“HOWL.camera: into the darkroom” is a conversation between data poet Ross Goodwin and Elisabeth Sweet about HOWL.camera as an algorithmic evolution of Allen Ginsberg’s iconic poem “Howl, for Carl Solomon” published in 1956 and of Muses & Self, an exhibition of Allen Ginsberg’s photographs shown at Fahey/Klein Gallery in 2023.

Ross Goodwin released HOWL.camera on February 19, 2025 as a generative + participatory work on fx(hash).

the transcript of the conversation which took place on February 19, 2025 at Fahey/Klein Gallery in Los Angeles, CA is presented here.

very special thanks to Fahey/Klein, the Allen Ginsberg Estate, and the Tezos Foundation for supporting the project + event.

– – –

[the conversation began with a group reading of “Howl, for Carl Solomon” in which guests read lines from the poem, cut up and randomly placed beneath their seats.]

Aya reads from “Howl, for Carl Solomon”

ELISABETH SWEET: “Howl” is a cry for the promised, for the promising, for the destitute, for the forgotten, the surrendered. For the brightest who fell into chaotic hells, who lost hope, and paid with their lives but for what?

I read each line of “Howl” as a headline, each a descriptor of a person Ginsberg knew – he’s telling a story with each line.

Take this one, for example:

“who threw potato salad at CCNY lecturers on Dadaism and subsequently presented themselves on the granite steps of the madhouse with shaven heads and harlequin speech of suicide, demanding instantaneous lobotomy,”

This line is about Carl Solomon, the man who Ginsberg dedicated the poem to. Apparently he really did throw potato salad at a lecturer at a university in New York, and I don’t know if it was a professor at Dadaism, but it was an act of defiance, an act of protest. Carl Solomon spent a lot of his life in and out of psychiatric wards, that’s actually where he and Ginsberg met. Allen Ginsberg’s mother had a long history of mental health troubles. She was committed at the [psychiatric] hospital while Carl Solomon was also there, and that’s where the two met. Eventually, Carl Solomon turned himself into the state and opted for a lobotomy, in order to prevent himself going back into the cycles of the psychiatric hospital.

Some may think that this poem is old and outdated, and it is 70 years old, but I think that there’s so much to it that we can still take. You can read the poem and its verses as describing symbols and hyperbole, perhaps metaphor, but there is something very real here, there’s a depth. There are people in every single story that Ginsberg tells through this poem.

After he wrote “Howl” Ginsberg reflected,

I sat idly by my desk at the first floor window facing Montgomery Street [in San Francisco]…I began typing not with the idea of writing a formal poem, but stating my imaginative sympathies, whatever they were worth. As my loves were impractical and my thoughts relatively unworldly, I had nothing to gain, only the pleasure of enjoying on paper those sympathies most intimate to myself and most awkward in the great world of family, formal education, business and current literature.

[then later he adds:] I thought to disseminate a poem so strong that a clean Saxon four-letter word might enter high school anthologies permanently…

And so here we are today.

In some ways, 2024 bears a striking resemblance to 1956, when Ginsberg released this poem. We’re still in a post-war, cold war, never-ending war era. Racism and mental health have mutated into something that we collectively cannot live within, and powerful poetry is even more than ever.

When you look around the room, you see Mark Laita’s exhibition, Soft White Underbelly. This show captures this vulnerable, invisible parts of society that people like to forget about, that don’t fit the reality TV program, don’t fit the shiny get-ready-with-me videos or whatever is going on social media these days.

This is real, and these are the people that I believe Ginsberg would have connected with and who howls through the poem. And so, the fact that this exhibition is up is really, really powerful. Maybe the simulation is working.

But there are, of course, big differences between 1956 and 2024. One of the biggest changes is technology, how we use it, how it looks, how we connect with each other, and how we make art with it.

So, now, to HOWL.camera. I’ve been talking for a while, and we’re gonna hear from Ross soon. HOWL.camera is a project that blends several areas of poetry and technology from the classic poem itself to the poetic devices within it, from Markov chains to generative language models.

Elisabeth Sweet & Ross Goodwin

Ross Goodwin has given us a new perspective through which to read one of modern literature’s most significant works of poetry.

To begin, I’d like to focus on the powerful, yet surprisingly humble tools that Ross, this scion of algorithmic intelligence, has chosen. I don’t wanna say basic, but he’s chosen really fundamental tools to computing. And so I’d like to start with asking about the computational aesthetics of HOWL.camera, why you chose a small language model rather than a large language model. Maybe we can even get into a little bit of the ASCII aspect of it.

ROSS GOODWIN: Yeah, sure. There’s a lineage to pieces. I did a piece before this based on “The Wasteland” [by T.S. Eliot], exploring that relationship between a language model and the material the language model is based on – the relationship between those two things. They’re not the same thing, and it’s much more apparent when you’re dealing with just one poem.

You can use a large language model on one poem, but the result of that is going to echo the vast depths of the internet and the many, many books that a large language model is trained on. Just showing [the model] one poem over and over again after it’s read the whole internet isn’t going to shape the work as much as you might think. It’ll start writing in that style, but it’s still carrying forward everything it learned before. The other reason I went with simpler tools this time is because I’m really disgusted with the tech industry right now, to be honest.

I don’t endorse a lot of what’s happening right now, in the AI universe especially. I think a lot of it’s really destructive, or it can be really destructive. I want people to question the use of these massive, massive, massive language models for virtually any application.

Don’t get me wrong, I think they’re great, I use them myself. In many ways, they’re very powerful tools. I think of technology, capital-T Technology, as a neutral concept. You know, it’s like the weather, or even more so it’s just a force of nature. It’s also very embedded in our humanity.

But, lower case-t technology is never neutral. I think that the way the tech industry is responding to current events, and the way that things are unfolding, especially with the dismantling of DEI initiatives, is really disgusting.

So I wanted to take a step back from using the cutting edge tools that are out on the frontier of computer science, which is where I started my practice 10 years ago. Those tools were very different 10 years ago, the cutting edge was a different place.

I wanted to reevaluate the need for such tools in the common context like this, because the idea behind HOWL.camera is not to make like robo Allen Ginsberg back from the dead.

ELISABETH SWEET: It’s explicitly not that!

ROSS GOODWIN: Yeah, the idea is to pay tribute, an homage to the poem. It’s supposed to encourage reading it in different ways and looking at it in fragments rather than looking at the whole thing, or looking even in ways that aren’t intended.

There’s a big gap between intent and perception. I think a lot of artists, art critics, and people in the art world will recognize that an artist can tend to a lot of different things in their work, but the way it’s perceived is ultimately what’s important in terms of its impact.

“Howl” is perceived in myriad ways and has been a powerful cultural icon for quite a long time. I don’t think that we can speak to which parts of it speak to which people. So looking at it through a fragmented lens lends itself to a perspective that embraces the perception side while looking through the intent side.

Ross Goodwin

ELISABETH SWEET: Really well said, I appreciate that. Something that’s really sticking out to me is that this smaller language model gets more specific, which can then be more personal.

And so through the personal, we say at theVERSEverse – well, this is not just us, people say this – but that the more personal you go the more universal your work becomes. And I think that that is something that I see in this work, where you’re rewriting “Howl” but very much from the bowels of “Howl” itself.

You’re not necessarily pulling from everything that’s happened over the last 70 years, you’re really looking at what language is and how language forms. And a little bit later we’ll talk about the photograph aspect and how that kind of comes into everything as well, how the photographs are then also reading the work.

Though maybe we’ll just talk about that now! Thinking about how a collector of this work chooses a photograph and then that photograph becomes the page upon which HOWL.camera reads and rewrites Ginsberg’s “Howl.” Then this ASCII visual comes up, and then within 10 seconds, the piece is screaming, howling the poem into existence, and howling the poem onto your screen. Can you talk a bit about that?

ROSS GOODWIN: I think it’s important to talk about ASCII visualization, because it is a reverberation, right? [That reverberation] is literally in the algorithm, that’s what it’s doing. It’s cycling through a set of characters that represent grayscale values, represent pixels in the image. And what it’s doing is pretty simple, honestly. It’s a couple lines of code to do that ASCII noise.

It’s more than that to set up the whole grid, and to get it all to sync, but just to make that ASCII is only a couple lines of code, and it’s a relatively simple operation. But what it represents is this reverberation. “Howl” has a reverberation to it, has an energy to it that transcends the medium that it exists in. And I think that a lot of Ginsberg’s photographs speak to his experiences that inform that poem. And the way in which those photographs can reverberate physically in the ASCII space, I think it speaks to a lot of those themes.

ELISABETH SWEET: I love that. It’s like an anaphora, which is in poetry, it’s a poetic device that uses repetition. So in “Howl” a lot of the lines in the first third start with “who” and then you have the section that starts “I’m with you in Rockland” at the beginning of every line. So this is an interesting anaphora in more of a technological sense.

ROSS GOODWIN: That’s a really good analogy. It’s literally one because the characters are repeating in the same way. And then I would also say that in photography, there’s a lot of importance in terms of AI, and the connection between the invention of the camera and the world we live in now. People who lived in the 19th century and witnessed the invention of the photographic camera had a lot of the same misgivings that people today have about artificial intelligence. In that they saw something, which had taken a measure of human effort, become automated and it made everyone really disconcerted.

People thought it was going to kill painting dead, but if anything, it set painting free in a way. It led to modernism. It led to painters really transcending photo-realism with the abstraction of more and more photo-realistic paintings. This is similar to medieval paintings – they’re not very realistic. Then you have Renaissance paintings, made when they were taking apart dead bodies to figure out how to paint a human form effectively, thinking about space and how bodies move through space and how things reflect light. And all of those things fed into the invention of the camera. Then the camera changed the system in which it was born into.

So I think AI is very similar. In large language models, you trade the camera for writing.

ELISABETH SWEET: So if photography freed painting, what does AI free in poetry?

ROSS GOODWIN: I think that is yet to be written. It could potentially lead to a world where people can write beyond themselves. I feel like we’re already in this world where I can write something better using Claude and ChatGPT or a fine-tuned model, if I use it in the right way, than I’d be able to write myself. And I think that we can look back to other historical examples, like the typewriter. You know, Nietzsche was the first philosopher to use a typewriter, and he thought that it didn’t just change the way he thought about writing, but it changed the way he thought in general. He called it “a mechanized word” or something like that.

There’s a really good book called Gramophone, film, typewriter by Friedrich Kittler that talks about this. But, yeah, the typewriter also feeds back into HOWL.camera because typewriter art is sort of the origin of ASCII art. People still make visual art using typewriters, and I’ve always found that tradition really fascinating visually and aesthetically.

ELISABETH SWEET: Thank you. I’d love now to kind of turn to the content of the poetry.

Early in our conversations about this project, we acknowledged that there are words in this poem that are considered offensive today. So thinking about censorship and how to bring “Howl” into today’s world, do we take out words that have caused grief and that we now recognize are insulting and derogatory?

We had a lot of conversations around this, as theVERSEverse and with Ross, and eventually we left it up to Ross to decide. I’d like to revisit that, and understand where your thought process went.

But before I do, I just want to make a tie-in that I can’t help but see with the Obscenity Trials that Lawrence Ferlinghetti the publisher of “Howl” underwent because of “Howl” [and other publications]. In the State of California, [Ferlinghetti] was brought to court for publishing obscene works that were suspected to be disruptive to society. He stood trial, and people brought evidence to court about how “Howl” had cultural value, and how “Howl” said something that had to be said. Eventually Lawrence won, and was then able to publish works that were censored previously like Henry Miller’s Tropic of Cancer. So without that trial, without that process, we might not have these other works that have really shaped modern literature in big ways.

And so thinking about how to censor “Howl” or if that’s appropriate was a really interesting thing for me to consider throughout this process, and especially now that the release has happened, so I’d love to hear your thoughts on it.

ROSS GOODWIN: It’s a really complicated question. For example, in 2018, I worked on this project called “Please Feed the Lions” with Es Devlin in London. If you’ve ever been to Trafalgar Square in London, there are four of these giant lion sculptures in bronze that were put there in the 19th century that symbolized the British monarchy and lots of things that I don’t know because I’m not British. Anyways, we put in a fifth, dayglo red lion. You could feed it a word and it would roar poetry back at you. A poem would appear on a screen in its mouth, and then at night it was projected all over the lion and Nelson’s Column behind it.

I hadn’t thought much about censoring the people who were putting words into the machine. Until the day I showed up in London, and my colleague from Google, who was sponsoring the project, came to me and said, so the mayor of London’s a bit worried that there’s going to be a little group of skinheads snickering in the corner of Trafalgar Square as all their code words go up Nelson’s column in giant white letters. And so I spent the next few nights on what’s called a whitelist approach. But that was a very particular case where we were trying to prevent a publicity disaster and a harm from taking place in a very public venue.

“Howl” is something that exists, and it’s not a symbol of the British monarchy, or a symbol of anything super mainstream at all. It’s really a symbol of counterculture. And I did not think it was appropriate to censor any of “Howl” in this project.

I came to that decision not to censor rap music when I was training models on rap music back in 2016, because that language is part of the medium in this case. I think there’s a balance, because you can’t say, oh, the machine said that, I didn’t say that, when you’re the creator of a system like this. You have to take ownership of the language. If your self-driving car kills somebody, it’s the engineer’s fault, probably. You have to take ownership of the language, but that becomes a really complicated issue I think when you’re dealing with something like “Howl.”

So it was a challenge to decide, but ultimately I just didn’t touch the poem at all. In this case, it felt obscene to tamper with it.

ELISABETH SWEET: Thank you. That was my gut feeling during it, during our discussions, but, you know, it is a question. I’m interested to see in the outputs how much those words are represented, or how much they could even be detected.

ROSS GOODWIN: The other thing is that the 1950s happened, right? We can’t erase the past, even if it’s super haunted and cursed. The truth is that “Howl” is a document that exists from a moment in time, and it reflects that moment. In the same way that 1984 is really more about what was going on in 1948 when the book was written. Science Fiction and literature and poetry, they reflect the times in which they are written and the fears and hopes and aspirations of those times. I think “Howl” is a really good example of something like that.

ELISABETH SWEET: I think that bringing that into the modern day, letting [those works] speak through and to understand what we’ve learned since then, and considering how those messages and those lessons still apply can show us how we’ve changed and evolved and grown since then.

I’m glad you brought up the Es Devlin project, because the next thing I want to ask is about this theme of participation that I see throughout your work. theVERSEverse curated a piece for an exhibition we had in Paris a few years ago, where Ross turned a car into a pen, specifically a 1995 Porsche 993.

He turned the car into a pen and asked the car to write a poem about the drive from the Eiffel Tower to Versailles. So, in this sense, you are inviting this machine to really do the writing.

In the Es Devlin piece, as you mentioned, you invite random Londoners or people traveling through London to feed words to a lion that roars poetry back at you.

I know that there are other examples of participation in your work, and I really noticed this with you. Especially with this collection, it seemed really important to you, at least I understood it as such, that the collector chooses the photograph that the algorithm then reads the poem through.

Why is participation important to you in this work specifically, in HOWL.camera? And then more broadly, how does poetry invite participation by people in ways that we might not always see so directly?

ROSS GOODWIN: You know, I’m a student of language and vision, and I think that something I’ve learned is that language is a participatory process. We’re all participating in the creation and destruction of our languages all the time, even if we don’t know we’re doing it.

Language is a very amorphous thing. We use roads that are older than the languages that name them. There are other things that we take for granted that are much older than the languages we speak. Language is something that’s constantly in flux.

And I think that to deny your role as a participant in the future of that is to deny your humanity in many ways and your participation in society. I think it’s really important for us to think about ourselves as active participants in the evolution of language.

That’s always been something I’ve thought about with this kind of work. So, I like work to be interactive and participatory. I think that it’s about elevating the conversations we’re having about language and technology and the way those things intersect.

The way to start a conversation, in my opinion, is to facilitate more through interaction.

ELISABETH SWEET: And why in HOWL.camera was it important to you that the photograph was chosen by the collector rather than some other aspect or some other mode?

ROSS GOODWIN: Well, you know, photography is a very subjective medium. Weirdly enough – people don’t think about it this way, and it’s going to sound absurd to some of you – but photography is the most abstract form of art.

[Photographers] are taking a slice of reality, they’re not deciding what goes on a canvas. The painter engineers their work in every way. A photographer has to find it, essentially. Or make it through Herculean efforts of staging and set design, but photography is a very abstract pursuit.

In HOWL.camera, I wanted people to feel like they were choosing the frame in which they were going to absorb this poem. I have other work that’s similar to this, work where I’ve used webcams to show people their own reflection, essentially like a mirror.

Good art is like a mirror. For something to be art, it has to reflect the human experience. Otherwise, it’s pretty useless. Somebody told me that a long time ago. And I really take it to heart. I think it’s true. Even through the lens of technology. If you’re making art that’s for AI, you’re not really making art, you’re making something else. And while I think that would be okay, art is about human experience, and it’s a reflection of humanity.

ELISABETH SWEET: And we can more deeply experience something if we’re participating in it.

ROSS GOODWIN: I think so, yeah. I think lived experience is the ultimate. A lot of my work is very performative in a way people don’t expect, I think.

ELISABETH SWEET: What do you mean by that?

ROSS GOODWIN: Well, I think of coding as a performance. You’re like, sitting at a keyboard and your hands are up doing this [marionette gestures], and thinking and solving problems in real time. Something I think about a lot is how a computer is an instrument that you can play like a musical instrument. A lot of technologies become that over time. Engineers build platforms and artists instrumentalize them. I like thinking about computers as an instrument that you can play, most of all. That’s my favorite way to frame the device.

ELISABETH SWEET: Thank you. Now I have literally no idea what time it is. But I would love to open up for questions from the audience, if anyone here has questions for – hand! Let’s go!

Peter Wu+ asks Ross Goodwin questions about HOWL.camera

PETER WU+: Thank you so much. That was a great talk. Thank you so much, Ross. Thank you, Elisabeth. One really basic question, and then a more complicated one. The first basic question is: are the images that were used for HOWL.camera, are they self-portraits that Ginsberg took? They’re of him?

ROSS GOODWIN: Yes there are self portraits.

PETER WU+: So he took them.

ELISABETH SWEET: So he took all the photos, and some of them are self portraits.

PETER WU+: Okay, so I mean, who are the subjects? Are they mostly him or are there other poets or?

ROSS GOODWIN: It’s the whole Beat Generation practically.

ELISABETH SWEET: There’s Willian Burroughs, Patti Smith, Jack Kerouac, there’s everyone, it’s incredibly iconic.

ROSS GOODWIN: Yeah, it’s this whole group of people. It’s really interesting to me, especially with Burroughs. Burroughs is an interesting figure. We talked about the cut-up method before. But like, Burroughs murdered his wife, and a lot of people gloss over that, by accident in theory. Still, it’s a really dark origin.

The story is, they were at a bar in Mexico and she supposedly challenged him to a game of William Tell. They all had guns, and he tried to shoot a glass off the top of her head, but he shot her in the face. They had to flee the country and he got away with it.

He was a really famous writer at that time, and he has a lot to say about it obviously.

ELISABETH SWEET: He says that without that experience, he would never have been able to write what he ended up writing, the best works of his life. It’s something he expressed remorse for, but still.

ROSS GOODWIN: Yeah, it’s interesting, it’s a complicated history.

ELISABETH SWEET: Super complicated.

PETER WU+: The second question is, the idea of… Okay, so you talked about how ChatGPT could help you and make something that’s even better than you could write. There’s this idea of perfection, right? Or this idea of “what is humanity” in this thing? Where do you define that line? Because mistakes in general are human. There’s going to be a point, I see in work, in art, and poetry, that we’re going to take all these things, even spelling mistakes or whatever. For example, when I’m writing a grant, and I feed it through AI, and it gives me bullet points, I will make sure that it’s not “too AI.” I’ll put some other things in purposely to make it seem more human. And I think that we’re going to go even more in that direction. Where do you feel that?

ROSS GOODWIN: I have a lot to say about that. I think that first of all, the idea of perfection is a delusion. Art is not about achieving perfection. It’s about pushing culture forward. There are a lot of perspectives on that. But I guess what I’ll say is that the type of machine learning that I do falls in this category that we call constructive machine learning, which is where you’re not trying to get it to a perfect state, you’re not optimizing. Optimization, the process of computational optimization – the idea is you reach an optimum. You reach the highest point on the graph and the lowest point, whatever you’re shooting for.

But what if the graph, what if the surface you’re looking at is in so many dimensions, that you don’t really know what the highest or lowest points are?

PETER WU+: The ladder never ends.

ROSS GOODWIN: The ladder never ends. There’s a theory that there’s an infinite number of dimensions. It just keeps going and going and going. So at what point is the local optima, or the local minima, or the local maxima, versus the global minima, the global maxima. These are different words that people in different fields have used.

It’s interesting to highlight the difference between the word optima and minima or maxima, which is the difference between how computer programmers describe the lowest and highest points on the graph. Something can be the highest point, but perfection is an entirely different question.

In terms of ChatGPT and how we work editing computer output, I think what’s interesting about this is that writing can be a subtractive process now. You can chip away at the block of marble until you find the statue inside, rather than just trying to build the statue up layer by layer. And I think that’s okay, and that’s a logical evolution of form. I think to resist that as the next step is not the right answer, but neither is to accept it without questioning it.

I think we all have a role to play in the discussion that happens now about how these systems are designed and what they’re used for. Because there are a lot of people who are saying that AI is for the defense industrial complex, it’s for murder, essentially. And a lot of people in the tech industry say that now actually. If you don’t believe that should be true, and you disagree with that perspective, I think that those voices need to be heard and they are being heard now.

PETER WU+: Thank you.

OSCAR SHARP: You observed one thing, that there are roads that are older than the languages we use to name them. You gave me shivers with that one. It’s been a while. Glad to feel them again. The name Ross is much older than Ross Goodwin. Should people, or gulfs, or mountains, care what they are named? And if so, why?

ROSS GOODWIN: Yeah, I think so. I think names are important. Names are points in a very high dimensional space that we call language. And names represent a landmark in that space, a definitive point. But they can also be little prisons, right? You’re drawing a box around some concepts when you name it. And the associations a name has, within a culture, within a language, really shape the way that thing is, the way that thing is treated by others, and the way that thing exists within the system that describes it.

So, yeah, I think names are really important. One of my favorite books from back in the days was Freakonomics by Steven Levitt and Stephen J. Dubner, and they talk about that, about an example where this guy names his kids Winner and Loser, and the guy named Winner ended up in prison, and the guy named Loser ended up being a successful guy, who goes by Lou. Or at least he was more successful than his brother, Winner.

But names aren’t your fate, they aren’t etched in stone. And that goes back to what I said before about language being this amorphous thing that we’re all shaping together.

Your name in a way is the most direct thing you can shape with what you do, the way you move through the world, the way you treat other people, and how other people think about you. That’s the biggest contribution you’re going to make to language, probably. That’s an overstatement, but yeah.

ELISABETH SWEET: But we don’t choose our names, I mean we can, but –

OSCAR SHARP: I did.

ROSS GOODWIN: So, just a second. As background, Oscar and I have a long history together. We made the movie Sunspring together in 2016. So, if you want to know about that movie or about that film in general, talk to this gentleman.

OSCAR SHARP: Although we didn’t name it.

ROSS GOODWIN: We didn’t name it, that’s very true. It was named by the sci-fi and filmmaking contest that we entered it in.

PETER WU+: Really? What movie is it?

ROSS GOODWIN: It’s the first movie written by AI. We made it to the Sci-Fi London 48-hour film festival in 2016, when we were both in grad school which is where we met, at NYU. Oscar was in film school and I was in this program called ITP, it’s like art school for engineers or engineering school for artists.

PETER WU+: Were you using TensorFlow?

ROSS GOODWIN: TensorFlow did not exist yet.

PETER WU+: Oh, Jesus.

OSCAR SHARP: This is old school.

ROSS GOODWIN: Yeah, [TensorFlow] was just coming out at that point. It came out the following year, in that timeframe.

OSCAR SHARP: We made another one a year later which predicted the writers’ strike in the face of artificial intelligence six years before the writers’ strike in the face of artificial intelligence.

ROSS GOODWIN: We used Torch.

PETER WU+: Oh, Jesus.

ROSS GOODWIN: Torch was the one, which is the Lua framework.

Oscar Sharp asks Ross Goodwin a question about names and the significance of naming things and people

PETER WU+: So for HOWL.camera, how big was the dataset that you used?

ROSS GOODWIN: It’s just literally “Howl” – just the poem – and there’s nothing else that it’s trained on.

PETER WU+: So when it’s generating so many images with the text, is that predetermined?

ROSS GOODWIN: Nope, it’s going to generate more and more forever, it’s all gonna be different. But they are fragments, so every five letters is directly from the poem. And it’s a character-level model, not word-level model. So it’s actually stitching words together as it goes.

It kind of knows where the word is, not because it’s literally defined, but because a word is sort of where space occurs. The space between spaces.

ELISABETH SWEET: It also pops the seams of each verse. There will be a word, and then that word is taken and brought down to this word.

PETER WU+: From each one?

ELISABETH SWEET: So, when you look at the work, different verses will start to appear, and then all of a sudden it bursts into the ASCII performance of “Howl,” and if you leave the code running, it will eventually be populated entirely by verse that is constantly changing and being reassembled and re-aligned based on the Markov chains that Ross designed.

MARGARET MURPHY: Can I ask a question that might help answer some of your questions?

PETER WU+: Please!

MARGARET MURPHY: Okay, great. In terms of the ASCII and then the code or the poem that’s generated, it’s doing so in the style based on, I would assume, highlights, shadows, contrast, gradients of the images in some ways, or it’s mapped through that. So obviously he has a huge archive of photographs. What was the selection process like for choosing the 54 that ended up being part of the renders?

ROSS GOODWIN: I used all the ones they let me use. I wanted to really share the whole range of photographs and not try to curate them. I thought about curating them and doing some other stuff, but I thought it would be more powerful to use them all.

ELISABETH SWEET: And Nick, correct me but all the photographs were from Muses & Self? Yeah, so Muses & Self was the Ginsberg exhibition that was here in 2023. All those photos were used in this collaboration. When you go to the Fahey/Klein website, which we’re going to link to on the HOWL.camera website, you can find the captions for each of the works. You can see who is actually in it and when the photo was taken. There’s more context there that we’re going to build in.

ROSS GOODWIN: There will be a website for HOWL.camera, it’s just under construction.

LAURENCE FULLER: So it doesn’t know the meaning of the word in a broader sense, just within the context of the poem it understands?

ROSS GOODWIN: It’s literally going: what letter comes next, what letter comes next, what letter comes next? And if a letter came next more often than another letter, it chooses that letter. There’s sort of an equation by which it decides which letter. The equation is called Unsmooth, Maximum Likelihood, which is the version of the Markov model that I discovered in 2015. I think it’s the highest quality Markov output you can get. Or it’s just the most interesting to me because it exists at the character level. You’re viewing the language more atomically when you look at the character level. Language isn’t even just made of characters, it’s made out of lines and dots. A character is a set of marks in space. And that’s true across all languages. I think the more we can drill down to the building blocks of language, the more we can understand it in certain ways.

MARCELA VIEIRA: Considering translation, when you translate a book or a poem, you are bringing a poem or novel into the language of the day. Do you consider what you are doing a kind of translation?

ROSS GOODWIN: Yeah, I think so. That’s a really interesting way to put it. I think it’s an adaptation more than a translation, but I think translation is another way to think about it. It’s more like translating two media into one. It’s kind of like a forked process or something. So yeah, adaptation and translation, I think those are adequate words to describe what’s happening.

AYA: So back to the small language model. My understanding, from what you described, it sounds like it’s making decisions more on frequency or representation of the word, how often it’s showing up. But I was wondering why you didn’t consider something more semantic, so really understanding what the word means. I wonder if that’s because it is a small language model. Is that the reason versus how large language models have that history of knowledge to understand semantics?

ROSS GOODWIN: I’m going to quote Oscar. Oscar has this great quote that I quote all the time, which is that meaning is overrated. At least in art.

OSCAR SHARP: Wow, I haven’t said that in years.

ROSS GOODWIN: You haven’t said it in years, but I think you should say it all the time.

PETER WU+: How do you feel about it now?

ELISABETH SWEET: What does it mean to you now?

ROSS GOODWIN: It’s not to say that meaning is meaningless or meaning is bad or that meaning is less than, it’s just that meaning is not the only thing that is important in a piece of artwork. And it goes back to that thing I said about intent and perception, those are the two halves of a piece’s meaning that you drill down to at any level. It’s this really crazy fractal shaped thing that you can constantly describe in more and more detail and never really get to the answer. Meaning is, I think, a rabbit hole when you’re talking about systems like this.

The one thing I would say is you’re right to point out that a statistical model is going to be biased in ways that the original author or Ginsberg wouldn’t be, it’s going to be biased in ways we don’t expect.

In response to that, the Unsmooth Maximum Likelihood model that I use is my authorship on this project: the choice of algorithm that I’m presenting. Part of my role in authoring this is deciding the aesthetic of what’s going to come out.

I think that the Unsmooth Maximum Likelihood model, for whatever reason, is more representative, it does a better job of capturing the essence of a piece of work when it’s presented this way.

view the collection on fx(hash)