This week, we sat down with published poet Andrew Levy, a professor at The City University in New York. We discussed using AI to assist in writing poetry and creative writing. We also ran one of Andrew's poems, The Putative Tiring of Light, through an AI text generator and we discuss the output on the episode.
Andrew's most recent publication, Artifice in the Calm Damages, can be found here:
Check out more of our work with GPT-3 below:
Practical Applications of AI & NLP for Automated Text Generation
Modern AI Text Generation: An Exploration of GPT-3, Wu Dao 2.0 & other NLP Advances
Deep: Hi there I'm Deep Dhillon. Welcome to your AI injection, the podcast where we discuss state of the art techniques and artificial intelligence with a focus on how these capabilities are used to transform organizations, making them more efficient, impactful, and successful. Welcome back to your AI injection. This week's episode, we'll continue our explorations in AI assisting in creativity and art. This week, we've got Andrew Levy. Andrew's a published poet and professor at the City University in New York. Levy's most recent publication Artifice In The Calm Damages is in his own words, 175 pages of poetry, prose, essays, graphics, and the uncategorizable. We took some of Andrew's work and passed it through an AI text generator. We'll be discussing what came out as well as some of Andrew's thoughts on using AI to assist in creative writing. So Andrew, get us up to speed on your background in poets and how you approach writing and how that might be different than what most folks think of when they think of poetry and poets.
Dr. Levy: So I basically, you know, didn't even get turned on the fact that anybody even wrote poetry until I was about 20, because of my encounter with a professor who was an editor of a great magazine called Vort. And it was formatted where every issue was dedicated to two writers, he would have an extensive long interview and then the interview would be followed with a sample of their work and then maybe half a dozen pieces by contemporaries who could say something about their work and put it into a context, right? So they were experimentalists. For example, one of the writers would've been Jackson McClow, who was a friend and contemporary John Cage. They were both using chance generated methods for producing their work. Both of them later in their life actually did experiment with computer generated texts, using early programs where they could do diastasis and acrostics. And so that, that's kind of like my grounding in the idea of art as experimental.
Deep: Kind of along those lines. Like, I mean, within the world of experimentalism, what do you think the role of the machine is? You know, I mean, whether it's the pencil, you know, and pen or, you know, the writing, um, the laptop or other tools.
Dr. Levy: It, it's funny, you mentioned that now, because I hadn't thought about this in a long time, a well known poet, Charles Bernstein. I remember that he and I had a conversation back in the eighties, you know, when people were getting their first PCs, right. That had about as much power as a pencil. And it was interesting at the time because the people that I knew that were getting computers first were some of the poets, not because I think they thought they necessarily were gonna write their poetry, you know, in their primitive word processing program. But I think just the whole idea of the machine just titillated people's imagination somehow. And then also with the advent of the internet, that possibility to communicate instantaneously at great people was very enticing. But what I remember is that I was talking with him and he was like, I'm not gonna write my poetry on the computer. I'm still doing that by hand. For some reason, some of the people that I thought of as the most experimental, it was like this kind of, uh, hesitancy to see the machine as a way to facilitate the composition of the poetry.
Deep: So, you know, I know I, um, I exposed you to very large, uh, scale projects that are going on to train machines on, in essence, the sum of human publicly available writings in to learn language and say things like most folks are quite amazed. The first time they see the output of these systems, they're just like, wow, this very much feels like an actual human. We're just seeing like an uncanny output from these machines. And I guess a question for you is like, what was your first reaction when, you know, when you started seeing stuff like this?
Dr. Levy: Well, my most intensive introduction to that is through another friend and colleague, whose name is Andrew Klobuchar. And he's a professor at N J I T kind of his specialization is, uh, digital poetry and algorithmic derived work. And a few years ago he was showing me some where these artists slash performers were using computers in live performances and they were writing code to treat and generate language. And then they were also doing this with music. And the thing that caught me up gave me a little bit of trouble in totally appreciating what he showed me or shared with me was a big emphasis of this idea of improvisation. They consider themselves improvisers. They would talk about their coding, et cetera. And these performances as works of improvisation.
Deep: You're listening to your AI injection, brought to you by xyonix.com, that's X Y O N I X.com. Check out our website for more content, or if you need help injecting AI into your organization.
Kinda, I wanted to go back to your idea of identity, cuz I think this is like a core part of the, you know, the question here is like with these systems, like what does it mean to have an identity when you've assimilated all of the writings of all of the humans that have ever put anything out that winds up on the internet, like what's the identity there?
Dr. Levy: That's a good question. Like, you know, I don't even know if identity is the right word. You know, there's another writer who is working on, I guess, some kind of career review of my body of work, right. So far. And he wrote me and he said, you know, the question I still have is who is Andrew Levi? You know? And he is like, what's your identity? Who is the poet here behind these words? And I think it was because I problematize that issue of identity of the author authorial identity in, in the work itself. And some of the writing in the poetry directly address that.
Deep: Decades ago. We used to talk a lot about this. I, I remember a bunch of our conversations of the, you know, the role of the poet as the, you know, the channeler or the anntenae, if you will. And.
Dr. Levy: Like the Spicer or spicearean radio. Idea.
Deep: Yeah, yeah. The stripping out of the ego and you're basically like a probe into this world and you're just kind of like trying to funnel it, which is very anti traditional poetry in many ways. Right. Because traditionally, like it's kind of all about the assertion of I identity and you're sort of intentionally stripping out well.
Dr. Levy: Yeah, no, you're right. You're right. But also that traditional idea is complicated because who basically dictated their work to them. And as you know, Yates has a whole book that he says was delivered to him that way called a vision. Right. Uh, in which his wife basically was the channel through which the book was delivered.
Deep: If you saw this project by Ross Goodwin who loads up this black Cadillac and he, he uh, basically turns it into a quote antenna or channeler, if you will, it's got a video camera up on, on the front, it's got microphone, GPS coordinates, and he's trained up kind of natural language generation from these signals. And then he recreates some of the road trip of, of Kerouac, you know, in, um.
Dr. Levy: Yeah, I think I read an article about this.
Deep: Yeah. Yeah. He calls it one the road instead of on the road. And then the machine is, I mean, it's kind of literally, it's hard for us to think of the machine as having an ego in this sense, cuz it is it's, it's almost closer from the poetic sense of what we're talking about, but it generates some pretty wild, amazing, insightful stuff. And it's all based on what it sees while it drives across, you know, Kerouac's journey and it's seeing things, it's hearing things it's writing verse along the way, when we think of identity, it's kind of got an identity, at least in the sense that it's this thing that drove across the country, you know? Right. It's also got an identity in the, a sense that it's a collection of everything it's been trained on. But yeah, I mean, I think one of the places I wanted to take this conversation to was what are tools that could really help, I can like dig into this elaborate world.
Dr. Levy: So you remind me of the kind of question I wanted to ask you was in your mind, in your imagination and what would you imagine these AI systems to be able to be?
Deep: Yeah. I mean, that's a hard question to answer, cuz I think at the end of the day, like none of us really know how these algorithms work. I mean we know how they work. Yeah. We can author their architectures. Like we know, okay, well there's, you know, 28 layers and you know, like the, the lower layers or this and the higher layers of that. Yeah. And we can sort of model them and we can sometimes like stick little probes in there to see like what are examples of what this layer is seeing. And that layer is seeing, but we don't really know what they do. We can't wrap our heads around the 175 billion parameters these are like knobs and dials that this model iteratively tuned over many, many iterations as it was trying to fit a cost function. Basically trying to predict the next sequence of characters in words, across the entire sum of human knowledge. And in the process of that, it built up this network that we can't really wrap our heads around it any more than one of us can look in our own brains and figure out how our neurons are firing. We don't know that we just know what we trained it on. We know what that does in smaller, more predictable context.
Dr. Levy: Wait a minute. Are you saying this system is tuned and trained to digest different kinds of information for largely predictive purposes?
Deep: Yeah. That's exactly what they're they're prediction machines that predict future character and word sequences. If you were brain dead and now all you see is text and then you were only given the task to predict what the next word would be based on the text you've seen, that's where these systems are at the end of the day.
Dr. Levy: How and why have they been designed to be prediction machines?
Deep: Because as a result of that, you end up with a neural network structure that in essence understands language. So what what's happening in this system? Like when we give it your text, it's like your poem. Yeah. It's taking that poem and it's like navigating this complex information space that it's constructed because it's just the first time we've really built things where they're so powerful that we're just surprised by them. It's like kind of uncanny. So that seed text, if you will, is kind of basically getting it to navigate that find patterns in all the language that's ever been written. That's closer to you and what you are writing. One of the things that's happening is it's also kind of in a sense sort of like ignoring a lot of the other stuff you wanna just like read that poem of yours, the punitive tiring of light. And if you can recite that for us and then we'll look at what the machine's doing and then let's talk about it. Like, you know, what's interesting and what's not whether it's you or whether it's not you.
Dr. Levy: Yeah. All right. So the title of this piece is the punitive tiring of light. It's dedicated to William Fuller. "So the paradox of capitalist labor is a fine thing, but is not the place of sanctity. One dwells in the recognition of a Demi Quintes hours of toil. That very with position or time that eventually leaves one's pocket ghost of the departed slowed to an adagio paralyzed by its haste, everything else, rots in hell in the unenviable position of having to study work in points of departure, perhaps some form of manual control and neither very rational dialogue with one's material takes the optometrist job that smells of laundry soap and mildew along the hallway where the spiders hang and wait. I mean that literally and metaphorically adults who cry when they die did not have mothers who would comfort them. What is constant? If it exists is not really constant. We're not people we're lithographs. We're only real between acts the best of our domestic implications. Allow one to expire, re blend it with desire. Part of the unstoppable drift in time in thus operative, imaginative trees, becoming stars, creating space climbing down from my tree. The sole of my right foot keeps the only knowledge it retains on the top of my sandals. Striving in both construction and surface may be something ordained to service, a cactus garden, birds humming insect in the needles, secu by a regimen of idleness in abyss of saffron fertilizer, Exxon tanks, soap plant spilling into key streams, neuro molecular death, turning down the lender of last resort. Genius conjectures on lonely fantasies accidentally imprison in retirement to avoid an apology. Seeing that intimacy is effortless. It has no need to come inside one's cabin and waving its feet in safety deposit, envelopes of the divine, which heartbroken calls from a huge bedroom and cries. The impeded stream is the one that sings."
Deep: Awesome. So I'm gonna read for you the quote ending. So we fed this into G P T 3 and GPT3 three comes back and continues your poem and says, "It is easy to believe in the end of the world and even easier not to, I know it will happen because my father told me, so the first half was a lie and he would never tell me that part. But I have an idea about what happened before then. we were all naked and running around when someone invented clothes, it didn't matter who did it or how they came up with clothing after that, everything changed.
Dr. Levy: Yeah. And looking at that now I'm like going, wait a minute, have, have you, and I made a mistake, are we sure this is the language that was generated from this poem and not one or the other two?
Deep: No, this was, this. Is it. This is it.
Dr. Levy: Right. And so it's kinda like, you know, so this is really interesting, you know, when I first read it and got back to you, I think this is the one where I said, well, this one's funny. And by funny, I met interesting, right? Like the other one seemed more prosaic. They almost seemed like kind of literal explanations of the content of maybe something in one of the other two pieces. Whereas this got off and went in a slightly different direction. And also, you know, on the level of some of the word choice content here, I mean, it says, my father told me so well, there's no mention of a parent figure in the poem. There's no mother there's no father. Yeah. So it's like, how did, how did that come up? It's funny. And it's kind of almost nonsensical in parts and yet it still makes sense. And then all this stuff about lying the first half was a lie. Uh, we were all naked. We were all running around and then somebody invented close. I mean, that sounds like a joke, right? It's like, that's the part that made me laugh. I was like, what? You know, cuz I like that, you know, that was kind of funny. And um, it turns out that this same poet, William Fuller and I started a actual collaboration. So what I'm gonna do is on my, my next contribution. I'm gonna put this in Without telling him it's origin. Okay. And I'm gonna see what happens. And then I'm gonna tell him what its source is and how it was generated. And then I'm just curious to see like, is he gonna be pick up on anything in this because this is possible. I think whether written by me or somebody else or by the computer where some of the other pieces that we tried. Right. Didn't really gel, you know? Uh. Yeah. They didn't capture the kind of poetic. Yeah. That that's what is actually what I would call kind of imaginative. Right.
Deep: Yeah. And I think, you know, going back to that question of tooling, so there's this idea of levels in, in creation, you know, and, and I kind of envision like of the next 10 years, you know, we're gonna see an awful lot of creative tooling emerge. That's gonna allow people to like walk up levels and uh, and interact at the kind of levels they want to. And I think you'll have the ability, you know, at this, what I would call this as kind of a high level, right. Like you're saying, Hey, finish my poem. Like that's a high level act, but there's other levels like where you're just like kind of, you know, like there's simple stuff. Like give me some synonyms here. Well, you also made me think about, you know, the use of constraints to generate work is like the APO group, right. The famous artist writers' group and yeah, they would same thing. They would set up very crazy constraints, like right. Entire novel without using the vow. E that's. I mean, I really love this idea of you've got some constraints that you put down or some tooling that helps you tap into the us higher, deeper level of understanding of language that you can kind of reverse engineer into some tooling that ultimately helps put you in uncomfortable places or, you know, as a writer or helps challenge you in some way that, you know, is, is kind of, you know, just out of, of your comfort zone. I mean, that feels to me like a type of tool, that's very different than like what we've got so far.
Dr. Levy: And right. A kind of maybe last question for you. Okay. That I think kind of segues on that. And it's from the poem that I just read. So the first two lines, the paradox of cap labor is a fine thing, but is not the place of sanctity. So can AI be this place of sanctity in our culture?
Deep: I mean, that's up to us, right? Uh, machine learning, artificial intelligence as of today and maybe the next five years, my guess is it's pretty much in the tool department. So if you use it to build a video recommendation engine that optimizes engagement on YouTube, that's a capitalist endeavor. But if you use it to have an assisted poetic experience and the poetry you're writing is transcended in some way or helping you get there, then you could say that the AI is a thing of sanctity.
Dr. Levy: I just heard that in what you were saying before that this could be a really creative endeavor in tool. You know, all creative endeavors can be monetized, I guess, but in a way for the individual artist working who is not able to monetize anything, it could be like a very different kind of tool, a very different experience.
Deep: AI's going to be the most prolific presence in our We just might not see it as such in the same way. We don't think about it. I mean, that's, what's gonna happen with this stuff. And it already is like, there's a bajillion parts of your life. Every day, you touch your phone where machine learning, AI systems are constantly impacting your life. You just don't think about it that way. But going back to your safety argument, like to take it even a step further, like one of the things these systems do, then this would kind of go directly to kind of like my personal sense of spirituality. I don't know. I don't know how we got there on this podcast, but like what the heck, why not be there in my sort of spiritual sensibility is the goal quote of a life is like where you're not thinking of yourself as this body and this individual perspective, but you're thinking of it as everything, all the living beings and humans and, you know, and creation. And you're trying to process reality through the lens of what's good and wrong, right. And whatever for that collective entity. And when I look at these systems, that's kind of by default what we're doing these days, right? We're by default a similar all of humanity into these language models, by default into these imagery models. And then now we're starting to couple the two with like, you know, text and images and eventually it'll be audio and all the senses. And, and then they're always kind of by default because it's just, we're always looking for more and more and more data. They're they're by default gonna take the collection at least of humanity. And so I don't know, like, is that at a sanctity? I mean, I feel like there's some lesson there, you know, in terms of like the way we're thinking about knowledge in these machines as being collective and not like individualistic, I think there's some lesson to be learned about translating that back into how we actually, as our own constrained human minds think, but, but yeah,
That's all for this week. Thank you so much, Andy, for chatting with us. If you're interested in reading more of his work, check out Andy's book artifice in the calm damages. And if you're interested in learning more about GT3 and other AI generator texts capabilities, we've got a blog article up on our website with more details on this topic, you can find that at xyonix.com/articles, that's xyonix.com/articles. That's all for this episode. I'm Deep Dhillon, your host saying check back soon for your next AI injection. In the meantime, if you need help injecting AI into your business, reach out to us at xyonix.com. That's x-y-o-n-i-x.com. Whether it's text, audio, video, or other business data, we help all kinds of organizations like yours automatically find and operationalize transformative insights.