A conversation with the Melbourne-based poet and researcher on robot poetry and what humans can learn from computers.
Oscar Schwartz is thinking about whether robots can write poetry. His website bot or not asks users to guess whether poems are written by algorithms or humans. This Turing test is disconcertingly hard, and it raises all kinds of questions: what is ‘human-like’ poetry? Could computers teach creative writing? And what does it mean to be human in an age of digital creations?
* * *
JOAN FLEMING: Earlier this year you launched bot or not, a Turing test for poetry, where readers have to guess whether a poem is written by a human or by a computer. What kind of a response have you gotten to bot or not?
OSCAR SCHWARTZ: Unexpectedly bot or not went viral. I spend a lot of time on the Internet and I’ve learnt that there is no predicting what is going to be popular and what isn’t. I wouldn’t have thought bot or not would be popular: it’s about poetry, and not many people read much poetry these days. But, for whatever reason, it went viral… We’ve had something like 250,000 unique visitors to the website. It’s been covered by Washington Post, The Awl, Huffington Post, Popular Science and a number of other publications. In the first few months of it being launched I became addicted to Google Analytics.
But probably the most satisfying thing about its popularity was watching my friends, who don’t know anything about poetry, play it. I was at a friend’s house. He works in the finance sector, and isn’t much of a reader. I showed him the site and started playing. Next thing I know he was shouting: “God damn it! Gertrude Stein got me again!” I never thought I’d hear him say that.
“Next thing I know he was shouting: ‘God damn it! Gertrude Stein got me again!’ I never thought I’d hear him say that.”
JF: The test stumped me. I got to 7-correct, 8-incorrect, before I stopped. I nailed Tao Lin as human, but I was sure Bukowski and Tristan Tzara were computers. What are the characteristics, do you think, of “computer-like” poetry written by humans?
OS: That question is basically the central question of my whole thesis, so I haven’t really got an answer for it as yet. But here’s what I think so far. The first thing to remember is that the results of bot or not don’t tell us what actually is human or computer-like in poetry, but what people perceive as human or computer-like. So the most human-like human poet is William Blake. This might not be a surprise. His poetry seems to possess a very human type of wisdom. But Gertrude Stein was voted as one of the least human-like human poets. Does this mean her poetry has less wisdom, or just that the wisdom comes in a more difficult form? I think maybe people perceive Blake as human and Stein as not because we’ve been culturally conditioned to perceive poetry as a Romantic, rhyming pursuit. Anything that deviates from that, for those how haven’t studied modern poetry, is unfamiliar and therefore “not human.”
JF: What are the limits of the bot or not test?
OS: I’ve been grappling with this question. A crisis I’ve been having is whether or not to add a comments section to the bot or not website that allows people to give the reasons as to why they think a poem is human or computer-like. As it stands, it’s just a binary option: yes or no. Computer or human. I think I’ll keep it like this though, because it kind of embodies the way that humans are progressively being reduced to data points in a digital world. It’s a little perverse to treat a human opinion like binary data, but it also raises interesting questions: are humans simply complex machines?
An example of a poetry test on bot or not.*
JF: Funny thing about Gertrude Stein—I read her work, and it feels absolutely, characteristically Steininan. Are there any computer-generated poetry machines that produce surprising and interesting poems that have this quality of cohesion and recognisability? As in, “Oh, that’s a racter poem” or “that is such a Janus Mode poem.” Could an algorithm generate a whole series of poems, or a whole book, or several books—an oeuvre!—that would feel characteristic enough to convince its reader that it is written by a human being?
OS: I guess what you’re asking there is whether a computer could ever develop a sense of ‘style’. Technically, the answer is almost certainly yes. In fact, that’s exactly what an algorithm is: it’s a style of writing that is systematised and never strayed from. An algorithm will have the most consistent style you can ever imagine, because it will never stray from its own rules.
But, and this is a big but: maybe style is not about consistency but about development, i.e. when I read Gertrude Stein’s collected works I notice changes in her style. These changes, coupled with the biographical facts around her life, create an aura around her aesthetic progression. I believe that this aura is what we call the ‘voice’. In fact, I would say that it is a poet’s ability to break their own rules that gives them a voice to being with… Maybe with the development of artificial neural networks and machine learning a computer could go through this process of aesthetic development… I don’t think we’re there yet.
JF: Besides a “type of wisdom,” what are the characteristics, do you think, of a human-like poem? What are the characteristics of a computer-like poem?
OS: According to my research so far, the poems that people perceive as being human-like are: simple, prosaic, rhyme, use words with short syllables, speak about death or an experience of the body, are written in a traditional form, appear to impart some often platitudinal moral lesson and use concrete rather than abstract language. Poems people perceive as being computer-like possess: irregular syntactic or grammatical structures, longer words, a lack of consistency in the type of words used, more abstract words, no rhyme, a lack of narrative, have a ‘meaning’ that is much harder, or almost impossible, to interpret.
JF: You are a scholar, and you are also a poet. Would you consider robot poetry one of your influences?
OS: It’s an intellectual influence, definitely. The concept of a computer writing poetry has enabled not only my scholarly writing, but a lot of poetry. It just gives me this new way of perceiving language and humans that opens up all of these new approaches to poetry… and I guess reality and existence too. Being a poet, computer-generated poetry provides constant existential doubt. That being said: I’m not interested in reading poetry produced by a computer, because, at this stage, I don’t think a computer wants to write poetry. And that’s what interests me most about other people’s writing: trying to figure out why the hell they actually wanted to sit down and write. My favourite poetry is that which seems to come from someone having an irrepressible desire to communicate their experience of existence with another person who is also existing. I don’t really like poetry that is prodigiously technical or whatever. I like amateur poetry, expression, etc.
“My favourite poetry is that which seems to come from someone having an irrepressible desire to communicate their experience of existence with another person who is also existing.”
JF: Many intelligent people are insisting that we must prepare ourselves for the age of the automatons—in all parts of commerce and society, not just self-driving cars, but robot baristas, doctors, lawyers. I feel troubled, even repulsed, by the idea of an automaton teaching creative writing or literature courses. But when I think about how much I repeat myself in first-year tutorial discussions and feedback (“use concrete imagery!” “avoid abstractions!”), then maybe it’s not such a ridiculous idea. Any thoughts on this?
OS: Machines have been taking human jobs for millennia. The computer’s capacity for simulation of intelligent human functions means that this moment of automation is qualitatively different from past moments of automation, I think. For this reason, the question of whether to automate something will become a moral/philosophical question rather than a question of engineering and programming. That is to say, in considering whether we should make computer teachers the question shouldn’t be: can a computer teach creative writing, but rather, ought a computer teach creative writing? This transition from ‘can a computer do x’ to ‘ought a computer do x’ will be an interesting transition.
JF: What, if anything, do you believe the digital turn can show us about what it means to be human?
OS: I think maybe more than anything it has taught us how little we actually know about ourselves. For example, Aristotle and then later the Enlightenment thinkers believed that rational thought was what separated humans from animals, and represented the highest function in a human being. We’ve now made computers that possess a far superior capacity for abstract reasoning. In fact, those qualities that were once considered to be ‘the highest’ or the most human have been the easiest for a computer to simulate. What has been difficult for the computer to simulate are simple things, like telling the difference between a cat and a coffee mug, or moving around a room. In other words, the digital turn has taught us that the easy things are hard and the hard things are easy. I think, for me, it’s made me more aware of the complexity of my body.