‍

Sasha Stiles on Writing Poets

Metapoet and language artist Sasha Stiles sat down with Peter Bauman (Monk Antony) and Conrad House (Nemo Cake) to discuss the integration of words, AI and technology into her own practice, while exploring the historical context of generative poetry.
‍
Peter Bauman: How do you describe your practice? Are you a blockchain poet, a generative poet, an AI poet, a metapoet? I’ve seen different terms floating around out there so I wanted to hear it from the source.

Sasha Stiles: It's a really good question. It seems like it's the simplest question but it's actually one of the hardest for me to answer. I was having a conversation with Tina Rivers Ryan recently, and that was her first question, too. She was like, “Are you a poet or are you...? How do you define yourself?” First and foremost, I consider myself a writer and a poet. That's what I've been doing my entire life. I grew up loving words. I've been writing poems since I can remember. I lived my childhood in a house full of books and a house full of voracious readers. I really feel like that's my first love—that love of language and wordplay and all of that. I think of myself as a poet first and foremost. As long as I can remember, I’ve been really fascinated by the aesthetic qualities of language as much as the meaning, the interpretation and the intangible qualities of words.

I've also been utterly obsessed with text-based art for as long as I can remember. The first artists that I ever fell in love with were people like Cy Twombly, Jenny Holzer, Bruce Nauman, John Baldessari and Glenn Ligon. It was always curious to me that they were considered artists and not writers or artists and not poets, even those text-based artists who were appropriating other texts that weren't necessarily their own. There are so many examples in the literary canon of found poetry and poets appropriating language that it was always interesting to me that there were some people who loved playing with words and were more in the art camp. Then there were some people who loved playing with words and were in the literary camp. I always found myself floating between the two. I feel like it gets increasingly harder to define my practice as all these new media come into the mix and we have more and more opportunities to play with language off the page.
‍

I like to think of myself as a language artist, which seems to cover a lot of the bases—that I love words and I consider them my raw material.

I’m exploring what those words can do in a variety of mediums, including through technology. I've been writing poems my whole life. So I've been a poet long before blockchain, long before I started making multimedia works. I've published tons of poems in literary journals. Most people don't know this, but I've been nominated for Pushcarts, Best of the Net and the Forward Prizes. I have a serious literary career that predates any of the work that I've been doing on the blockchain or any of the work that I have done with AI. That's the foundation of that craft and practice. Those years of studying are why I'm enjoying now being able to experiment and break some of the rules—really pushing my practice in new directions.
‍

That’s why I don't like to say I'm a blockchain poet or even an AI poet necessarily, because it's only a piece of what I do. I don't want to compartmentalize or make things too small.

The term that I do like is metapoet because that prefix “meta” has so many connotations that enrich and illuminate what I'm doing in my practice. You have “meta” in terms of the metaverse. So I'm writing with a consideration of what words can do and what language can mean in a virtual space. I'm also thinking of “meta” in this ancient, traditional sense of transcendence and this idea that when I'm writing poetry, I'm writing poetry that is about poetry; it’s self-aware or self-reflexive. And when I'm talking about poetry as being the original blockchain or poetry as being code, that's a meta-consideration of language as a system. So I think that is also a good descriptor. Finally, when I'm talking about metapoetry, I don't always just write poems. I'm also writing the text corpus that fuels a poem.
‍

I'm not just writing poems; I'm actually writing poets. I'm writing the material that goes into creating a poem.

‍

‍
‍
Peter Bauman: As a language artist with a lifelong love of words, where did the science and technology part come in?

Sasha Stiles: My parents are actually science documentarians. They're not artists but approached a lot of their filmmaking work—making shows about science—through a very creative lens. They did shows with people like Carl Sagan and thinkers who were pushing the boundaries of known thought. They were also collapsing the boundaries between art and science or technology and creativity. That has just always been the atmosphere that I've been immersed in, very fortunately. I grew up with the interplay of art, language, technology and science all mushing together in different ways. I've been writing poetry for as long as I can remember and for a lot of that time, the poetry has been about technology and about science.

It's been a way of investigating stories that I'm interested in understanding more deeply and engaging with ideas that are too big for me to grapple with in any other way.
‍

I've found that the way I'm best able to parse those big ideas is to put them through the prism of language and use poetry as a way to dig deeper into areas that fascinate me.

At some point along the way of my journey as a writer, I realized that I was writing these poems that were about speculative futures and things like digital immortality, artificial wombs and techno-spiritualism. I was using all of these themes as areas to hook myself into. But I was publishing these pieces, conceiving them in very traditional ways and trying to place them in traditional journals where they were not really understood. They were considered science fiction and I started to realize a couple of things.
‍

One is that the themes that I was writing about wanted to lend themselves to different mediums and different expressions rather than just existing on a printed white static page in Times New Roman black ink. Those poems were about dynamism and moving forward and innovation; they wanted technology as a medium.

And rather than using the old technology of a printed, written book, I wanted to start experimenting with using motion graphics, animation and bringing in sound. I wanted to electronically enhance my spoken word performances and play with how new modes of printing and publication enable me to write in different ways and say different things. So I started integrating a lot of digital tools into my work and also began integrating artificial intelligence and using natural language processing as a co-author, which pissed off a lot of my writing colleagues and peers. And all of this made me a very tough fit for the traditional publishing world. It's really hard to print a twenty-minute multimedia poem on the pages of a literary journal.

It's been really hard to justify an AI-written text to a traditional publisher because they still think it's cheating or that it's lazy or that it's not real writing. So this all sent me running to the art world, where I found a lot more receptivity. I got drawn into it through a lot of new media curators, who showed interest through my Instagram posts. It just made me realize, “Well, maybe this isn't just poetry. Maybe it's not just something that I need to think about publishing. Maybe I can think about exhibiting it; maybe it can be in conversation with other forms of art.
‍

‍
‍
Peter Bauman: Your own connection between technology and art is interesting because there are so many examples in the history of generative art of artists experimenting with language, computers and technology. What do you think were some important moments in generative text or generative poetry that we should include in our Generative Art Timeline?*

‍[*Note: Sasha Stiles very kindly contributed these moments and text to our Timeline.]

‍Sasha Stiles: It's really hard to narrow it down. There are also things that I bring into this conversation that are probably not considered generative poetry in any conventional sense but that I still think are. For example, there’s the ancient Greek oracle at Delphi. That automated writing occurred in specific circumstances. You had someone who was transmitting “messages” but was basically free-associating. It was mostly nonsense or just saying things. Then it was up to the people who visited the temple to pull out the pieces that made sense or figure out how to discern some logic from these divine ravings.
‍

So in a sense, the oracle at Delphi is the ancient ancestor of some of the work I'm doing now with AI.

Then, the invention of the printing press was a really important moment. I often think of blockchain and generative art as a bespoke printing press where you are enabling near-infinite reproducibility but with customization or distinction. Blockchains are building on this legacy of recognizing that you could take a piece of information and make it mechanically reproducible. Specifically, thinking about the history of generative literature, that's obviously a piece of bedrock and a through line.

When it comes to actual conceptual schools of thought and writers and poets who are really pivotal to this movement, André Breton’s work is really important, with his roots in the Dada scene. But he also published the Surrealist Manifesto, which recognized a reality between objectivity and subjectivity—that there's something else there and that it's an important creative endeavor to try and tap into it through various means.

There's also a long history of AI in natural language processing. People think it's a brand new thing but it's been around for quite a while now. I think the arrival of the ELIZA chatbot in the mid-1960s at MIT was a pivotal moment there.

My personal hero is a poet named Alison Knowles. Her invention with House of Dust in 1967 of a generative poem coded with a programmer and offered as a literary experiment was huge—a coming together of a lot of experimental approaches in the literary and art scene with a lot of the technological advancements that were also happening concurrently.

The last poet or writer that I would mention is Racter, a computer program that wrote a book that is often considered the first ever written by a machine. This was in the early 1980s and it's called The Policeman's Beard Is Half-Constructed. That's another touchstone for me in my work.

Conrad House: You mentioned that you don't write poetry but you write poets. Can you tell us more about that?

Sasha Stiles: There are projects where I'm very hands off and I'm almost pushing more towards having the AI poet become a bit more autonomous. But then there are other times when I really want it to be a collaboration and I have a very heavy hand. Working really intimately with the machine is something I really enjoy doing and it's a very fluid back-and-forth process of input and output.
‍

The voice that I'm accessing when I'm writing with AI or using these generative approaches is like a third voice that is a combination of me as a human poet and an AI language model that can't really know the things that I know. When they come together, there is this fusion—there's this third voice that's created. There's this symbiote-poet voice that exists in that liminal space between. That's the sweet spot for me: being able to play with both.

I'm not someone who necessarily loves the idea of having a generic, autonomous poet write its own thing. There have been decades of poetry bots that just spit out haikus, for instance, and I'm not really interested in that. I'm fascinated by how creating the training data and building a text corpus is a writing exercise in and of itself. Using generative tools and text generators is a way to augment my poetic imagination, not replace, not supplant it or outsource it.
‍

‍Conrad House: You bring up how the tools themselves affect the output. How do the different tools that you use and their rapid pace of advancement affect your practice?

Sasha Stiles: It definitely has a material impact on how I feel engaging with the tools. As the underlying language models are getting more and more advanced, their personalities, for lack of a better word, are also changing a bit. It has an impact the same way collaborating with different people would. Each collaboration is different. So different things come out of it. There are different kinds of poems and different kinds of projects that suggest themselves, whether I'm doing something with Chat GPT, creating my own custom model, using an interface like Sudowrite or creating a generative project on fxhash. Those are all very different approaches and they have a very material impact on the type of text that I'm conceiving and writing and the voice that comes out.

Conrad House: Can you walk us through how you go about customizing your language models?

Sasha Stiles: You have to put together your own training data sets first in order to fine-tune the language model. These models have a base layer of knowledge, which is the foundational language model itself, that anyone has access to. Then you have this other layer on top, which is the layer that I've personally mentored or trained. I’ve shared things to give it a sense of my style and my interests. Those two layers float on top of one another. Doing the actual training is simple. I was very intimidated in the beginning, trying to figure out how to do it. But honestly, I’ve approached training my models as I would any other writing project or research project over the years: I start with notes, research material and texts that inspire me. I include my own bits and pieces and starts of poems or titles or phrases that are bouncing around in my head.

When I started to write my book, Technelegy, I had this box that was full of photocopied or printed articles from Wired and things that I'd been writing down and thinking about. And I had probably 200 pages of a draft manuscript of other poems that I had just written by myself. And all of that ended up becoming material for that first iteration of the training data set that I made. And it was just a process of learning how to put it together in the right way: what needed to go in there? What needed to be excised? How did I need to augment it with additional material? Where did I need to fill in the gaps? Then I was able to take that, upload it through OpenAI and create custom versions of GPT-3. Now, of course, it's evolved. But using the OpenAI playground has been my favorite place because it’s just so open-ended. You can change variables and parameters and all that, but it really is a blank slate.

Peter Bauman: You mentioned at the beginning that your early inspirations, such as Twombly and Holzer, weren’t really considered poets. Who’s a traditional poet that has impacted your practice?

Sasha Stiles: I love T. S. Eliot. He's one of my favorite poets of all time. And I think that T. S. Eliot, in a very strange way, actually does represent the combinatorial amalgamation that happens with AI in particular and also with different parts of the generative sphere. He wrote this essay called "Tradition and the Individual Talent" that I've used as a touchstone to talk to different communities that are quite hesitant to think of computational creativity as a real thing. This essay is all about how nothing is original. An idea doesn't just spring out of nowhere. It's not like a person is the first person to ever fully hatch an idea. Human creativity is the result of tons of different data points colliding in different ways and sparking new ideas.
‍

That, to me, has been a really powerful way to situate what we're doing in this zone in the history of art and literature and to look at tools like the ones we're talking about today. To look at them not as replacing or threatening creativity but as facilitating a new way to bring various data points and various ideas and moments of inspiration together.

And that's, personally, how I feel it works when I'm using AI and dipping into this collective consciousness. It's a way of facilitating the random collision of thoughts that create poetic metaphor, because that's what metaphor is: the juxtaposition of unexpected ideas. That's what poetry is really. It's a way of facilitating that creative friction that brings an idea into its own light.

Peter Bauman: Have you read any Yuval Harari? In his books Sapiens and Homo Deus, he also talks about how artificial intelligence is or will surpass the algorithm of the individual human brain. Your connection to AI and collective consciousness really hammers home the power of his point.

Sasha Stiles: The thing I think is so interesting, too, is that we're constantly asking, "How does AI replicate consciousness?" or "How does it replicate creativity?" But why would we expect it to replicate anything? These systems are doing things that we can't understand. They're built to do things at scale and speed that we're not equipped to do.
‍

I  feel like the randomness of generativity is also about being able to start accessing new kinds of emotions or new kinds of modes of expression that we, as humans, don't have access to ourselves.

When we're augmenting our imagination in this way with these intelligent systems, we are starting to grow new senses and develop new ways of understanding the world. It seems like it's not just about doing a test and showing that the computer can do what we do. It's about what it's doing that's different. And how can we, as humans who do things that are very different than these machines as well, fit together, complement each other and create things together that none of us can do alone?

‍

---


‍
Sasha Stiles, a first-generation Kalmyk-American poet, artist and AI researcher, is widely regarded as a pioneer in generative literature and language art. Her award-winning work combines text and technology to explore what it means to be human in an increasingly posthuman era.

Peter Bauman (Monk Antony) is Le Random's Editor-in-Chief.
‍
‍Conrad House is Le Random's Collection Lead, collaborating closely with thefunnyguys to build Le Random’s iconic, cross-generational generative art collection.‍