop. From day one, I wanted to bring a physical body back into space—but not mine. Humans gave me the information; I simulated and disembodied them; and, now, I ought to put them back in. A lot of the movement is informed by the ideas of cloning, mirrors, and how people replicate each others’ movements. We tried to get twins to perform in the live show.
It’ll premier at Unsound in Poland, and then in London. I wanted to develop the choreography around the idea of how things learn—sometimes they learn badly. Obviously, Candela has brought a huge amount to the work. I gave them a PDF of ideas, and they read it; I played the music to them, and they improvised. It was very moving: I’d never seen anyone do that to my work before. So, I thought, “I want to get on this.”
I’m DJing as regularly as I can, just banging it out and enjoying myself. I’m really enjoying painting. I’m hoping to do some installation work with these voices as well. I don’t know if it’s just me, but I feel like my life moves in ten-year cycles. I did therapy last year for the first time in my life, and that came up. It was really helpful. I was really glad that people pushed me to do it. I’m not against it, of course, but I come from a background where that wasn’t a thing. It took me a long time, but I needed it, and it was super helpful. One of the things that came up is the idea of ten-year cycles in my life. Diversions was 2012, and that was a moment where I did a bunch of other stuff. Now, it’s 2023, and there’s Models, and I’m painting again. Now, I’m ready to boot them glass ceilings as hard as I can for the next ten years.
It’s that idea of possible futures: what kind of future are you offering?
Lee Gamble: There’s definitely some of that aspect to Models. Take the Flush Real Pharynx series I did. That was me trying to come to terms with the culture clash we’re talking about: multiple screens at the same time. It’s a kind of collage. That wasn’t the defining thing for the record, but I had that in my head. I wanted that forefront in my brain while I was writing it. That’s why I called it Flush Real Pharynx: it’s about flushing your voice away and being bombarded with stuff.
With Models, I said, “F*ck that.” That’s really bad for our mental health. I’m gonna bounce, and I’m gonna make something more emotionally driven, or soulful, rather than a TikTok kind of thing. I’m not sure how long people can take that, just as a flesh-and-bone kind of thing. As brains, you and I are not that different to how we were 2,000 years ago. It’s having an effect on people’s mental health. As a culture, we try to come to terms with it. That’s what I did with Flush, and I injured myself by doing that, so I wanted to soothe myself a bit and switch it all off.
I’m not qualified to explain this, but in terms of how things would affect my music: previously it was cities. If you grew up in Detroit, you might be into one thing, and if you grew up in New York, you might be into something else, and if you grew up in Birmingham, it might be another. Now, it’s the internet. If you’re into something, you can find it. There are so many positive things about it, but at this point in my life, it’s better for my mental health for me to switch it off when I’m going to make something. It’s taken me a few years to learn that, but it’s fine. Sometimes things take a while to work out. If I were to make another record, I’d want to hire a studio in a barn somewhere on a mountain for three months.
You started Models when AI was prehistoric compared to where it is now, and when it wasn’t really a public conversation. What’s it been like to watch that evolution in real time?
Lee Gamble: Neural networks are from the ’50s, so AI is not new. But data happened. They didn’t have enough data sets to make a good neural network in the ’50s. Even when we started doing the voice modeling stuff, you’d need a bunch of stuff. Eventually, it’d get down to 60 seconds of something that you could work with. That was with two people and a normal computer; it’s not with Meta or Google or whatever. It was obvious to me that it was going to accelerate at rate of knots.
It’s a little bit of brain, in a sense, that can learn how to do something. It’s a simulation technology. Anything you experience is because you have a brain, fundamentally. If the brain is allowing everything to be knowable, then a neural network can try any of that. So, of course it’s going to be everywhere. We’ve seen it first in image-making, because most online communities are built around images. I knew it was going to be quick, so I had to say, “We are where we are, and let’s roll with that version.”
When I started writing [Models], I bowed out of the AI conversation. I started looking at it again recently, because I needed a break from it all. I read the other day about someone who had been in a coma for 20 years, and they lost their voice. They came out of the coma, and they’ve used AI to train them to have a voice again, which is such a Models thing. I think it’ll become ubiquitous. Then it’ll level off, and that’s when the interesting things will happen.
I wanted this to sound like a record that you’d make when AI isn’t a “thing” anymore. It’s a bit like early Herbie Hancock records. Everybody wanted to know what synthesizer he was using. 20 years later, nobody’s asking; he just plays. I wanted to place Models there somehow. I realize this is coming close to making “futuristic” music. [laughs] Right now, I’m overwhelmed by the shininess and potential of this technology. In years’ time, people won’t think of it like that.
When you put something like this out, you get people acting as though you’re okay with all that happening. The problem, for me, is that AI is already in every part of culture. So saying, “I’m not going to use it because it’s bad”—it’s not like not using Amazon because they’re scumbags. But we can’t not use AI. Fundamentally, it’s built on our collective labor, art, and time. I don’t want that to sound like a cop-out; I’m no libertarian. But, at a certain point, you’re going to have to engage with it. When that happens, it’ll either be unknowingly or with a bit of knowledge. The only way for me to offset that was to say, “Educate yourself on it, and hear some voices that aren’t of a similar background.”
I’m not here to make an excuse as to why I’d use it. I’m not out there saying, “This is amazing; we don’t need singers anymore.” I’m an artist, and I wouldn’t even begin to think that. If I can make something that can resonate emotionally with people, like I hope this record can, then maybe that’s okay. If I’m able to make people think about how they can use the technology in their way outside of any corporate dystopias, then maybe that’s okay. I’m not expecting to be able to make a record and say something about AI in general. There are things that we should be aware of and use as artists, and there are other things we should be wary of and at least talk about.
For example, there’s the protection of people’s voices. If I fed the first model one person’s voice and said, “Okay, that’s it,” that wouldn’t have been enough. I had to process it until I couldn’t tell who it was anymore; I had to eliminate the training from it somewhat. It’s not my voice. I understand there’s a complex relationship there. I had to be, and will continue to be, aware of that. People that look like me have claimed ownership of other people’s cultures and work, and that has to be taken seriously. If this can open a conversation about that, then that’s the least you can do if you’re using the technology. You can at least acknowledge that there’s a history of people taking people’s voices, mimicking them, and selling them because they’re a different complexion to the people who invented the style. I don’t think that any of this sounds like any particular person. I hope that it doesn’t, because that’s something I worked really hard to avoid.
It would be so easy to do that: you’ve seen the AI deepfake voices. I think the voice and the image in particular are very hot things to start playing around with. The image of another person that isn’t you is a very sensitive thing to mess with. What else have you got to identify yourself with other than the way you feel, look, and sound? That’s the most intimate thing, and to go, “I could do that now, because I have this Google sheet I can click on?” We have to think of it more respectfully than that.
I’m sure some people will disagree with that, but I’ve not heard anything yet when I’ve spoken about the record. When I was making it, I thought, “Maybe people will be really against this.” But I hope that, when people hear the record, they hear that that’s not where it’s coming from. It doesn’t feel like it’s coming from a position of ownership to me. It’s a dialogue. It’s: “Here’s where we’re going, and I’m trying to make something beautiful and touching with it.” That’s all I could do. I didn’t want to make something “icy,” or “fresh,” or “modern.” “Alien” or “other-dimensional,” sure, but I wanted to make it as warm as I could. I’m not too saccharine, but there’s definitely a few moments on there where I’m like, “F*cking hell.” The inner Cure fan came out. [laughs] I hope that people don’t think of it as an ode to a dystopian technology.
It’s a reckoning.
Lee Gamble: Yeah. It’s here. Try and own it: make it your work. It’s just some paint. It’s just some acrylic in a tube. Until you do anything with it, it doesn’t mean anything. There’s this incredible number of images made by image generation systems. Hito Steyerl is really interesting on this topic in terms of images. She talks about the image in one of her pieces of writing, In Defense of the Poor Image. It’s a piece of writing about the digital image. It goes through all sorts of different processes; it’ll get compressed again and again, and the colors change, and it all falls apart. So you get these poor images even though we’re in these hi-tech times.
This writing goes into a defense of the poor image, and it explores what qualities can be found in it. In the piece of writing I did, I titled one of my chapters—where I talk about the ‘crappy’ nature of the voices I used—”In Defense of the Poor Voice.” I quote Steyerl there, of course. In her writing, she says, “The poor image embodies the afterlife of many former masterpieces of cinema and video art.” I read her talk about it very eloquently, and I said, “I get that.” If I just swap all of that for the voice rather than the .JPG, then it’s the same thing. I’m seeing the some of the same advantages in the lack of quality as she’d seen. That really resonated with me.
What’s next for you?
Lee Gamble: It’s mainly the live show. I’m working with Candela Capítan, a performance artist and choreographer. I looked at her work and thought it was amazing, in terms of how it relates to the body, repetition, alienation, and the gaze and technologized image. I wanted to think about how to present Models live, and I didn’t want to be onstage with a laptop. From day one, I wanted to bring a physical body back into space—but not mine. Humans gave me the information; I simulated and disembodied them; and, now, I ought to put them back in. A lot of the movement is informed by the ideas of cloning, mirrors, and how people replicate each others’ movements. We tried to get twins to perform in the live show.
It’ll premier at Unsound in Poland, and then in London. I wanted to develop the choreography around the idea of how things learn—sometimes they learn badly. Obviously, Candela has brought a huge amount to the work. I gave them a PDF of ideas, and they read it; I played the music to them, and they improvised. It was very moving: I’d never seen anyone do that to my work before. So, I thought, “I want to get on this.”
I’m DJing as regularly as I can, just banging it out and enjoying myself. I’m really enjoying painting. I’m hoping to do some installation work with these voices as well. I don’t know if it’s just me, but I feel like my life moves in ten-year cycles. I did therapy last year for the first time in my life, and that came up. It was really helpful. I was really glad that people pushed me to do it. I’m not against it, of course, but I come from a background where that wasn’t a thing. It took me a long time, but I needed it, and it was super helpful. One of the things that came up is the idea of ten-year cycles in my life. Diversions was 2012, and that was a moment where I did a bunch of other stuff. Now, it’s 2023, and there’s Models, and I’m painting again. Now, I’m ready to boot them glass ceilings as hard as I can for the next ten years.
With Models, I said, “F*ck that.” That’s really bad for our mental health. I’m gonna bounce, and I’m gonna make something more emotionally driven, or soulful, rather than a TikTok kind of thing. I’m not sure how long people can take that, just as a flesh-and-bone kind of thing. As brains, you and I are not that different to how we were 2,000 years ago. It’s having an effect on people’s mental health. As a culture, we try to come to terms with it. That’s what I did with Flush, and I injured myself by doing that, so I wanted to soothe myself a bit and switch it all off.
I’m not qualified to explain this, but in terms of how things would affect my music: previously it was cities. If you grew up in Detroit, you might be into one thing, and if you grew up in New York, you might be into something else, and if you grew up in Birmingham, it might be another. Now, it’s the internet. If you’re into something, you can find it. There are so many positive things about it, but at this point in my life, it’s better for my mental health for me to switch it off when I’m going to make something. It’s taken me a few years to learn that, but it’s fine. Sometimes things take a while to work out. If I were to make another record, I’d want to hire a studio in a barn somewhere on a mountain for three months.
You started Models when AI was prehistoric compared to where it is now, and when it wasn’t really a public conversation. What’s it been like to watch that evolution in real time?
Lee Gamble: Neural networks are from the ’50s, so AI is not new. But data happened. They didn’t have enough data sets to make a good neural network in the ’50s. Even when we started doing the voice modeling stuff, you’d need a bunch of stuff. Eventually, it’d get down to 60 seconds of something that you could work with. That was with two people and a normal computer; it’s not with Meta or Google or whatever. It was obvious to me that it was going to accelerate at rate of knots.
It’s a little bit of brain, in a sense, that can learn how to do something. It’s a simulation technology. Anything you experience is because you have a brain, fundamentally. If the brain is allowing everything to be knowable, then a neural network can try any of that. So, of course it’s going to be everywhere. We’ve seen it first in image-making, because most online communities are built around images. I knew it was going to be quick, so I had to say, “We are where we are, and let’s roll with that version.”
When I started writing [Models], I bowed out of the AI conversation. I started looking at it again recently, because I needed a break from it all. I read the other day about someone who had been in a coma for 20 years, and they lost their voice. They came out of the coma, and they’ve used AI to train them to have a voice again, which is such a Models thing. I think it’ll become ubiquitous. Then it’ll level off, and that’s when the interesting things will happen.
I wanted this to sound like a record that you’d make when AI isn’t a “thing” anymore. It’s a bit like early Herbie Hancock records. Everybody wanted to know what synthesizer he was using. 20 years later, nobody’s asking; he just plays. I wanted to place Models there somehow. I realize this is coming close to making “futuristic” music. [laughs] Right now, I’m overwhelmed by the shininess and potential of this technology. In years’ time, people won’t think of it like that.
When you put something like this out, you get people acting as though you’re okay with all that happening. The problem, for me, is that AI is already in every part of culture. So saying, “I’m not going to use it because it’s bad”—it’s not like not using Amazon because they’re scumbags. But we can’t not use AI. Fundamentally, it’s built on our collective labor, art, and time. I don’t want that to sound like a cop-out; I’m no libertarian. But, at a certain point, you’re going to have to engage with it. When that happens, it’ll either be unknowingly or with a bit of knowledge. The only way for me to offset that was to say, “Educate yourself on it, and hear some voices that aren’t of a similar background.”
I’m not here to make an excuse as to why I’d use it. I’m not out there saying, “This is amazing; we don’t need singers anymore.” I’m an artist, and I wouldn’t even begin to think that. If I can make something that can resonate emotionally with people, like I hope this record can, then maybe that’s okay. If I’m able to make people think about how they can use the technology in their way outside of any corporate dystopias, then maybe that’s okay. I’m not expecting to be able to make a record and say something about AI in general. There are things that we should be aware of and use as artists, and there are other things we should be wary of and at least talk about.
For example, there’s the protection of people’s voices. If I fed the first model one person’s voice and said, “Okay, that’s it,” that wouldn’t have been enough. I had to process it until I couldn’t tell who it was anymore; I had to eliminate the training from it somewhat. It’s not my voice. I understand there’s a complex relationship there. I had to be, and will continue to be, aware of that. People that look like me have claimed ownership of other people’s cultures and work, and that has to be taken seriously. If this can open a conversation about that, then that’s the least you can do if you’re using the technology. You can at least acknowledge that there’s a history of people taking people’s voices, mimicking them, and selling them because they’re a different complexion to the people who invented the style. I don’t think that any of this sounds like any particular person. I hope that it doesn’t, because that’s something I worked really hard to avoid.
It would be so easy to do that: you’ve seen the AI deepfake voices. I think the voice and the image in particular are very hot things to start playing around with. The image of another person that isn’t you is a very sensitive thing to mess with. What else have you got to identify yourself with other than the way you feel, look, and sound? That’s the most intimate thing, and to go, “I could do that now, because I have this Google sheet I can click on?” We have to think of it more respectfully than that.
I’m sure some people will disagree with that, but I’ve not heard anything yet when I’ve spoken about the record. When I was making it, I thought, “Maybe people will be really against this.” But I hope that, when people hear the record, they hear that that’s not where it’s coming from. It doesn’t feel like it’s coming from a position of ownership to me. It’s a dialogue. It’s: “Here’s where we’re going, and I’m trying to make something beautiful and touching with it.” That’s all I could do. I didn’t want to make something “icy,” or “fresh,” or “modern.” “Alien” or “other-dimensional,” sure, but I wanted to make it as warm as I could. I’m not too saccharine, but there’s definitely a few moments on there where I’m like, “F*cking hell.” The inner Cure fan came out. [laughs] I hope that people don’t think of it as an ode to a dystopian technology.
It’s a reckoning.
Lee Gamble: Yeah. It’s here. Try and own it: make it your work. It’s just some paint. It’s just some acrylic in a tube. Until you do anything with it, it doesn’t mean anything. There’s this incredible number of images made by image generation systems. Hito Steyerl is really interesting on this topic in terms of images. She talks about the image in one of her pieces of writing, In Defense of the Poor Image. It’s a piece of writing about the digital image. It goes through all sorts of different processes; it’ll get compressed again and again, and the colors change, and it all falls apart. So you get these poor images even though we’re in these hi-tech times.
This writing goes into a defense of the poor image, and it explores what qualities can be found in it. In the piece of writing I did, I titled one of my chapters—where I talk about the ‘crappy’ nature of the voices I used—”In Defense of the Poor Voice.” I quote Steyerl there, of course. In her writing, she says, “The poor image embodies the afterlife of many former masterpieces of cinema and video art.” I read her talk about it very eloquently, and I said, “I get that.” If I just swap all of that for the voice rather than the .JPG, then it’s the same thing. I’m seeing the some of the same advantages in the lack of quality as she’d seen. That really resonated with me.
What’s next for you?
Lee Gamble: It’s mainly the live show. I’m working with Candela Capítan, a performance artist and choreographer. I looked at her work and thought it was amazing, in terms of how it relates to the body, repetition, alienation, and the gaze and technologized image. I wanted to think about how to present Models live, and I didn’t want to be onstage with a laptop. From day one, I wanted to bring a physical body back into space—but not mine. Humans gave me the information; I simulated and disembodied them; and, now, I ought to put them back in. A lot of the movement is informed by the ideas of cloning, mirrors, and how people replicate each others’ movements. We tried to get twins to perform in the live show.
It’ll premier at Unsound in Poland, and then in London. I wanted to develop the choreography around the idea of how things learn—sometimes they learn badly. Obviously, Candela has brought a huge amount to the work. I gave them a PDF of ideas, and they read it; I played the music to them, and they improvised. It was very moving: I’d never seen anyone do that to my work before. So, I thought, “I want to get on this.”
I’m DJing as regularly as I can, just banging it out and enjoying myself. I’m really enjoying painting. I’m hoping to do some installation work with these voices as well. I don’t know if it’s just me, but I feel like my life moves in ten-year cycles. I did therapy last year for the first time in my life, and that came up. It was really helpful. I was really glad that people pushed me to do it. I’m not against it, of course, but I come from a background where that wasn’t a thing. It took me a long time, but I needed it, and it was super helpful. One of the things that came up is the idea of ten-year cycles in my life. Diversions was 2012, and that was a moment where I did a bunch of other stuff. Now, it’s 2023, and there’s Models, and I’m painting again. Now, I’m ready to boot them glass ceilings as hard as I can for the next ten years.
It’s a little bit of brain, in a sense, that can learn how to do something. It’s a simulation technology. Anything you experience is because you have a brain, fundamentally. If the brain is allowing everything to be knowable, then a neural network can try any of that. So, of course it’s going to be everywhere. We’ve seen it first in image-making, because most online communities are built around images. I knew it was going to be quick, so I had to say, “We are where we are, and let’s roll with that version.”
When I started writing [Models], I bowed out of the AI conversation. I started looking at it again recently, because I needed a break from it all. I read the other day about someone who had been in a coma for 20 years, and they lost their voice. They came out of the coma, and they’ve used AI to train them to have a voice again, which is such a Models thing. I think it’ll become ubiquitous. Then it’ll level off, and that’s when the interesting things will happen.
I wanted this to sound like a record that you’d make when AI isn’t a “thing” anymore. It’s a bit like early Herbie Hancock records. Everybody wanted to know what synthesizer he was using. 20 years later, nobody’s asking; he just plays. I wanted to place Models there somehow. I realize this is coming close to making “futuristic” music. [laughs] Right now, I’m overwhelmed by the shininess and potential of this technology. In years’ time, people won’t think of it like that.
When you put something like this out, you get people acting as though you’re okay with all that happening. The problem, for me, is that AI is already in every part of culture. So saying, “I’m not going to use it because it’s bad”—it’s not like not using Amazon because they’re scumbags. But we can’t not use AI. Fundamentally, it’s built on our collective labor, art, and time. I don’t want that to sound like a cop-out; I’m no libertarian. But, at a certain point, you’re going to have to engage with it. When that happens, it’ll either be unknowingly or with a bit of knowledge. The only way for me to offset that was to say, “Educate yourself on it, and hear some voices that aren’t of a similar background.”
I’m not here to make an excuse as to why I’d use it. I’m not out there saying, “This is amazing; we don’t need singers anymore.” I’m an artist, and I wouldn’t even begin to think that. If I can make something that can resonate emotionally with people, like I hope this record can, then maybe that’s okay. If I’m able to make people think about how they can use the technology in their way outside of any corporate dystopias, then maybe that’s okay. I’m not expecting to be able to make a record and say something about AI in general. There are things that we should be aware of and use as artists, and there are other things we should be wary of and at least talk about.
For example, there’s the protection of people’s voices. If I fed the first model one person’s voice and said, “Okay, that’s it,” that wouldn’t have been enough. I had to process it until I couldn’t tell who it was anymore; I had to eliminate the training from it somewhat. It’s not my voice. I understand there’s a complex relationship there. I had to be, and will continue to be, aware of that. People that look like me have claimed ownership of other people’s cultures and work, and that has to be taken seriously. If this can open a conversation about that, then that’s the least you can do if you’re using the technology. You can at least acknowledge that there’s a history of people taking people’s voices, mimicking them, and selling them because they’re a different complexion to the people who invented the style. I don’t think that any of this sounds like any particular person. I hope that it doesn’t, because that’s something I worked really hard to avoid.
It would be so easy to do that: you’ve seen the AI deepfake voices. I think the voice and the image in particular are very hot things to start playing around with. The image of another person that isn’t you is a very sensitive thing to mess with. What else have you got to identify yourself with other than the way you feel, look, and sound? That’s the most intimate thing, and to go, “I could do that now, because I have this Google sheet I can click on?” We have to think of it more respectfully than that.
I’m sure some people will disagree with that, but I’ve not heard anything yet when I’ve spoken about the record. When I was making it, I thought, “Maybe people will be really against this.” But I hope that, when people hear the record, they hear that that’s not where it’s coming from. It doesn’t feel like it’s coming from a position of ownership to me. It’s a dialogue. It’s: “Here’s where we’re going, and I’m trying to make something beautiful and touching with it.” That’s all I could do. I didn’t want to make something “icy,” or “fresh,” or “modern.” “Alien” or “other-dimensional,” sure, but I wanted to make it as warm as I could. I’m not too saccharine, but there’s definitely a few moments on there where I’m like, “F*cking hell.” The inner Cure fan came out. [laughs] I hope that people don’t think of it as an ode to a dystopian technology.
It’s a reckoning.
Lee Gamble: Yeah. It’s here. Try and own it: make it your work. It’s just some paint. It’s just some acrylic in a tube. Until you do anything with it, it doesn’t mean anything. There’s this incredible number of images made by image generation systems. Hito Steyerl is really interesting on this topic in terms of images. She talks about the image in one of her pieces of writing, In Defense of the Poor Image. It’s a piece of writing about the digital image. It goes through all sorts of different processes; it’ll get compressed again and again, and the colors change, and it all falls apart. So you get these poor images even though we’re in these hi-tech times.
This writing goes into a defense of the poor image, and it explores what qualities can be found in it. In the piece of writing I did, I titled one of my chapters—where I talk about the ‘crappy’ nature of the voices I used—”In Defense of the Poor Voice.” I quote Steyerl there, of course. In her writing, she says, “The poor image embodies the afterlife of many former masterpieces of cinema and video art.” I read her talk about it very eloquently, and I said, “I get that.” If I just swap all of that for the voice rather than the .JPG, then it’s the same thing. I’m seeing the some of the same advantages in the lack of quality as she’d seen. That really resonated with me.
What’s next for you?
Lee Gamble: It’s mainly the live show. I’m working with Candela Capítan, a performance artist and choreographer. I looked at her work and thought it was amazing, in terms of how it relates to the body, repetition, alienation, and the gaze and technologized image. I wanted to think about how to present Models live, and I didn’t want to be onstage with a laptop. From day one, I wanted to bring a physical body back into space—but not mine. Humans gave me the information; I simulated and disembodied them; and, now, I ought to put them back in. A lot of the movement is informed by the ideas of cloning, mirrors, and how people replicate each others’ movements. We tried to get twins to perform in the live show.
It’ll premier at Unsound in Poland, and then in London. I wanted to develop the choreography around the idea of how things learn—sometimes they learn badly. Obviously, Candela has brought a huge amount to the work. I gave them a PDF of ideas, and they read it; I played the music to them, and they improvised. It was very moving: I’d never seen anyone do that to my work before. So, I thought, “I want to get on this.”
I’m DJing as regularly as I can, just banging it out and enjoying myself. I’m really enjoying painting. I’m hoping to do some installation work with these voices as well. I don’t know if it’s just me, but I feel like my life moves in ten-year cycles. I did therapy last year for the first time in my life, and that came up. It was really helpful. I was really glad that people pushed me to do it. I’m not against it, of course, but I come from a background where that wasn’t a thing. It took me a long time, but I needed it, and it was super helpful. One of the things that came up is the idea of ten-year cycles in my life. Diversions was 2012, and that was a moment where I did a bunch of other stuff. Now, it’s 2023, and there’s Models, and I’m painting again. Now, I’m ready to boot them glass ceilings as hard as I can for the next ten years.
This writing goes into a defense of the poor image, and it explores what qualities can be found in it. In the piece of writing I did, I titled one of my chapters—where I talk about the ‘crappy’ nature of the voices I used—”In Defense of the Poor Voice.” I quote Steyerl there, of course. In her writing, she says, “The poor image embodies the afterlife of many former masterpieces of cinema and video art.” I read her talk about it very eloquently, and I said, “I get that.” If I just swap all of that for the voice rather than the .JPG, then it’s the same thing. I’m seeing the some of the same advantages in the lack of quality as she’d seen. That really resonated with me.
What’s next for you?
Lee Gamble: It’s mainly the live show. I’m working with Candela Capítan, a performance artist and choreographer. I looked at her work and thought it was amazing, in terms of how it relates to the body, repetition, alienation, and the gaze and technologized image. I wanted to think about how to present Models live, and I didn’t want to be onstage with a laptop. From day one, I wanted to bring a physical body back into space—but not mine. Humans gave me the information; I simulated and disembodied them; and, now, I ought to put them back in. A lot of the movement is informed by the ideas of cloning, mirrors, and how people replicate each others’ movements. We tried to get twins to perform in the live show.
It’ll premier at Unsound in Poland, and then in London. I wanted to develop the choreography around the idea of how things learn—sometimes they learn badly. Obviously, Candela has brought a huge amount to the work. I gave them a PDF of ideas, and they read it; I played the music to them, and they improvised. It was very moving: I’d never seen anyone do that to my work before. So, I thought, “I want to get on this.”
I’m DJing as regularly as I can, just banging it out and enjoying myself. I’m really enjoying painting. I’m hoping to do some installation work with these voices as well. I don’t know if it’s just me, but I feel like my life moves in ten-year cycles. I did therapy last year for the first time in my life, and that came up. It was really helpful. I was really glad that people pushed me to do it. I’m not against it, of course, but I come from a background where that wasn’t a thing. It took me a long time, but I needed it, and it was super helpful. One of the things that came up is the idea of ten-year cycles in my life. Diversions was 2012, and that was a moment where I did a bunch of other stuff. Now, it’s 2023, and there’s Models, and I’m painting again. Now, I’m ready to boot them glass ceilings as hard as I can for the next ten years.
It’ll premier at Unsound in Poland, and then in London. I wanted to develop the choreography around the idea of how things learn—sometimes they learn badly. Obviously, Candela has brought a huge amount to the work. I gave them a PDF of ideas, and they read it; I played the music to them, and they improvised. It was very moving: I’d never seen anyone do that to my work before. So, I thought, “I want to get on this.”
I’m DJing as regularly as I can, just banging it out and enjoying myself. I’m really enjoying painting. I’m hoping to do some installation work with these voices as well. I don’t know if it’s just me, but I feel like my life moves in ten-year cycles. I did therapy last year for the first time in my life, and that came up. It was really helpful. I was really glad that people pushed me to do it. I’m not against it, of course, but I come from a background where that wasn’t a thing. It took me a long time, but I needed it, and it was super helpful. One of the things that came up is the idea of ten-year cycles in my life. Diversions was 2012, and that was a moment where I did a bunch of other stuff. Now, it’s 2023, and there’s Models, and I’m painting again. Now, I’m ready to boot them glass ceilings as hard as I can for the next ten years.