WEBVTT
00:00:19.839 --> 00:00:21.359
Welcome to the future of you.
00:00:21.760 --> 00:00:31.120
Based in Berlin, Holly Herndon creates groundbreaking music that combines her own voice with cutting-edge tools like AI and other custom-built software.
00:00:31.359 --> 00:00:46.000
With a PhD from Stanford University and a deep-rooted passion for electronic composition, she has released acclaimed albums such as Platform and Proto, pushing the boundaries of how we think about sound, identity, and creativity.
00:00:46.240 --> 00:00:54.960
Through projects like Holly Plus, she invites us to rethink our relationship with technology art and ourselves in the digital age.
00:00:55.200 --> 00:01:02.880
And in this episode, I speak with her about how artificial intelligence is transforming our understanding of identity.
00:01:03.200 --> 00:01:12.079
Holly takes us on a fascinating journey from her early experiments with AI voice models to her groundbreaking exhibition at the Serpentine Gallery.
00:01:12.239 --> 00:01:24.959
What's particularly interesting is Holly's concept of identity play, where she's moved beyond thinking about AI as just a tool to seeing it as a medium for exploring and extending who we are.
00:01:25.280 --> 00:01:40.959
Her HollyPlus project, which allows others to perform through her, and her provocative ex-Hairy Mutant X character, challenge our assumptions about authenticity and control in the digital age, which you know is just up my street.
00:01:41.280 --> 00:01:49.680
This conversation is a mind-expanding journey into the future of identity, where technology does indeed transform who we can be.
00:01:49.920 --> 00:01:52.400
I hope you enjoy it as much as I did.
00:01:56.719 --> 00:01:59.599
So, Holly Herndon, welcome to the Future Review.
00:01:59.680 --> 00:02:01.200
It's amazing to have you here.
00:02:01.519 --> 00:02:02.640
Thanks so much for having me.
00:02:02.959 --> 00:02:03.439
Not at all.
00:02:03.599 --> 00:02:05.519
I've been fascinated by your work for a long time.
00:02:05.599 --> 00:02:11.120
And I say, you know, welcome to Holly, but uh welcome to Holly Plus as well, and as perhaps several other identities.
00:02:11.280 --> 00:02:20.879
Um Holly, could you just give us a brief description or a resume of where you're at, what you're doing at the minute, and uh like the kind of field you operate in.
00:02:21.039 --> 00:02:24.159
Or maybe, maybe that's not easy to define.
00:02:24.479 --> 00:02:26.159
Yeah, I'm I'll I'll give it a shot.
00:02:26.240 --> 00:02:27.599
Um, so I'm an artist.
00:02:27.759 --> 00:02:30.159
I uh I have a background in computer music.
00:02:30.319 --> 00:02:32.240
Um I that's what I studied.
00:02:32.319 --> 00:02:35.919
Um and I've been working with uh AI for about 10 years now.
00:02:36.000 --> 00:02:43.360
And more recently, I've kind of uh made my foray into the um to the art world with a solo exhibition at the Serpentine in London.
00:02:43.520 --> 00:03:00.960
And my exploration of AI for the past 10 years has taken me down some um pretty psychedelic journeys, um, one of which being exploring my identity in AI models and um this idea of being able to create models um based on my own IP.
00:03:01.280 --> 00:03:05.520
It's an idea I call identity play where other people can kind of perform through me.
00:03:05.680 --> 00:03:09.840
It's a project called Holly Plus, who you also welcomed to the program here.
00:03:10.159 --> 00:03:13.360
So I would say that I'm an artist working with technology.
00:03:13.599 --> 00:03:14.719
Yes, because you studied that.
00:03:14.800 --> 00:03:16.000
Did you study technology?
00:03:16.080 --> 00:03:16.800
Am I right in saying?
00:03:17.120 --> 00:03:19.360
Yeah, it's technically I studied computer music.
00:03:19.439 --> 00:03:28.479
So um there's this really special program at Stanford where I did my PhD, which is um it's a very technical program, but it's also housed within the music department.
00:03:28.639 --> 00:03:37.680
So I really feel like I'm a kind of translator between um technical fields and cultural fields, and that's where I feel the most comfortable.
00:03:38.000 --> 00:03:41.360
And maybe that is also more akin to the future.
00:03:41.520 --> 00:03:44.560
We might, we might get onto that towards towards the end of the conversation.
00:03:44.719 --> 00:03:50.479
But I first um discovered you through Holly Plus, and it wasn't really your TED talk or anything.
00:03:50.560 --> 00:04:04.479
It was just, I think it was a few years ago when I was researching the book, and I couldn't find that many people who were exploring identity, but there were um a couple of people in the sort of art techno music field.
00:04:04.719 --> 00:04:08.800
How did you come about to conceive of Holly Plus?
00:04:09.199 --> 00:04:16.800
Well, as I mentioned, I've been working with AI for about 10 years, and this is really before it became really productized.
00:04:16.879 --> 00:04:22.240
So when most people think of AI today, they think of ChatGPT or Dali or Midjourney.
00:04:22.399 --> 00:04:27.360
These are really large general models that are trained by scraping data from the internet.
00:04:27.839 --> 00:04:32.480
But back in the day, none of those products were available, so you really had to train your own models.
00:04:32.639 --> 00:04:36.000
And one of the first questions would be: what do you want to train the model on?
00:04:36.160 --> 00:04:37.199
What's the data set?
00:04:37.439 --> 00:04:46.399
And you know, with a background in composition and understanding the music history of kind of sampling, I didn't feel comfortable using anyone else's data.
00:04:46.480 --> 00:04:53.199
So I just wanted to use my own data, put together an ensemble where we recorded training data specifically for training a model.
00:04:53.360 --> 00:04:59.519
And when I started working with specific voice models, I felt the most comfortable training on my own voice.
00:04:59.680 --> 00:05:10.240
So a voice model is essentially an AI model that learns all of the sounds of your particular voice that can then emulate and sing in your natural singing voice.
00:05:10.399 --> 00:05:13.279
And so I felt weird asking someone else to do that.
00:05:13.519 --> 00:05:16.560
It made the most sense to kind of experiment on my own voice.
00:05:16.720 --> 00:05:19.519
And I am a vocalist, you know, I've released records where I'm singing.
00:05:19.680 --> 00:05:21.920
I never really thought of myself as purely a vocalist.
00:05:22.000 --> 00:05:26.000
I feel like there are many, many people with much more beautiful voice than I have.
00:05:26.160 --> 00:05:32.399
But this felt like the kind of, I don't know, the ethical approach to use my own IP in that in that instance.
00:05:32.720 --> 00:05:33.040
Yes.
00:05:33.279 --> 00:05:47.920
And and how important is the emerging sort of field of decentralized technologies, whether it's AI or blockchain or the new era, if you like, of less sort of uh linear hierarchical technologies and the new sort of networked technologies?
00:05:48.160 --> 00:05:50.560
Has that made much more of your work possible?
00:05:50.879 --> 00:05:55.199
Yeah, I I think this has been a really fundamental part of what I've been working on.
00:05:55.360 --> 00:06:06.000
I mean, one way that we like to think about it, so I released an album in um 2015 called Platform, and then an album in 2019 called Proto, short for protocols.
00:06:06.079 --> 00:06:20.879
And I see this kind of shift from the platform era to the protocol era, the platform being the, you know, you think of like Spotify or Instagram or these specific products that are, that are platforms that kind of determine how artistic expression is distributed.
00:06:20.959 --> 00:06:29.600
And so different decisions that the platform owners will make have a direct impact on how media is made to optimize for that platform.
00:06:29.839 --> 00:06:35.439
I like to think about us entering this kind of protocol era where we're working at a layer underneath.
00:06:35.519 --> 00:06:51.439
And if you think about protocols being, you know, something like Ethereum would be a protocol that you can actually build applications on top of, but you can also think of a protocol as like just a way of behaving, like certain kind of like social manners that we have are also just kind of like social protocols.
00:06:51.519 --> 00:06:54.240
It's a kind of rule set or an understanding of how things work.
00:06:54.399 --> 00:07:05.040
So I like to think about the creation and distribution and all of the kind of facets that go into the art making process as part of the protocol, as part of the artwork itself.
00:07:05.279 --> 00:07:13.040
And yeah, I'm hoping that we're entering an era where more artists are able to have more control over all of those facets of their output.
00:07:13.360 --> 00:07:15.839
Yeah, do you think many do at the moment?
00:07:16.000 --> 00:07:23.199
I mean, there are some people working in this space, but not many who as are as high profile and prolific as you.
00:07:23.600 --> 00:07:25.759
Do you think that's going to change over time now?
00:07:26.319 --> 00:07:31.680
Um, yeah, I mean, I think a lot of the work that I've done requires really boring research.
00:07:32.560 --> 00:07:35.120
And I think a lot of people, you know, fair enough.
00:07:35.680 --> 00:07:42.879
People are busy and they don't want to have to read uh machine learning papers and watch, you know, academic conferences on YouTube.
00:07:42.959 --> 00:07:45.680
That's like a pretty, pretty specific thing.
00:07:45.839 --> 00:07:52.240
But um yeah, I mean, I think there, yeah, there's a technical barrier for entry for a lot of these things.
00:07:52.480 --> 00:08:02.480
But I think as, you know, I feel like we learned so much from the internet becoming a really ubiquitous, all-present part of our daily lives.
00:08:02.560 --> 00:08:14.000
And I think people started to kind of wake up as to how technical decisions or platform decisions, design decisions were impacting their personal lives really directly.
00:08:14.240 --> 00:08:18.480
I think people have a desire to have more agencies within these systems.
00:08:18.720 --> 00:08:26.079
So yeah, I do think people are are becoming more and more interested in understanding the tech and understanding how they can have agency with it.
00:08:26.399 --> 00:08:31.759
That's interesting because some of your work in the past has focused on the theme of surveillance, hasn't it?
00:08:31.920 --> 00:08:35.759
You know, how we are being like monitored or surveilled, privacy public.
00:08:36.159 --> 00:08:48.639
What does it mean, do you think, to be machine readable, to be noticed, to be captured, monitored, analyzed, and maybe even redefined or represented or represented in this kind of emerging tech world?
00:08:48.879 --> 00:08:49.919
Yeah, that's a great question.
00:08:50.080 --> 00:08:54.320
I actually have a book coming out in the next couple weeks called All Media is Training Data.
00:08:54.399 --> 00:08:59.759
And that that doesn't some people read that and they they they think I'm saying all it is is training data.
00:08:59.840 --> 00:09:00.960
And that's not what I'm saying.
00:09:01.039 --> 00:09:18.960
I'm saying that, yeah, as soon as something is captured in media, like, you know, this conversation that we're having, or as soon as you take a photograph or a video of anything, then it becomes machine legible, and then it can become part of a training canon, you know, to train a model that you may never even interact with.
00:09:19.120 --> 00:09:28.240
So a lot of the data that we're placing in social media, online, you know, this is already going to train different models that we're just not seeing or necessarily interacting with.
00:09:28.399 --> 00:09:33.679
I think it's a pretty profound moment in our kind of on our human timeline.
00:09:34.320 --> 00:09:36.320
Yeah, I would completely agree.
00:09:36.480 --> 00:10:05.120
I mean, I think people are kind of used to like, oh, maybe something's extracted from a PDF about me or a Wikipedia page, but all this kind of digital phenotyping, treating us almost like patients, capturing some of the atmospheric or environmental data around us, the way we move, the way I was talking to somebody about, you know, motion prints in VR, for example, the way in which our gesture with our head and our hands, well, you know, it's so individual that it's like a fingerprint and they call it motion print.
00:10:05.360 --> 00:10:07.519
I feel like sometimes there's nowhere to escape.
00:10:07.679 --> 00:10:16.879
Maybe that's the dystopian view of it, and obviously you're looking at it primarily through the opportunities, but it is a very, as you say, very profound shift, isn't it?
00:10:17.200 --> 00:10:24.639
Yeah, I think I think it all depends on kind of where we end up in terms of, you know, who's using this data and for what purpose.
00:10:24.879 --> 00:10:33.679
You know, on the kind of dystopian side, okay, you just have giant corporations that are just, you know, scraping up all of our data and they're monetizing it in ways that we are don't participate in.
00:10:33.759 --> 00:10:35.600
And that's that's the kind of like dark image.
00:10:35.679 --> 00:10:52.960
But there's also a, I think, a beautiful way to approach AI that requires a lot of um work being put in the public domain, having public domain models that we can then fine-tune and build on top of, participating and contributing to something that feels like it's collective, like a collective accomplishment.
00:10:53.120 --> 00:11:02.879
I like to think of AI as us in aggregate, as a collective human accomplishment rather than a kind of like alien other that's here to just uh replace us.
00:11:03.120 --> 00:11:16.480
So I think if we can think about it as a kind of really sophisticated coordination and collaboration mechanism, and we can find ways to to make it work for everyone and be fair, then I think it maybe will be less scary for people.
00:11:16.799 --> 00:11:22.320
I like that idea of complicity, that's the balance between the human and the machine elements.
00:11:22.480 --> 00:11:33.679
I think you've done that really well with Holly Plus, because obviously you've invited people to sort of be the um, I don't know, the stewards of your voice in the way that they deliver it, haven't you?
00:11:33.840 --> 00:11:39.039
I mean, and I think that's uh I don't know what was what was the reception to that when you suggested that?
00:11:39.120 --> 00:11:56.399
Because I I saw a lot of people online saying, oh, but what does this mean for sort of rights, copyright, royalties, you know, and and actually I I think, and correct me if you're wrong, but you were saying you can use my voice as long as you credit it or give me attribution, but what you do with it is up to you artistically, maybe to a point.
00:11:56.720 --> 00:11:57.200
Exactly.
00:11:57.279 --> 00:11:58.480
And that's what works for me.
00:11:58.639 --> 00:12:00.080
That's not going to work for everyone.
00:12:00.159 --> 00:12:07.279
And that kind of goes back to the conversation we were having earlier about the platform dynamics and you know, this kind of one size fits all for artists.
00:12:07.360 --> 00:12:16.799
I hope that we move to a situation where people are able to really make specific decisions about their own practice, the economics that make sense for them and their communities.
00:12:17.039 --> 00:12:20.879
For me, I was fine with just taking a permissive approach to my IP.
00:12:21.039 --> 00:12:23.440
I was more curious to see how people would use it.
00:12:23.600 --> 00:12:27.440
But I could imagine another vocalist might really not want someone to use their voice.
00:12:27.519 --> 00:12:28.799
And that's totally understandable.
00:12:28.879 --> 00:12:30.799
And that that right should be there.
00:12:30.960 --> 00:12:33.360
At the moment, it's a really gray area.
00:12:33.519 --> 00:12:37.759
You have personality rights as something that has more legal history.
00:12:37.919 --> 00:12:49.679
So if somebody recorded something and then used my name as an artist name, you could sue on those grounds, but it's less difficult to sue on the on the training of a model using my my voice.
00:12:49.919 --> 00:12:55.200
So we'll see where all of this lands in terms of legal, but yeah, that's just what felt right to me.
00:12:55.360 --> 00:12:58.480
I wanted to just experiment and make it as open as possible.
00:12:58.799 --> 00:13:08.000
Did anybody do anything with your voice that you weren't happy with or maybe surprised you or was innovative in a way that you hadn't really expected?
00:13:08.320 --> 00:13:10.960
You know, no one did anything that I was offended by.
00:13:11.120 --> 00:13:15.840
You know, also the voice that we made public was a I I have multiple voice models.
00:13:15.919 --> 00:13:21.279
One of them is more kind of language specific, another one is a real-time instrument that people can sing through.
00:13:21.440 --> 00:13:27.279
And then the one that we made public is a is a more kind of it it deals in more kind of like ambience and texture.
00:13:27.360 --> 00:13:34.879
So it can take a polyphonic input, any kind of sample you can drag onto the website and it'll transfer it into a kind of choir of hollies.
00:13:34.960 --> 00:13:41.279
And because it's more abstract, I wasn't really dealing with having to edit out hate speech or anything like that.
00:13:41.440 --> 00:13:49.600
Of course, that's always a always a question when you're dealing with the the open internet, but really I was surprised at the kind of beautiful approaches that people took to it.
00:13:49.679 --> 00:14:07.360
I mean, one person recorded their train ride and the kind of screeching of the train of the train tracks to the metal, and then transferred that into my voice, and it made this gorgeous choir with the most amazing overtones and just things like that that I wouldn't have thought of myself were just a really nice surprise.
00:14:07.600 --> 00:14:15.679
You see, I think this is one of the really interesting dimensions of AI, that it gives you another perspective once you apply these tools or use these tools.
00:14:15.840 --> 00:14:35.039
And I think Eric Schmidt, of all people, said this about AI agents that I don't know if you agree, but that they will give us another perspective to blend with the perspectives that are delivered through our own human senses and give us a picture or an understanding or even a window into the world that is richer or we might not have had access to before.
00:14:35.440 --> 00:14:48.559
Yeah, I mean, AI models essentially are just able to consume and crunch and make sense of a much broader or a vaster, if I can say that, um, amount of data than the human brain can.
00:14:48.799 --> 00:14:57.919
And so I rather than this kind of replacement thing, I see it as an aid to help us understand the vast amount of amounts of data that we're able to collect these days.
00:14:58.080 --> 00:14:59.679
Uh so yeah, I I think that's true.
00:14:59.759 --> 00:15:04.159
It's also just kind of like a strange mirror to ourselves, a strange reflection.
00:15:04.320 --> 00:15:07.120
It's not exactly us, it's a kind of mutant form of us.
00:15:07.200 --> 00:15:11.519
And so by interacting with that mirror world, we understand ourselves a little bit more.
00:15:12.080 --> 00:15:12.879
It's weird, isn't it?
00:15:12.960 --> 00:15:16.080
Because it's got an intimacy, but also an alien nature to it.
00:15:16.240 --> 00:15:19.759
It is that weird, maybe that's the uncanny thing about it.
00:15:20.639 --> 00:15:25.120
Yeah, I think the I think the intimacy is that it is actually human created.
00:15:25.200 --> 00:15:37.120
It's all of the data is from humans, and so it's basically you're just interacting, you're collaborating with vast amounts of people, and our brains haven't really evolved to be able to conceive of of that.
00:15:37.200 --> 00:15:48.399
You know, we come from tribal groups where we're used to having only so many kind of um social connections, and now all of a sudden we're able to collaborate with millions and millions of people at once.
00:15:48.480 --> 00:15:51.360
And that that I think that that scrambles our brains a little bit.
00:15:51.679 --> 00:15:56.159
Which is one of the reasons I love the language that you use for these things like spawning.
00:15:56.480 --> 00:15:58.879
I love because it's biological, isn't it?
00:15:59.200 --> 00:16:00.480
Not technological.
00:16:02.000 --> 00:16:08.480
Yeah, so spawn uh it started out as spawn when I when I made proto and we were working with our ensemble in Berlin.
00:16:08.559 --> 00:16:11.840
We wanted to have a an ensemble member trained on our voices.
00:16:12.080 --> 00:16:17.200
And so we um we came up with the name Spawn and called her our AI baby.
00:16:17.600 --> 00:16:21.120
Um, and spawn has since kind of grown up into a verb.
00:16:21.200 --> 00:16:25.440
This word spawning is uh is what we like to use for some of these generative AI tools.
00:16:25.600 --> 00:16:35.440
So instead of sampling, which is a kind of one-to-one mechanical reproduction, we think about spawning as being able to create infinite media from a kind of seed source.
00:16:35.519 --> 00:16:40.480
And so spawning felt like a much more kind of descriptive version of what's happening there.
00:16:40.639 --> 00:16:57.039
And then spawning has evolved once again and is now the name of an organization that I co-founded with my partner Matt Dryhurst and um Jordan Meyer, who's a machine learning engineer, where we're working to uh advocate for um data manners in the machine learning space.
00:16:57.200 --> 00:17:02.320
So spawn, spawning, spawning has had quite an evolution over the last several years.
00:17:02.559 --> 00:17:03.840
I love it, yeah.
00:17:04.079 --> 00:17:05.119
So so is that right?
00:17:05.200 --> 00:17:07.920
So it allows artists to opt in and out of data sets.
00:17:08.160 --> 00:17:08.720
How does that work?
00:17:08.880 --> 00:17:10.960
How does an artist access that?
00:17:11.279 --> 00:17:11.519
Sure.
00:17:11.680 --> 00:17:15.119
So at the moment it's in the image space and uh in the text space.
00:17:15.279 --> 00:17:19.920
And so you're able to go in and at the moment we don't have opt-in, but we do have opt-out.
00:17:20.160 --> 00:17:28.400
So if you don't want for your images on your website, for example, to be included in machine learning training, you can add that to your website using our tools.
00:17:28.559 --> 00:17:34.960
You can see if your data is already being used in training sets by going to have ibentrained.com.
00:17:35.200 --> 00:17:36.640
Yeah, just check out spawning.
00:17:36.720 --> 00:17:38.559
We we have several different things in the works.
00:17:38.640 --> 00:17:41.920
We're also working on uh the public domain at the moment.
00:17:42.079 --> 00:17:45.279
We see that as a really important first step for an opt-in.
00:17:45.839 --> 00:17:56.400
So if you have a public domain base layer model, then you can fine-tune that model on your own work and feel like you as an artist have actual ownership over that model.
00:17:56.559 --> 00:18:06.880
So you can then sell the model or you can license the model or you can come up with whatever economic interaction that you want around that model, knowing that you haven't infringed on anyone else's copyright.
00:18:07.039 --> 00:18:11.359
We see that as a really important first step to getting a lot of artists into the space.
00:18:11.680 --> 00:18:12.720
That's amazing.
00:18:12.960 --> 00:18:18.000
So have you got a big tech team then, or is it just you and Matt and a few collaborators doing this?
00:18:18.240 --> 00:18:20.079
It is definitely not just me and Matt.
00:18:20.240 --> 00:18:22.960
We have a whole team of wonderful people.
00:18:23.119 --> 00:18:26.480
You know, we've always been collaborative oriented.
00:18:26.640 --> 00:18:27.279
Can I say that?
00:18:27.440 --> 00:18:28.880
We've always been collaborators.
00:18:28.960 --> 00:18:32.640
Um, and I've I've always really enjoyed working with teams of engineers.
00:18:32.799 --> 00:18:34.799
So spawning is its own kind of thing.
00:18:34.880 --> 00:18:39.039
It's run in the United States, and there's a whole team of people working on that full-time.
00:18:39.200 --> 00:18:45.440
The work that we did at The Serpentine, there are three different novel models that we're working with for that show.
00:18:45.599 --> 00:18:49.599
And we're working with EarCAM, which is a research institute in Paris.
00:18:49.759 --> 00:18:58.319
Um, we're working with Algamous, which is a research team in Lille, Stable Audio, which is a decentralized kind of audio research organization.
00:18:58.400 --> 00:19:04.640
So we have collaborators and engineers that we're in constant contact with that really make everything possible.
00:19:04.960 --> 00:19:05.920
That's amazing.
00:19:06.079 --> 00:19:07.680
That really is a network, isn't it?
00:19:07.839 --> 00:19:09.920
A kind of vibrant, dynamic network.
00:19:10.079 --> 00:19:11.920
Let's talk about your work at the Serpentine.
00:19:12.000 --> 00:19:13.680
I think it's on till February, isn't it?
00:19:13.759 --> 00:19:14.079
Mm-hmm.
00:19:14.319 --> 00:19:18.240
Because I really want I yes, I'm I'm hoping to come and see it before it it finishes in February.
00:19:18.319 --> 00:19:19.359
I know it's live now.
00:19:19.599 --> 00:19:21.119
Could you tell us a little bit about it?
00:19:21.200 --> 00:19:23.759
Because it's quite the construction.
00:19:24.160 --> 00:19:24.559
It is.
00:19:24.640 --> 00:19:26.880
It was a uh yeah, it's quite the project.
00:19:27.119 --> 00:19:28.000
Beautiful as well.
00:19:28.240 --> 00:19:28.640
Thank you.
00:19:28.720 --> 00:19:30.319
It took over a year to put together.
00:19:30.480 --> 00:19:33.359
You know, we worked really closely with the arts and technology team there.
00:19:33.519 --> 00:19:41.359
Hans Ulrich is so amazing at understanding what's happening and staying on top of all of the kind of cutting-edge developments.
00:19:41.440 --> 00:19:46.319
And so he really trusted us with the vision to work with the arts and technology team there.
00:19:46.559 --> 00:19:53.839
And what we wanted to do is so many shows really focus on just the output of a model because it's so impressive.
00:19:54.160 --> 00:20:01.519
What we wanted to do was focus on the process of the making of a model because we see making models as an artwork in and of themselves.
00:20:01.839 --> 00:20:05.359
Every step in the process, there's a moment for artistic intervention.
00:20:05.519 --> 00:20:17.519
So the creation of really deliberate training data as artworks, that going into, you know, training with uh choirs across the UK, those are performances, those are artworks.
00:20:17.759 --> 00:20:25.920
The decisions that we make in categorizing the data for the model, the model itself, and in interacting with the model as an output.
00:20:26.160 --> 00:20:31.599
Each of those phases we really wanted to show and give the visitor an experience of in the space.
00:20:31.839 --> 00:20:35.359
So we put together this really wild um production.
00:20:35.519 --> 00:20:41.920
Um, we wanted to make a choral AI model that could be collectively owned by a group of choirs.
00:20:42.000 --> 00:20:45.200
So we recorded 15 choirs across the UK.
00:20:45.519 --> 00:20:49.279
We hopped in a car and we drove around and we recorded all of these choirs.
00:20:49.519 --> 00:20:57.599
And so when you go into the exhibition in one of the rooms, you can hear the kind of field recording and the ethnographic uh recordings that we were making on site.
00:20:57.759 --> 00:21:02.880
So you can hear how the training of a model, it's not just these kind of random data points.
00:21:02.960 --> 00:21:09.119
It's actually people being creative and making meaning and making work together in a space.
00:21:09.279 --> 00:21:10.799
That's what these data points are.
00:21:10.880 --> 00:21:13.599
They're not these cold little bits of media.
00:21:13.839 --> 00:21:15.440
Um, so you can experience that.
00:21:15.519 --> 00:21:19.599
And then we created an organ out of uh GPU fans.
00:21:19.680 --> 00:21:27.279
So the sound of our studio is just the constant whirring of GPU uh fans that's cooling our GPU as it's crunching all the numbers.
00:21:27.359 --> 00:21:34.319
So you kind of can experience the compression phase and the power that goes into to training these models through the organ.
00:21:34.559 --> 00:21:42.240
And you can hear the output of the model that we created with all of the recordings across the UK and our own personal archive.
00:21:42.319 --> 00:21:49.279
We made a model that we can prompt and create fully coherent, beautiful choral songs, infinite choral songs.
00:21:49.359 --> 00:21:51.839
You can hear those spatialized in the space.
00:21:52.000 --> 00:21:54.559
And then also the visitor can interact with the model.
00:21:54.720 --> 00:21:59.759
That's a really important piece of this that these models they're designed to be interacted with.
00:22:00.079 --> 00:22:00.240
With.
00:22:00.319 --> 00:22:09.279
So we wanted to allow the public to see what that feels like without introducing a screen because there's so much screen work when it comes to people interacting with AI.
00:22:09.440 --> 00:22:19.839
So we wanted to eliminate the screen and just use microphones so people can sing in through a microphone and they can kind of navigate through the choral model and hear the choral model singing back to them.
00:22:20.000 --> 00:22:33.279
So we hope that the visitor gets the experience of the kind of full stack training of a model and maybe can see themselves in that and how it might be beautiful to collaboratively train a model, you know, with other people.
00:22:33.920 --> 00:22:35.119
It really is amazing.
00:22:35.279 --> 00:22:42.240
Um is that idea of the process being creative and also collaborative and transparent to people?
00:22:42.640 --> 00:22:47.519
A bit like your idea about protocols being habits and ethics.
00:22:48.160 --> 00:22:48.640
It is.
00:22:49.039 --> 00:22:49.759
A hundred percent.
00:22:49.839 --> 00:22:50.079
Yeah.
00:22:50.160 --> 00:22:54.319
I mean, I feel like the design of the exhibition, we were thinking very closely.
00:22:54.480 --> 00:22:55.839
What what is the protocol?
00:22:56.000 --> 00:22:56.799
What's the approach?
00:22:56.960 --> 00:23:04.720
What's the kind of rules that we want to put together for how this functions in a space, for how we deal with the individual participants?
00:23:04.880 --> 00:23:09.039
That was a really important part of the kind of protocol design of the show.
00:23:09.279 --> 00:23:15.359
And one piece of it, it might sound really bureaucratic and kind of boring, but we worked really closely with the Serpentine.
00:23:15.440 --> 00:23:26.319
Their team is really at the forefront of thinking through, from a legal perspective, how you can co-own and um collectively kind of shepherd a data set.
00:23:26.480 --> 00:23:38.240
So they they put a lot of energy into legal work and legal frameworks to think through how can we invite the people who participated in these recordings to play a part in stewarding this data moving forward.
00:23:38.480 --> 00:23:39.279
So yeah.
00:23:39.680 --> 00:23:40.720
What did they think of it?
00:23:40.799 --> 00:23:42.160
What were the participants?
00:23:42.480 --> 00:23:46.480
You know, it was such a wonderful, wide range of people.
00:23:46.640 --> 00:23:54.960
A lot of people were really surprised at the variety of ages that were involved because this was a I think a lot of a lot of projects would just record in London.
00:23:55.039 --> 00:24:15.039
And we did do some recordings in London, but we really hit the road and went across the UK and we went to rural communities and we met with older people and people who had never heard of AI, and we tried to get a really kind of broad cross-section of experience, and people had really interesting thoughts and concerns.
00:24:15.279 --> 00:24:21.200
And in a way, it was a it was informative to us to understand, you know, kind of where people are at.
00:24:21.279 --> 00:24:26.799
But I think we also were helping people think through their own data footprint online.
00:24:27.200 --> 00:24:39.920
Like some people, for example, would come to the project and be a little bit wary of training in AI, but that same person might upload their music to YouTube, not understanding that that is already training in AI, just without their knowledge.
00:24:40.079 --> 00:24:45.839
So there was a lot of information that we were able to share with people to better understand how this landscape is unfolding.
00:24:46.160 --> 00:24:47.119
It's interesting, isn't it?
00:24:47.200 --> 00:24:57.440
Because I think it's been really difficult to get people to engage with the idea or the notion that identity is transformed by the media that you use.
00:24:57.599 --> 00:25:04.480
And I guess that's a great example of that actually happening, although you might not necessarily use the language identity or personal identity.
00:25:04.640 --> 00:25:08.079
Maybe people start to realise that is happening to them.
00:25:08.160 --> 00:25:13.279
Like elements of the self are extended or augmented or whatever.
00:25:13.359 --> 00:25:15.440
I think I think that's happened with avatars as well.
00:25:15.599 --> 00:25:18.400
I think now people are messing about and playing with avatars.
00:25:18.480 --> 00:25:20.640
People can start to understand, oh, hang on a minute.
00:25:20.799 --> 00:25:28.559
This can happen to my voice, or my voice can, you know, suddenly I can speak Japanese uh in my own accent, even though I can't, sort of thing.
00:25:28.799 --> 00:25:32.559
Do you think it's going to get to a point where there is a better understanding of all this?
00:25:32.880 --> 00:25:40.480
Yeah, I mean, it's that's something that I'm that I think about a lot, that I'm that I'm actively working to help people understand.
00:25:40.880 --> 00:25:49.599
One of the things that I think about is the kind of difference between the data point and the actual person, like the fleshy, messy person that's in the real world.
00:25:49.759 --> 00:25:54.400
So one of the projects that I worked on this past year was called X Harry Mutant X.
00:25:55.440 --> 00:26:00.079
And so that's a project that we um exhibited at the Whitney Biennial this year.
00:26:00.319 --> 00:26:13.759
So this is a really funny kind of super superhero character that I created for myself because I, my, my name, Holly Herman, me as an artist, I meet the threshold of being a concept in some of these large models.
00:26:13.839 --> 00:26:19.200
So if you go to an image model and you type in my name, a kind of essence of me will appear.
00:26:19.519 --> 00:26:29.039
And because these models are trained by just scraping the open internet, it's just collected a bunch of images of me and created a kind of pastiche of what it understands of me.
00:26:29.119 --> 00:26:37.119
So because I'm not as famous as someone like Dolly Parton, Dolly Parton would have all of her full features and would be much more fleshed out.
00:26:37.279 --> 00:26:40.240
Me, I just kind of appear as an orange blob.
00:26:40.400 --> 00:26:47.519
So, you know, where wherever my face might always be different, but I'll always have this giant orange blob of hair.
00:26:47.599 --> 00:26:54.400
So I was really asking myself, okay, how much agency do I have in this, in determining who I am in these systems?
00:26:54.559 --> 00:27:00.400
If this is gonna be the net, if this is gonna be the next layer of the internet, do I get to decide who I am here?
00:27:00.559 --> 00:27:09.119
And so when I did some experiments and tried introducing some images of me that were outside of that with entirely different hair, it was kind of thrown out as noise.
00:27:09.279 --> 00:27:16.079
So I realized if I wanted to have any agency here, I had to almost lean into my cliche, lean into my pastiche.
00:27:16.480 --> 00:27:24.240
So we designed this wild costume X Harry Mutant X, which is this kind of golem superhero with giant orange hair.
00:27:24.480 --> 00:27:32.960
And then I created a model with my partner Matt Dryhurst of that character that people can actually go to the Whitney website and prompt through.
00:27:33.119 --> 00:27:39.599
And no matter what you prompt, you'll always come up with something kind of X Harry Mutant X like in the image.
00:27:39.680 --> 00:27:42.400
It's kind of infected with my giant orange hair.
00:27:42.559 --> 00:27:45.440
And each of these images are tagged with my name.
00:27:45.680 --> 00:28:01.359
So all of these images then are online, and the next time a large image model scrapes the internet and tries to understand me as a concept, it will encounter all of these giant X Harry Mutant X images of me, and that will transform who I am in the model.
00:28:01.680 --> 00:28:18.400
So it's a kind of long-form performance between me and the public and machine learning companies, and it's a really fun way to kind of explore identity and who I am and and the kind of difference between the person and the data points and the digital footprint that we leave on the internet.
00:28:19.039 --> 00:28:23.119
Oh, I love that because this is the area I'm really interested in.
00:28:23.200 --> 00:28:31.039
And it reminds me a little bit of database animals, Azuma about the Otaku and the Otaku's obsession with anime and mango and obviously.
00:28:31.200 --> 00:28:43.359
And I'm I really see, let me know if you agree or not, but this sort of breakdown of us as people into this database of kare mo, mo elements, like the attributes, the elements that make us up.
00:28:43.440 --> 00:28:51.119
And I mean, you just said, you know, orange hair, for example, and they're very much like these characters having loved elements, care mo.
00:28:51.519 --> 00:28:58.640
And how do these elements get mixed, remixed, arranged, and rearranged together, and then spun out in different contexts?
00:28:58.880 --> 00:29:18.240
And the idea is that in this fan world, there is no original, there are no copies, like the parodies are just as important as the original character, but these elements are kind of constructed and reconstructed and rearranged in lots of different ways, and that's that's maybe how we should think about ourselves in the digital environment.
00:29:18.480 --> 00:29:26.640
I mean, I know it's a kind of you know atomistic sort of way in these individual attributes, but it is that a way forward for us to think about the self?
00:29:26.960 --> 00:29:29.359
Yeah, I think that's a really interesting reference.
00:29:29.519 --> 00:29:36.640
I think it kind of intersects with this idea of fandoms and different characters that allow themselves to be kind of remixed and reinterpreted.
00:29:36.799 --> 00:29:43.359
Machine learning models are designed to be interacted with, they're designed to be prompted.
00:29:43.599 --> 00:29:55.440
And so this kind of idea of authorship just being a one-way stream or just being one person, you know, in order to make the model, you're usually drawing from many, many people's creative input in the first place.
00:29:55.519 --> 00:30:04.480
And then when you're inviting people to interact with the model, then you have this kind of like fandom moment where people, the new work is it's both you and it's the person who prompted.
00:30:04.559 --> 00:30:06.880
And then it's this interesting blending.
00:30:06.960 --> 00:30:13.519
And I think, yeah, I think that we can learn a lot from how fandoms have proliferated in the last several years.
00:30:13.759 --> 00:30:15.119
Yeah, no, totally.
00:30:15.279 --> 00:30:21.759
You mentioned collaboration, and I know obviously you've you've spoken a few times about the the people and groups that you work with.
00:30:21.920 --> 00:30:23.920
What about your podcast, Interdependent?
00:30:24.240 --> 00:30:24.880
Yes, yes.
00:30:25.039 --> 00:30:25.599
Oh gosh.
00:30:25.759 --> 00:30:26.480
Interdependence.
00:30:26.559 --> 00:30:29.279
I miss interdependence, but we just kind of ran out of time.
00:30:29.359 --> 00:30:31.279
It was such a big part of our COVID time.
00:30:31.519 --> 00:30:32.720
Will you carry on doing it?
00:30:32.799 --> 00:30:34.720
Because you had some fascinating guests.
00:30:35.119 --> 00:30:36.079
I really enjoyed it.
00:30:36.160 --> 00:30:40.160
I would love to be able to, yeah, maybe Holly Plus can take it on the Sunday.
00:30:40.720 --> 00:30:42.319
That's the point of our avatars, isn't it?
00:30:42.559 --> 00:30:45.519
Go off and do things or agents that we haven't got time to do.
00:30:45.839 --> 00:30:46.319
Exactly.
00:30:46.480 --> 00:30:48.559
You know, I won't, I I'll never say never.
00:30:48.640 --> 00:30:53.920
At the moment, we we don't have time to to start it again, but it's something that I I very much miss.
00:30:54.079 --> 00:30:59.200
It was one of those things during COVID where we were so used to being touring musicians and traveling around so much.
00:30:59.279 --> 00:31:01.759
We missed, you know, meeting new people all the time.
00:31:01.839 --> 00:31:06.799
And it was one of those like socially acceptable ways to call someone that you find interesting and just talk for a while.
00:31:07.759 --> 00:31:09.359
That's why I that's why I got in touch with you.
00:31:09.599 --> 00:31:10.240
Yeah, exactly.
00:31:10.319 --> 00:31:10.559
Yeah.
00:31:10.640 --> 00:31:14.160
So it's honestly it's a great, it's a great, it was a great social outlet for us.
00:31:14.240 --> 00:31:15.440
And I definitely miss it.
00:31:15.680 --> 00:31:22.799
But is there anybody on there that you thought, or an idea or a concept that was just so out there, challenging or amazing and inspiring?
00:31:23.039 --> 00:31:24.880
Oh gosh, I don't know if I would single out.
00:31:25.119 --> 00:31:26.480
There were just so many good ones.
00:31:26.640 --> 00:31:31.440
Um, you know, one person that for some reason is popping in my mind right now is Jacob Horn from Zora.
00:31:31.839 --> 00:31:41.200
He's been thinking about these kind of decentralized identities and co-ownership ideas and and IP and new ways to approach that for a really long time.
00:31:41.279 --> 00:31:44.480
And I always he comes from a different, very different perspective than I do.
00:31:44.559 --> 00:31:47.759
And I always feel like it's a fruitful exchange with him.
00:31:48.079 --> 00:31:59.279
So, Holly, I know I've got to let you go in a minute, but what do you think the future of identity holds if we were to look sort of 20 or 30 years ahead or like mid-century?
00:31:59.519 --> 00:32:07.440
Where do you think we'll be in terms of understanding ourselves, representing ourselves, like expressing ourselves as artists like yourself?
00:32:08.079 --> 00:32:10.559
I think this is a really interesting philosophical question.
00:32:10.720 --> 00:32:22.000
I almost wish my partner, Matt Dryhurst, would be able to answer it because he's been thinking a lot about how the concept of identity and individuality has really shifted throughout the ages of humanity.
00:32:22.160 --> 00:32:25.920
There's this interesting book that he's been reading called The Creation of the Individual.
00:32:26.079 --> 00:32:46.640
Um, so I think one thing that might change is how we perceive ourselves as individual identities and might perceive ourselves more as kind of networks and relationships with other people and also just our own biology, how we might understand that depending on the the way that you look at someone, if like with a naked eye, it looks like a single person.
00:32:46.720 --> 00:32:55.279
But if you look through a microscope, there's all these different interconnected desires and microorganisms that are kind of interconnectedly working together.
00:32:55.440 --> 00:33:03.599
So, in a way, we as selves are also a kind of web of collaborators making our whole bodies work.
00:33:03.759 --> 00:33:06.640
I think that identity is going to get really weird.
00:33:06.880 --> 00:33:08.799
And I I very much welcome that.
00:33:09.039 --> 00:33:32.240
Yeah, it's exciting, but it's also threatening because if we have eight, nine billion people on the planet who are all in an identity crisis, because as you say, if it does get a bit weird and we can't understand the interior of ourselves or selves as um identity continues to be thought of in a more pluralistic way, let's say, I wonder what will happen to people, whether they'll be able to cope with it.
00:33:32.559 --> 00:33:51.200
Yeah, you know, I think not to be like super wide-eyed and uh overly optimistic, but I think that's one of the roles of art, honestly, to help people understand or to help people work through the feelings of of alienation in this like super technologically mediated world that we live in.
00:33:51.359 --> 00:33:54.480
And I hope that we can contribute to that in a meaningful way.
00:33:54.720 --> 00:33:56.400
Oh, I think you do wonderful work there.
00:33:56.480 --> 00:33:57.200
Um thank you.
00:33:57.279 --> 00:33:59.039
It's been fascinating to speak to you.
00:33:59.119 --> 00:34:04.319
And I'm sure it's it's only the tip of the iceberg, what's to come, from Holly and Co.
00:34:04.720 --> 00:34:05.279
So thank you.
00:34:05.440 --> 00:34:06.160
Thank you, Holly.
00:34:06.240 --> 00:34:07.359
It's been great to speak to you.
00:34:07.599 --> 00:34:08.880
Thanks so much, Tracy.
00:34:14.639 --> 00:34:18.000
Thank you for listening to The Future of You, hosted by me, Tracy Follows.
00:34:18.239 --> 00:34:22.079
Be sure to check out the show notes for more info about the topics we covered today.
00:34:22.239 --> 00:34:26.880
If you enjoyed this episode, please like and subscribe wherever you get your podcasts.
00:34:27.039 --> 00:34:30.639
And if you know someone who would love this episode, please share it with them.
00:34:30.880 --> 00:34:36.639
For more on the future of identity in a digital world, visit futuremade.group slash the future of you.
00:34:36.960 --> 00:34:41.760
And to explore the future of everything else, head over to futuremade.group.
00:34:41.840 --> 00:34:45.519
The Future of View podcast is produced by Big Tent Media.
00:00:19.839 --> 00:00:21.359
Welcome to the future of you.
00:00:21.760 --> 00:00:31.120
Based in Berlin, Holly Herndon creates groundbreaking music that combines her own voice with cutting-edge tools like AI and other custom-built software.
00:00:31.359 --> 00:00:46.000
With a PhD from Stanford University and a deep-rooted passion for electronic composition, she has released acclaimed albums such as Platform and Proto, pushing the boundaries of how we think about sound, identity, and creativity.
00:00:46.240 --> 00:00:54.960
Through projects like Holly Plus, she invites us to rethink our relationship with technology art and ourselves in the digital age.
00:00:55.200 --> 00:01:02.880
And in this episode, I speak with her about how artificial intelligence is transforming our understanding of identity.
00:01:03.200 --> 00:01:12.079
Holly takes us on a fascinating journey from her early experiments with AI voice models to her groundbreaking exhibition at the Serpentine Gallery.
00:01:12.239 --> 00:01:24.959
What's particularly interesting is Holly's concept of identity play, where she's moved beyond thinking about AI as just a tool to seeing it as a medium for exploring and extending who we are.
00:01:25.280 --> 00:01:40.959
Her HollyPlus project, which allows others to perform through her, and her provocative ex-Hairy Mutant X character, challenge our assumptions about authenticity and control in the digital age, which you know is just up my street.
00:01:41.280 --> 00:01:49.680
This conversation is a mind-expanding journey into the future of identity, where technology does indeed transform who we can be.
00:01:49.920 --> 00:01:52.400
I hope you enjoy it as much as I did.
00:01:56.719 --> 00:01:59.599
So, Holly Herndon, welcome to the Future Review.
00:01:59.680 --> 00:02:01.200
It's amazing to have you here.
00:02:01.519 --> 00:02:02.640
Thanks so much for having me.
00:02:02.959 --> 00:02:03.439
Not at all.
00:02:03.599 --> 00:02:05.519
I've been fascinated by your work for a long time.
00:02:05.599 --> 00:02:11.120
And I say, you know, welcome to Holly, but uh welcome to Holly Plus as well, and as perhaps several other identities.
00:02:11.280 --> 00:02:20.879
Um Holly, could you just give us a brief description or a resume of where you're at, what you're doing at the minute, and uh like the kind of field you operate in.
00:02:21.039 --> 00:02:24.159
Or maybe, maybe that's not easy to define.
00:02:24.479 --> 00:02:26.159
Yeah, I'm I'll I'll give it a shot.
00:02:26.240 --> 00:02:27.599
Um, so I'm an artist.
00:02:27.759 --> 00:02:30.159
I uh I have a background in computer music.
00:02:30.319 --> 00:02:32.240
Um I that's what I studied.
00:02:32.319 --> 00:02:35.919
Um and I've been working with uh AI for about 10 years now.
00:02:36.000 --> 00:02:43.360
And more recently, I've kind of uh made my foray into the um to the art world with a solo exhibition at the Serpentine in London.
00:02:43.520 --> 00:03:00.960
And my exploration of AI for the past 10 years has taken me down some um pretty psychedelic journeys, um, one of which being exploring my identity in AI models and um this idea of being able to create models um based on my own IP.
00:03:01.280 --> 00:03:05.520
It's an idea I call identity play where other people can kind of perform through me.
00:03:05.680 --> 00:03:09.840
It's a project called Holly Plus, who you also welcomed to the program here.
00:03:10.159 --> 00:03:13.360
So I would say that I'm an artist working with technology.
00:03:13.599 --> 00:03:14.719
Yes, because you studied that.
00:03:14.800 --> 00:03:16.000
Did you study technology?
00:03:16.080 --> 00:03:16.800
Am I right in saying?
00:03:17.120 --> 00:03:19.360
Yeah, it's technically I studied computer music.
00:03:19.439 --> 00:03:28.479
So um there's this really special program at Stanford where I did my PhD, which is um it's a very technical program, but it's also housed within the music department.
00:03:28.639 --> 00:03:37.680
So I really feel like I'm a kind of translator between um technical fields and cultural fields, and that's where I feel the most comfortable.
00:03:38.000 --> 00:03:41.360
And maybe that is also more akin to the future.
00:03:41.520 --> 00:03:44.560
We might, we might get onto that towards towards the end of the conversation.
00:03:44.719 --> 00:03:50.479
But I first um discovered you through Holly Plus, and it wasn't really your TED talk or anything.
00:03:50.560 --> 00:04:04.479
It was just, I think it was a few years ago when I was researching the book, and I couldn't find that many people who were exploring identity, but there were um a couple of people in the sort of art techno music field.
00:04:04.719 --> 00:04:08.800
How did you come about to conceive of Holly Plus?
00:04:09.199 --> 00:04:16.800
Well, as I mentioned, I've been working with AI for about 10 years, and this is really before it became really productized.
00:04:16.879 --> 00:04:22.240
So when most people think of AI today, they think of ChatGPT or Dali or Midjourney.
00:04:22.399 --> 00:04:27.360
These are really large general models that are trained by scraping data from the internet.
00:04:27.839 --> 00:04:32.480
But back in the day, none of those products were available, so you really had to train your own models.
00:04:32.639 --> 00:04:36.000
And one of the first questions would be: what do you want to train the model on?
00:04:36.160 --> 00:04:37.199
What's the data set?
00:04:37.439 --> 00:04:46.399
And you know, with a background in composition and understanding the music history of kind of sampling, I didn't feel comfortable using anyone else's data.
00:04:46.480 --> 00:04:53.199
So I just wanted to use my own data, put together an ensemble where we recorded training data specifically for training a model.
00:04:53.360 --> 00:04:59.519
And when I started working with specific voice models, I felt the most comfortable training on my own voice.
00:04:59.680 --> 00:05:10.240
So a voice model is essentially an AI model that learns all of the sounds of your particular voice that can then emulate and sing in your natural singing voice.
00:05:10.399 --> 00:05:13.279
And so I felt weird asking someone else to do that.
00:05:13.519 --> 00:05:16.560
It made the most sense to kind of experiment on my own voice.
00:05:16.720 --> 00:05:19.519
And I am a vocalist, you know, I've released records where I'm singing.
00:05:19.680 --> 00:05:21.920
I never really thought of myself as purely a vocalist.
00:05:22.000 --> 00:05:26.000
I feel like there are many, many people with much more beautiful voice than I have.
00:05:26.160 --> 00:05:32.399
But this felt like the kind of, I don't know, the ethical approach to use my own IP in that in that instance.
00:05:32.720 --> 00:05:33.040
Yes.
00:05:33.279 --> 00:05:47.920
And and how important is the emerging sort of field of decentralized technologies, whether it's AI or blockchain or the new era, if you like, of less sort of uh linear hierarchical technologies and the new sort of networked technologies?
00:05:48.160 --> 00:05:50.560
Has that made much more of your work possible?
00:05:50.879 --> 00:05:55.199
Yeah, I I think this has been a really fundamental part of what I've been working on.
00:05:55.360 --> 00:06:06.000
I mean, one way that we like to think about it, so I released an album in um 2015 called Platform, and then an album in 2019 called Proto, short for protocols.
00:06:06.079 --> 00:06:20.879
And I see this kind of shift from the platform era to the protocol era, the platform being the, you know, you think of like Spotify or Instagram or these specific products that are, that are platforms that kind of determine how artistic expression is distributed.
00:06:20.959 --> 00:06:29.600
And so different decisions that the platform owners will make have a direct impact on how media is made to optimize for that platform.
00:06:29.839 --> 00:06:35.439
I like to think about us entering this kind of protocol era where we're working at a layer underneath.
00:06:35.519 --> 00:06:51.439
And if you think about protocols being, you know, something like Ethereum would be a protocol that you can actually build applications on top of, but you can also think of a protocol as like just a way of behaving, like certain kind of like social manners that we have are also just kind of like social protocols.
00:06:51.519 --> 00:06:54.240
It's a kind of rule set or an understanding of how things work.
00:06:54.399 --> 00:07:05.040
So I like to think about the creation and distribution and all of the kind of facets that go into the art making process as part of the protocol, as part of the artwork itself.
00:07:05.279 --> 00:07:13.040
And yeah, I'm hoping that we're entering an era where more artists are able to have more control over all of those facets of their output.
00:07:13.360 --> 00:07:15.839
Yeah, do you think many do at the moment?
00:07:16.000 --> 00:07:23.199
I mean, there are some people working in this space, but not many who as are as high profile and prolific as you.
00:07:23.600 --> 00:07:25.759
Do you think that's going to change over time now?
00:07:26.319 --> 00:07:31.680
Um, yeah, I mean, I think a lot of the work that I've done requires really boring research.
00:07:32.560 --> 00:07:35.120
And I think a lot of people, you know, fair enough.
00:07:35.680 --> 00:07:42.879
People are busy and they don't want to have to read uh machine learning papers and watch, you know, academic conferences on YouTube.
00:07:42.959 --> 00:07:45.680
That's like a pretty, pretty specific thing.
00:07:45.839 --> 00:07:52.240
But um yeah, I mean, I think there, yeah, there's a technical barrier for entry for a lot of these things.
00:07:52.480 --> 00:08:02.480
But I think as, you know, I feel like we learned so much from the internet becoming a really ubiquitous, all-present part of our daily lives.
00:08:02.560 --> 00:08:14.000
And I think people started to kind of wake up as to how technical decisions or platform decisions, design decisions were impacting their personal lives really directly.
00:08:14.240 --> 00:08:18.480
I think people have a desire to have more agencies within these systems.
00:08:18.720 --> 00:08:26.079
So yeah, I do think people are are becoming more and more interested in understanding the tech and understanding how they can have agency with it.
00:08:26.399 --> 00:08:31.759
That's interesting because some of your work in the past has focused on the theme of surveillance, hasn't it?
00:08:31.920 --> 00:08:35.759
You know, how we are being like monitored or surveilled, privacy public.
00:08:36.159 --> 00:08:48.639
What does it mean, do you think, to be machine readable, to be noticed, to be captured, monitored, analyzed, and maybe even redefined or represented or represented in this kind of emerging tech world?
00:08:48.879 --> 00:08:49.919
Yeah, that's a great question.
00:08:50.080 --> 00:08:54.320
I actually have a book coming out in the next couple weeks called All Media is Training Data.
00:08:54.399 --> 00:08:59.759
And that that doesn't some people read that and they they they think I'm saying all it is is training data.
00:08:59.840 --> 00:09:00.960
And that's not what I'm saying.
00:09:01.039 --> 00:09:18.960
I'm saying that, yeah, as soon as something is captured in media, like, you know, this conversation that we're having, or as soon as you take a photograph or a video of anything, then it becomes machine legible, and then it can become part of a training canon, you know, to train a model that you may never even interact with.
00:09:19.120 --> 00:09:28.240
So a lot of the data that we're placing in social media, online, you know, this is already going to train different models that we're just not seeing or necessarily interacting with.
00:09:28.399 --> 00:09:33.679
I think it's a pretty profound moment in our kind of on our human timeline.
00:09:34.320 --> 00:09:36.320
Yeah, I would completely agree.
00:09:36.480 --> 00:10:05.120
I mean, I think people are kind of used to like, oh, maybe something's extracted from a PDF about me or a Wikipedia page, but all this kind of digital phenotyping, treating us almost like patients, capturing some of the atmospheric or environmental data around us, the way we move, the way I was talking to somebody about, you know, motion prints in VR, for example, the way in which our gesture with our head and our hands, well, you know, it's so individual that it's like a fingerprint and they call it motion print.
00:10:05.360 --> 00:10:07.519
I feel like sometimes there's nowhere to escape.
00:10:07.679 --> 00:10:16.879
Maybe that's the dystopian view of it, and obviously you're looking at it primarily through the opportunities, but it is a very, as you say, very profound shift, isn't it?
00:10:17.200 --> 00:10:24.639
Yeah, I think I think it all depends on kind of where we end up in terms of, you know, who's using this data and for what purpose.
00:10:24.879 --> 00:10:33.679
You know, on the kind of dystopian side, okay, you just have giant corporations that are just, you know, scraping up all of our data and they're monetizing it in ways that we are don't participate in.
00:10:33.759 --> 00:10:35.600
And that's that's the kind of like dark image.
00:10:35.679 --> 00:10:52.960
But there's also a, I think, a beautiful way to approach AI that requires a lot of um work being put in the public domain, having public domain models that we can then fine-tune and build on top of, participating and contributing to something that feels like it's collective, like a collective accomplishment.
00:10:53.120 --> 00:11:02.879
I like to think of AI as us in aggregate, as a collective human accomplishment rather than a kind of like alien other that's here to just uh replace us.
00:11:03.120 --> 00:11:16.480
So I think if we can think about it as a kind of really sophisticated coordination and collaboration mechanism, and we can find ways to to make it work for everyone and be fair, then I think it maybe will be less scary for people.
00:11:16.799 --> 00:11:22.320
I like that idea of complicity, that's the balance between the human and the machine elements.
00:11:22.480 --> 00:11:33.679
I think you've done that really well with Holly Plus, because obviously you've invited people to sort of be the um, I don't know, the stewards of your voice in the way that they deliver it, haven't you?
00:11:33.840 --> 00:11:39.039
I mean, and I think that's uh I don't know what was what was the reception to that when you suggested that?
00:11:39.120 --> 00:11:56.399
Because I I saw a lot of people online saying, oh, but what does this mean for sort of rights, copyright, royalties, you know, and and actually I I think, and correct me if you're wrong, but you were saying you can use my voice as long as you credit it or give me attribution, but what you do with it is up to you artistically, maybe to a point.
00:11:56.720 --> 00:11:57.200
Exactly.
00:11:57.279 --> 00:11:58.480
And that's what works for me.
00:11:58.639 --> 00:12:00.080
That's not going to work for everyone.
00:12:00.159 --> 00:12:07.279
And that kind of goes back to the conversation we were having earlier about the platform dynamics and you know, this kind of one size fits all for artists.
00:12:07.360 --> 00:12:16.799
I hope that we move to a situation where people are able to really make specific decisions about their own practice, the economics that make sense for them and their communities.
00:12:17.039 --> 00:12:20.879
For me, I was fine with just taking a permissive approach to my IP.
00:12:21.039 --> 00:12:23.440
I was more curious to see how people would use it.
00:12:23.600 --> 00:12:27.440
But I could imagine another vocalist might really not want someone to use their voice.
00:12:27.519 --> 00:12:28.799
And that's totally understandable.
00:12:28.879 --> 00:12:30.799
And that that right should be there.
00:12:30.960 --> 00:12:33.360
At the moment, it's a really gray area.
00:12:33.519 --> 00:12:37.759
You have personality rights as something that has more legal history.
00:12:37.919 --> 00:12:49.679
So if somebody recorded something and then used my name as an artist name, you could sue on those grounds, but it's less difficult to sue on the on the training of a model using my my voice.
00:12:49.919 --> 00:12:55.200
So we'll see where all of this lands in terms of legal, but yeah, that's just what felt right to me.
00:12:55.360 --> 00:12:58.480
I wanted to just experiment and make it as open as possible.
00:12:58.799 --> 00:13:08.000
Did anybody do anything with your voice that you weren't happy with or maybe surprised you or was innovative in a way that you hadn't really expected?
00:13:08.320 --> 00:13:10.960
You know, no one did anything that I was offended by.
00:13:11.120 --> 00:13:15.840
You know, also the voice that we made public was a I I have multiple voice models.
00:13:15.919 --> 00:13:21.279
One of them is more kind of language specific, another one is a real-time instrument that people can sing through.
00:13:21.440 --> 00:13:27.279
And then the one that we made public is a is a more kind of it it deals in more kind of like ambience and texture.
00:13:27.360 --> 00:13:34.879
So it can take a polyphonic input, any kind of sample you can drag onto the website and it'll transfer it into a kind of choir of hollies.
00:13:34.960 --> 00:13:41.279
And because it's more abstract, I wasn't really dealing with having to edit out hate speech or anything like that.
00:13:41.440 --> 00:13:49.600
Of course, that's always a always a question when you're dealing with the the open internet, but really I was surprised at the kind of beautiful approaches that people took to it.
00:13:49.679 --> 00:14:07.360
I mean, one person recorded their train ride and the kind of screeching of the train of the train tracks to the metal, and then transferred that into my voice, and it made this gorgeous choir with the most amazing overtones and just things like that that I wouldn't have thought of myself were just a really nice surprise.
00:14:07.600 --> 00:14:15.679
You see, I think this is one of the really interesting dimensions of AI, that it gives you another perspective once you apply these tools or use these tools.
00:14:15.840 --> 00:14:35.039
And I think Eric Schmidt, of all people, said this about AI agents that I don't know if you agree, but that they will give us another perspective to blend with the perspectives that are delivered through our own human senses and give us a picture or an understanding or even a window into the world that is richer or we might not have had access to before.
00:14:35.440 --> 00:14:48.559
Yeah, I mean, AI models essentially are just able to consume and crunch and make sense of a much broader or a vaster, if I can say that, um, amount of data than the human brain can.
00:14:48.799 --> 00:14:57.919
And so I rather than this kind of replacement thing, I see it as an aid to help us understand the vast amount of amounts of data that we're able to collect these days.
00:14:58.080 --> 00:14:59.679
Uh so yeah, I I think that's true.
00:14:59.759 --> 00:15:04.159
It's also just kind of like a strange mirror to ourselves, a strange reflection.
00:15:04.320 --> 00:15:07.120
It's not exactly us, it's a kind of mutant form of us.
00:15:07.200 --> 00:15:11.519
And so by interacting with that mirror world, we understand ourselves a little bit more.
00:15:12.080 --> 00:15:12.879
It's weird, isn't it?
00:15:12.960 --> 00:15:16.080
Because it's got an intimacy, but also an alien nature to it.
00:15:16.240 --> 00:15:19.759
It is that weird, maybe that's the uncanny thing about it.
00:15:20.639 --> 00:15:25.120
Yeah, I think the I think the intimacy is that it is actually human created.
00:15:25.200 --> 00:15:37.120
It's all of the data is from humans, and so it's basically you're just interacting, you're collaborating with vast amounts of people, and our brains haven't really evolved to be able to conceive of of that.
00:15:37.200 --> 00:15:48.399
You know, we come from tribal groups where we're used to having only so many kind of um social connections, and now all of a sudden we're able to collaborate with millions and millions of people at once.
00:15:48.480 --> 00:15:51.360
And that that I think that that scrambles our brains a little bit.
00:15:51.679 --> 00:15:56.159
Which is one of the reasons I love the language that you use for these things like spawning.
00:15:56.480 --> 00:15:58.879
I love because it's biological, isn't it?
00:15:59.200 --> 00:16:00.480
Not technological.
00:16:02.000 --> 00:16:08.480
Yeah, so spawn uh it started out as spawn when I when I made proto and we were working with our ensemble in Berlin.
00:16:08.559 --> 00:16:11.840
We wanted to have a an ensemble member trained on our voices.
00:16:12.080 --> 00:16:17.200
And so we um we came up with the name Spawn and called her our AI baby.
00:16:17.600 --> 00:16:21.120
Um, and spawn has since kind of grown up into a verb.
00:16:21.200 --> 00:16:25.440
This word spawning is uh is what we like to use for some of these generative AI tools.
00:16:25.600 --> 00:16:35.440
So instead of sampling, which is a kind of one-to-one mechanical reproduction, we think about spawning as being able to create infinite media from a kind of seed source.
00:16:35.519 --> 00:16:40.480
And so spawning felt like a much more kind of descriptive version of what's happening there.
00:16:40.639 --> 00:16:57.039
And then spawning has evolved once again and is now the name of an organization that I co-founded with my partner Matt Dryhurst and um Jordan Meyer, who's a machine learning engineer, where we're working to uh advocate for um data manners in the machine learning space.
00:16:57.200 --> 00:17:02.320
So spawn, spawning, spawning has had quite an evolution over the last several years.
00:17:02.559 --> 00:17:03.840
I love it, yeah.
00:17:04.079 --> 00:17:05.119
So so is that right?
00:17:05.200 --> 00:17:07.920
So it allows artists to opt in and out of data sets.
00:17:08.160 --> 00:17:08.720
How does that work?
00:17:08.880 --> 00:17:10.960
How does an artist access that?
00:17:11.279 --> 00:17:11.519
Sure.
00:17:11.680 --> 00:17:15.119
So at the moment it's in the image space and uh in the text space.
00:17:15.279 --> 00:17:19.920
And so you're able to go in and at the moment we don't have opt-in, but we do have opt-out.
00:17:20.160 --> 00:17:28.400
So if you don't want for your images on your website, for example, to be included in machine learning training, you can add that to your website using our tools.
00:17:28.559 --> 00:17:34.960
You can see if your data is already being used in training sets by going to have ibentrained.com.
00:17:35.200 --> 00:17:36.640
Yeah, just check out spawning.
00:17:36.720 --> 00:17:38.559
We we have several different things in the works.
00:17:38.640 --> 00:17:41.920
We're also working on uh the public domain at the moment.
00:17:42.079 --> 00:17:45.279
We see that as a really important first step for an opt-in.
00:17:45.839 --> 00:17:56.400
So if you have a public domain base layer model, then you can fine-tune that model on your own work and feel like you as an artist have actual ownership over that model.
00:17:56.559 --> 00:18:06.880
So you can then sell the model or you can license the model or you can come up with whatever economic interaction that you want around that model, knowing that you haven't infringed on anyone else's copyright.
00:18:07.039 --> 00:18:11.359
We see that as a really important first step to getting a lot of artists into the space.
00:18:11.680 --> 00:18:12.720
That's amazing.
00:18:12.960 --> 00:18:18.000
So have you got a big tech team then, or is it just you and Matt and a few collaborators doing this?
00:18:18.240 --> 00:18:20.079
It is definitely not just me and Matt.
00:18:20.240 --> 00:18:22.960
We have a whole team of wonderful people.
00:18:23.119 --> 00:18:26.480
You know, we've always been collaborative oriented.
00:18:26.640 --> 00:18:27.279
Can I say that?
00:18:27.440 --> 00:18:28.880
We've always been collaborators.
00:18:28.960 --> 00:18:32.640
Um, and I've I've always really enjoyed working with teams of engineers.
00:18:32.799 --> 00:18:34.799
So spawning is its own kind of thing.
00:18:34.880 --> 00:18:39.039
It's run in the United States, and there's a whole team of people working on that full-time.
00:18:39.200 --> 00:18:45.440
The work that we did at The Serpentine, there are three different novel models that we're working with for that show.
00:18:45.599 --> 00:18:49.599
And we're working with EarCAM, which is a research institute in Paris.
00:18:49.759 --> 00:18:58.319
Um, we're working with Algamous, which is a research team in Lille, Stable Audio, which is a decentralized kind of audio research organization.
00:18:58.400 --> 00:19:04.640
So we have collaborators and engineers that we're in constant contact with that really make everything possible.
00:19:04.960 --> 00:19:05.920
That's amazing.
00:19:06.079 --> 00:19:07.680
That really is a network, isn't it?
00:19:07.839 --> 00:19:09.920
A kind of vibrant, dynamic network.
00:19:10.079 --> 00:19:11.920
Let's talk about your work at the Serpentine.
00:19:12.000 --> 00:19:13.680
I think it's on till February, isn't it?
00:19:13.759 --> 00:19:14.079
Mm-hmm.
00:19:14.319 --> 00:19:18.240
Because I really want I yes, I'm I'm hoping to come and see it before it it finishes in February.
00:19:18.319 --> 00:19:19.359
I know it's live now.
00:19:19.599 --> 00:19:21.119
Could you tell us a little bit about it?
00:19:21.200 --> 00:19:23.759
Because it's quite the construction.
00:19:24.160 --> 00:19:24.559
It is.
00:19:24.640 --> 00:19:26.880
It was a uh yeah, it's quite the project.
00:19:27.119 --> 00:19:28.000
Beautiful as well.
00:19:28.240 --> 00:19:28.640
Thank you.
00:19:28.720 --> 00:19:30.319
It took over a year to put together.
00:19:30.480 --> 00:19:33.359
You know, we worked really closely with the arts and technology team there.
00:19:33.519 --> 00:19:41.359
Hans Ulrich is so amazing at understanding what's happening and staying on top of all of the kind of cutting-edge developments.
00:19:41.440 --> 00:19:46.319
And so he really trusted us with the vision to work with the arts and technology team there.
00:19:46.559 --> 00:19:53.839
And what we wanted to do is so many shows really focus on just the output of a model because it's so impressive.
00:19:54.160 --> 00:20:01.519
What we wanted to do was focus on the process of the making of a model because we see making models as an artwork in and of themselves.
00:20:01.839 --> 00:20:05.359
Every step in the process, there's a moment for artistic intervention.
00:20:05.519 --> 00:20:17.519
So the creation of really deliberate training data as artworks, that going into, you know, training with uh choirs across the UK, those are performances, those are artworks.
00:20:17.759 --> 00:20:25.920
The decisions that we make in categorizing the data for the model, the model itself, and in interacting with the model as an output.
00:20:26.160 --> 00:20:31.599
Each of those phases we really wanted to show and give the visitor an experience of in the space.
00:20:31.839 --> 00:20:35.359
So we put together this really wild um production.
00:20:35.519 --> 00:20:41.920
Um, we wanted to make a choral AI model that could be collectively owned by a group of choirs.
00:20:42.000 --> 00:20:45.200
So we recorded 15 choirs across the UK.
00:20:45.519 --> 00:20:49.279
We hopped in a car and we drove around and we recorded all of these choirs.
00:20:49.519 --> 00:20:57.599
And so when you go into the exhibition in one of the rooms, you can hear the kind of field recording and the ethnographic uh recordings that we were making on site.
00:20:57.759 --> 00:21:02.880
So you can hear how the training of a model, it's not just these kind of random data points.
00:21:02.960 --> 00:21:09.119
It's actually people being creative and making meaning and making work together in a space.
00:21:09.279 --> 00:21:10.799
That's what these data points are.
00:21:10.880 --> 00:21:13.599
They're not these cold little bits of media.
00:21:13.839 --> 00:21:15.440
Um, so you can experience that.
00:21:15.519 --> 00:21:19.599
And then we created an organ out of uh GPU fans.
00:21:19.680 --> 00:21:27.279
So the sound of our studio is just the constant whirring of GPU uh fans that's cooling our GPU as it's crunching all the numbers.
00:21:27.359 --> 00:21:34.319
So you kind of can experience the compression phase and the power that goes into to training these models through the organ.
00:21:34.559 --> 00:21:42.240
And you can hear the output of the model that we created with all of the recordings across the UK and our own personal archive.
00:21:42.319 --> 00:21:49.279
We made a model that we can prompt and create fully coherent, beautiful choral songs, infinite choral songs.
00:21:49.359 --> 00:21:51.839
You can hear those spatialized in the space.
00:21:52.000 --> 00:21:54.559
And then also the visitor can interact with the model.
00:21:54.720 --> 00:21:59.759
That's a really important piece of this that these models they're designed to be interacted with.
00:22:00.079 --> 00:22:00.240
With.
00:22:00.319 --> 00:22:09.279
So we wanted to allow the public to see what that feels like without introducing a screen because there's so much screen work when it comes to people interacting with AI.
00:22:09.440 --> 00:22:19.839
So we wanted to eliminate the screen and just use microphones so people can sing in through a microphone and they can kind of navigate through the choral model and hear the choral model singing back to them.
00:22:20.000 --> 00:22:33.279
So we hope that the visitor gets the experience of the kind of full stack training of a model and maybe can see themselves in that and how it might be beautiful to collaboratively train a model, you know, with other people.
00:22:33.920 --> 00:22:35.119
It really is amazing.
00:22:35.279 --> 00:22:42.240
Um is that idea of the process being creative and also collaborative and transparent to people?
00:22:42.640 --> 00:22:47.519
A bit like your idea about protocols being habits and ethics.
00:22:48.160 --> 00:22:48.640
It is.
00:22:49.039 --> 00:22:49.759
A hundred percent.
00:22:49.839 --> 00:22:50.079
Yeah.
00:22:50.160 --> 00:22:54.319
I mean, I feel like the design of the exhibition, we were thinking very closely.
00:22:54.480 --> 00:22:55.839
What what is the protocol?
00:22:56.000 --> 00:22:56.799
What's the approach?
00:22:56.960 --> 00:23:04.720
What's the kind of rules that we want to put together for how this functions in a space, for how we deal with the individual participants?
00:23:04.880 --> 00:23:09.039
That was a really important part of the kind of protocol design of the show.
00:23:09.279 --> 00:23:15.359
And one piece of it, it might sound really bureaucratic and kind of boring, but we worked really closely with the Serpentine.
00:23:15.440 --> 00:23:26.319
Their team is really at the forefront of thinking through, from a legal perspective, how you can co-own and um collectively kind of shepherd a data set.
00:23:26.480 --> 00:23:38.240
So they they put a lot of energy into legal work and legal frameworks to think through how can we invite the people who participated in these recordings to play a part in stewarding this data moving forward.
00:23:38.480 --> 00:23:39.279
So yeah.
00:23:39.680 --> 00:23:40.720
What did they think of it?
00:23:40.799 --> 00:23:42.160
What were the participants?
00:23:42.480 --> 00:23:46.480
You know, it was such a wonderful, wide range of people.
00:23:46.640 --> 00:23:54.960
A lot of people were really surprised at the variety of ages that were involved because this was a I think a lot of a lot of projects would just record in London.
00:23:55.039 --> 00:24:15.039
And we did do some recordings in London, but we really hit the road and went across the UK and we went to rural communities and we met with older people and people who had never heard of AI, and we tried to get a really kind of broad cross-section of experience, and people had really interesting thoughts and concerns.
00:24:15.279 --> 00:24:21.200
And in a way, it was a it was informative to us to understand, you know, kind of where people are at.
00:24:21.279 --> 00:24:26.799
But I think we also were helping people think through their own data footprint online.
00:24:27.200 --> 00:24:39.920
Like some people, for example, would come to the project and be a little bit wary of training in AI, but that same person might upload their music to YouTube, not understanding that that is already training in AI, just without their knowledge.
00:24:40.079 --> 00:24:45.839
So there was a lot of information that we were able to share with people to better understand how this landscape is unfolding.
00:24:46.160 --> 00:24:47.119
It's interesting, isn't it?
00:24:47.200 --> 00:24:57.440
Because I think it's been really difficult to get people to engage with the idea or the notion that identity is transformed by the media that you use.
00:24:57.599 --> 00:25:04.480
And I guess that's a great example of that actually happening, although you might not necessarily use the language identity or personal identity.
00:25:04.640 --> 00:25:08.079
Maybe people start to realise that is happening to them.
00:25:08.160 --> 00:25:13.279
Like elements of the self are extended or augmented or whatever.
00:25:13.359 --> 00:25:15.440
I think I think that's happened with avatars as well.
00:25:15.599 --> 00:25:18.400
I think now people are messing about and playing with avatars.
00:25:18.480 --> 00:25:20.640
People can start to understand, oh, hang on a minute.
00:25:20.799 --> 00:25:28.559
This can happen to my voice, or my voice can, you know, suddenly I can speak Japanese uh in my own accent, even though I can't, sort of thing.
00:25:28.799 --> 00:25:32.559
Do you think it's going to get to a point where there is a better understanding of all this?
00:25:32.880 --> 00:25:40.480
Yeah, I mean, it's that's something that I'm that I think about a lot, that I'm that I'm actively working to help people understand.
00:25:40.880 --> 00:25:49.599
One of the things that I think about is the kind of difference between the data point and the actual person, like the fleshy, messy person that's in the real world.
00:25:49.759 --> 00:25:54.400
So one of the projects that I worked on this past year was called X Harry Mutant X.
00:25:55.440 --> 00:26:00.079
And so that's a project that we um exhibited at the Whitney Biennial this year.
00:26:00.319 --> 00:26:13.759
So this is a really funny kind of super superhero character that I created for myself because I, my, my name, Holly Herman, me as an artist, I meet the threshold of being a concept in some of these large models.
00:26:13.839 --> 00:26:19.200
So if you go to an image model and you type in my name, a kind of essence of me will appear.
00:26:19.519 --> 00:26:29.039
And because these models are trained by just scraping the open internet, it's just collected a bunch of images of me and created a kind of pastiche of what it understands of me.
00:26:29.119 --> 00:26:37.119
So because I'm not as famous as someone like Dolly Parton, Dolly Parton would have all of her full features and would be much more fleshed out.
00:26:37.279 --> 00:26:40.240
Me, I just kind of appear as an orange blob.
00:26:40.400 --> 00:26:47.519
So, you know, where wherever my face might always be different, but I'll always have this giant orange blob of hair.
00:26:47.599 --> 00:26:54.400
So I was really asking myself, okay, how much agency do I have in this, in determining who I am in these systems?
00:26:54.559 --> 00:27:00.400
If this is gonna be the net, if this is gonna be the next layer of the internet, do I get to decide who I am here?
00:27:00.559 --> 00:27:09.119
And so when I did some experiments and tried introducing some images of me that were outside of that with entirely different hair, it was kind of thrown out as noise.
00:27:09.279 --> 00:27:16.079
So I realized if I wanted to have any agency here, I had to almost lean into my cliche, lean into my pastiche.
00:27:16.480 --> 00:27:24.240
So we designed this wild costume X Harry Mutant X, which is this kind of golem superhero with giant orange hair.
00:27:24.480 --> 00:27:32.960
And then I created a model with my partner Matt Dryhurst of that character that people can actually go to the Whitney website and prompt through.
00:27:33.119 --> 00:27:39.599
And no matter what you prompt, you'll always come up with something kind of X Harry Mutant X like in the image.
00:27:39.680 --> 00:27:42.400
It's kind of infected with my giant orange hair.
00:27:42.559 --> 00:27:45.440
And each of these images are tagged with my name.
00:27:45.680 --> 00:28:01.359
So all of these images then are online, and the next time a large image model scrapes the internet and tries to understand me as a concept, it will encounter all of these giant X Harry Mutant X images of me, and that will transform who I am in the model.
00:28:01.680 --> 00:28:18.400
So it's a kind of long-form performance between me and the public and machine learning companies, and it's a really fun way to kind of explore identity and who I am and and the kind of difference between the person and the data points and the digital footprint that we leave on the internet.
00:28:19.039 --> 00:28:23.119
Oh, I love that because this is the area I'm really interested in.
00:28:23.200 --> 00:28:31.039
And it reminds me a little bit of database animals, Azuma about the Otaku and the Otaku's obsession with anime and mango and obviously.
00:28:31.200 --> 00:28:43.359
And I'm I really see, let me know if you agree or not, but this sort of breakdown of us as people into this database of kare mo, mo elements, like the attributes, the elements that make us up.
00:28:43.440 --> 00:28:51.119
And I mean, you just said, you know, orange hair, for example, and they're very much like these characters having loved elements, care mo.
00:28:51.519 --> 00:28:58.640
And how do these elements get mixed, remixed, arranged, and rearranged together, and then spun out in different contexts?
00:28:58.880 --> 00:29:18.240
And the idea is that in this fan world, there is no original, there are no copies, like the parodies are just as important as the original character, but these elements are kind of constructed and reconstructed and rearranged in lots of different ways, and that's that's maybe how we should think about ourselves in the digital environment.
00:29:18.480 --> 00:29:26.640
I mean, I know it's a kind of you know atomistic sort of way in these individual attributes, but it is that a way forward for us to think about the self?
00:29:26.960 --> 00:29:29.359
Yeah, I think that's a really interesting reference.
00:29:29.519 --> 00:29:36.640
I think it kind of intersects with this idea of fandoms and different characters that allow themselves to be kind of remixed and reinterpreted.
00:29:36.799 --> 00:29:43.359
Machine learning models are designed to be interacted with, they're designed to be prompted.
00:29:43.599 --> 00:29:55.440
And so this kind of idea of authorship just being a one-way stream or just being one person, you know, in order to make the model, you're usually drawing from many, many people's creative input in the first place.
00:29:55.519 --> 00:30:04.480
And then when you're inviting people to interact with the model, then you have this kind of like fandom moment where people, the new work is it's both you and it's the person who prompted.
00:30:04.559 --> 00:30:06.880
And then it's this interesting blending.
00:30:06.960 --> 00:30:13.519
And I think, yeah, I think that we can learn a lot from how fandoms have proliferated in the last several years.
00:30:13.759 --> 00:30:15.119
Yeah, no, totally.
00:30:15.279 --> 00:30:21.759
You mentioned collaboration, and I know obviously you've you've spoken a few times about the the people and groups that you work with.
00:30:21.920 --> 00:30:23.920
What about your podcast, Interdependent?
00:30:24.240 --> 00:30:24.880
Yes, yes.
00:30:25.039 --> 00:30:25.599
Oh gosh.
00:30:25.759 --> 00:30:26.480
Interdependence.
00:30:26.559 --> 00:30:29.279
I miss interdependence, but we just kind of ran out of time.
00:30:29.359 --> 00:30:31.279
It was such a big part of our COVID time.
00:30:31.519 --> 00:30:32.720
Will you carry on doing it?
00:30:32.799 --> 00:30:34.720
Because you had some fascinating guests.
00:30:35.119 --> 00:30:36.079
I really enjoyed it.
00:30:36.160 --> 00:30:40.160
I would love to be able to, yeah, maybe Holly Plus can take it on the Sunday.
00:30:40.720 --> 00:30:42.319
That's the point of our avatars, isn't it?
00:30:42.559 --> 00:30:45.519
Go off and do things or agents that we haven't got time to do.
00:30:45.839 --> 00:30:46.319
Exactly.
00:30:46.480 --> 00:30:48.559
You know, I won't, I I'll never say never.
00:30:48.640 --> 00:30:53.920
At the moment, we we don't have time to to start it again, but it's something that I I very much miss.
00:30:54.079 --> 00:30:59.200
It was one of those things during COVID where we were so used to being touring musicians and traveling around so much.
00:30:59.279 --> 00:31:01.759
We missed, you know, meeting new people all the time.
00:31:01.839 --> 00:31:06.799
And it was one of those like socially acceptable ways to call someone that you find interesting and just talk for a while.
00:31:07.759 --> 00:31:09.359
That's why I that's why I got in touch with you.
00:31:09.599 --> 00:31:10.240
Yeah, exactly.
00:31:10.319 --> 00:31:10.559
Yeah.
00:31:10.640 --> 00:31:14.160
So it's honestly it's a great, it's a great, it was a great social outlet for us.
00:31:14.240 --> 00:31:15.440
And I definitely miss it.
00:31:15.680 --> 00:31:22.799
But is there anybody on there that you thought, or an idea or a concept that was just so out there, challenging or amazing and inspiring?
00:31:23.039 --> 00:31:24.880
Oh gosh, I don't know if I would single out.
00:31:25.119 --> 00:31:26.480
There were just so many good ones.
00:31:26.640 --> 00:31:31.440
Um, you know, one person that for some reason is popping in my mind right now is Jacob Horn from Zora.
00:31:31.839 --> 00:31:41.200
He's been thinking about these kind of decentralized identities and co-ownership ideas and and IP and new ways to approach that for a really long time.
00:31:41.279 --> 00:31:44.480
And I always he comes from a different, very different perspective than I do.
00:31:44.559 --> 00:31:47.759
And I always feel like it's a fruitful exchange with him.
00:31:48.079 --> 00:31:59.279
So, Holly, I know I've got to let you go in a minute, but what do you think the future of identity holds if we were to look sort of 20 or 30 years ahead or like mid-century?
00:31:59.519 --> 00:32:07.440
Where do you think we'll be in terms of understanding ourselves, representing ourselves, like expressing ourselves as artists like yourself?
00:32:08.079 --> 00:32:10.559
I think this is a really interesting philosophical question.
00:32:10.720 --> 00:32:22.000
I almost wish my partner, Matt Dryhurst, would be able to answer it because he's been thinking a lot about how the concept of identity and individuality has really shifted throughout the ages of humanity.
00:32:22.160 --> 00:32:25.920
There's this interesting book that he's been reading called The Creation of the Individual.
00:32:26.079 --> 00:32:46.640
Um, so I think one thing that might change is how we perceive ourselves as individual identities and might perceive ourselves more as kind of networks and relationships with other people and also just our own biology, how we might understand that depending on the the way that you look at someone, if like with a naked eye, it looks like a single person.
00:32:46.720 --> 00:32:55.279
But if you look through a microscope, there's all these different interconnected desires and microorganisms that are kind of interconnectedly working together.
00:32:55.440 --> 00:33:03.599
So, in a way, we as selves are also a kind of web of collaborators making our whole bodies work.
00:33:03.759 --> 00:33:06.640
I think that identity is going to get really weird.
00:33:06.880 --> 00:33:08.799
And I I very much welcome that.
00:33:09.039 --> 00:33:32.240
Yeah, it's exciting, but it's also threatening because if we have eight, nine billion people on the planet who are all in an identity crisis, because as you say, if it does get a bit weird and we can't understand the interior of ourselves or selves as um identity continues to be thought of in a more pluralistic way, let's say, I wonder what will happen to people, whether they'll be able to cope with it.
00:33:32.559 --> 00:33:51.200
Yeah, you know, I think not to be like super wide-eyed and uh overly optimistic, but I think that's one of the roles of art, honestly, to help people understand or to help people work through the feelings of of alienation in this like super technologically mediated world that we live in.
00:33:51.359 --> 00:33:54.480
And I hope that we can contribute to that in a meaningful way.
00:33:54.720 --> 00:33:56.400
Oh, I think you do wonderful work there.
00:33:56.480 --> 00:33:57.200
Um thank you.
00:33:57.279 --> 00:33:59.039
It's been fascinating to speak to you.
00:33:59.119 --> 00:34:04.319
And I'm sure it's it's only the tip of the iceberg, what's to come, from Holly and Co.
00:34:04.720 --> 00:34:05.279
So thank you.
00:34:05.440 --> 00:34:06.160
Thank you, Holly.
00:34:06.240 --> 00:34:07.359
It's been great to speak to you.
00:34:07.599 --> 00:34:08.880
Thanks so much, Tracy.
00:34:14.639 --> 00:34:18.000
Thank you for listening to The Future of You, hosted by me, Tracy Follows.
00:34:18.239 --> 00:34:22.079
Be sure to check out the show notes for more info about the topics we covered today.
00:34:22.239 --> 00:34:26.880
If you enjoyed this episode, please like and subscribe wherever you get your podcasts.
00:34:27.039 --> 00:34:30.639
And if you know someone who would love this episode, please share it with them.
00:34:30.880 --> 00:34:36.639
For more on the future of identity in a digital world, visit futuremade.group slash the future of you.
00:34:36.960 --> 00:34:41.760
And to explore the future of everything else, head over to futuremade.group.
00:34:41.840 --> 00:34:45.519
The Future of View podcast is produced by Big Tent Media.
