Credit: Gerd Altmann via Pixabay.
Natural language processing is, according to AI expert Melanie Mitchell, “probably the hardest problem for AI.”1 But natural language processors get help from human interlocutors, who tend to assume AI has human emotions and motivations. Programmers call it the ELIZA effect, named after a 1960s chatterbot that imitated a psychotherapist. Today’s NLPs draw on deep neural networks and learn from interactions, inviting us, ever more convincingly, to participate in the illusion of subjectivity. In this respect, AI is like lyric poetry. According to Jonathan Culler, “to read something as lyric” is “to convince ourselves that we are hearing a voice.”2 Poetry, like NLPs, creates what Culler calls “effects of voicing,” the impression of a speaker.
Contemporary American playwright Joyelle McSweeney experiments with these forms of illusory subjectivity in her verse drama. She is especially interested in self-generating or self-replicating texts—any language that is produced or reproduced on its own. Her play Dead Youth, or the Leaks explores textual agency in genetic and digital forms. McSweeney uses both kinds of coding as figures for poetry, which is itself a kind of code.
The play depicts the journey of a group of “dead youth,” recently rescued by Julian Assange, who hopes to “reboot” them. Along the way they meet Antoine de Saint-Exupéry, French aviator and author of The Little Prince, and Abdi Wali Abdulqadir Muse, one of the Somalian hijackers of the MV Maersk Alabama. Henrietta Lacks delivers the prologue and epilogue, bringing together an otherwise chaotic amalgam of events and references. This character is based on the historical figure whose cells were harvested without her knowledge for use in cancer treatment and medical research. Lacks’s cell line is a ready figure for texts that grow, change, and proliferate on their own.
The eponymous dead youth speak lines that seem scripted or programmed, as opposed to intentional. Early on, one says to another, “Some bug’s got into my brain. Some kind of post-death inflammation reanimating me. My circuits are firing like a firing squad. Repeating like a repeating rifle.”3 Just as the post-mortem movements observed in some corpses are sometimes taken as evidence that the person is still alive, the audience is tempted to take the dead youth’s speech as an indication of living consciousness. But they are more like automatons than living beings.
In a 2017 essay, McSweeney calls this kind of writing “posthumous poetics.”4 The term designates a text that survives and proliferates outside of the author’s control. I see it as a species of prosopopoeia. This “master trope of poetic discourse,” as de Man calls it, is the attribution of speech to an imaginary, deceased, or inanimate entity.5 McSweeney does this throughout the play, but her characters know they are being ventriloquized. Hers is a kind of postmodern prosopopoeia that draws the audience’s attention to the fact that it is an illusion. De Man uses the Greek etymology of the word—poien, meaning to make or create, and prosopon, meaning a face or a person—to derive an account of its function: “positing voice or face by means of language.”6 The trope enacts both “giving and taking away of faces […] figuration and disfiguration.”7 McSweeney’s play also “deals with the giving and taking away of faces”: it makes the dead speak and gives them faces. But anthropomorphism is not the right word for what McSweeney does; it is more like automatomorphism, a sham of artificial intelligence, rather than human subjectivity. It creates an interface, not a human face.
There are many instances of automatomorphism in the dead youth’s lines. Here is a characteristic example:
DEAD YOUTH 1: I only regret that I have but one life to give—
DEAD YOUTH 2: —to the hive mind. Bury me standing
DEAD YOUTH 1: In the checkout line. I don’t want to lose my place.
DEAD YOUTH 2: I’m double couponing.
DEAD YOUTH 1: I’m saving for a doublewide.8
Here they sound like they have been scripted by a Markov chain, a form of AI that predicts what letters or words come next based on the patterns in the data. This effect is particularly conspicuous when they finish each other’s sentences in nonsensical ways, as in “Bury me standing […] in the checkout line” above. It is as if they are reacting to sensory stimuli rather than the words’ meaning.
Elsewhere, the dead youth generate lists in response to some linguistic trigger. Often these lists seem like part of a ritual or game, as in the following exchange:
JULIAN ASSANGE: Now who can tell me what IV’s, hypos, bee-stings, and bone meal all have in common.
DEAD YOUTH: They are all vectors!
JULIAN ASSANGE: That’s right, geniuses. Of what?
DEAD YOUTH 1: Infections
DEAD YOUTH 2: Nutriment
DEAD YOUTH 3: Antigens
DEAD YOUTH 4: Information.9
The dead youth have answers ready, as if they have been programmed by Julian Assange ahead of time.
Sometimes they produce lists without prompting. These passages resemble the output of more sophisticated forms of AI, which are capable of creating their own rules and generating novel text. In Act Two, Julian Assange’s reference to his mother provokes the dead youth to generate a list of 56 more mothers. The litany begins with a seemingly random assortment of women:
My mother JonBenet and me
My mother Margaret Thatcher.
My mother Henrietta Lacks.
My mother Antigone.10
Then they move beyond the human to things like: “sleeper cell,” “human error / Yellow cake or Zyklon B.”11 This haphazard list mirrors the hidden process of trial and error through which many forms of AI learn. Many neural networks reproduce unusual words and phrases with disproportionate frequency, revealing the idiosyncrasies of the training dataset. This algorithm seems to have been trained on bleak news headlines from the last 30 or 40 years. As with real neural networks, there are outliers, like “Antigone,” especially in early stages of development.
As the list continues, there is a subtle movement from randomness to coherence. Groupings start to emerge. The first four entries are all women, and the second group are sources of hidden danger, like sleeper cells. Then the algorithm shifts, and they begin to describe a specific, hypothetical mother:
My mother bared midriff, dirndl, sari,
sandal, buckskin, wristwatch, hijab,
Who survived my birth
Whose idea of groceries
was a bottle of bleach or pills12
The descriptors seem random at first, but they eventually cohere into a vague storyline that admits of several variations. Maybe she was wearing a buckskin or a hijab, maybe she committed sabotage or robbery, but in any case, she was a bad mother who suffered a tragic downfall, bringing down an institution along the way, like Antigone. This passage is reminiscent of the way neural networks organize an enormous amount of data in order to discern and replicate patterns. Over time, and sometimes with course correction from a programmer, these AIs can produce better and more coherent results, although it is difficult, if not impossible, to determine what rule the algorithm is determining for itself. At the same time, juxtaposing clothes associated with marginalized identities—especially the “hijab”—with references to terrorism, is a troubling reminder of how many forms of AI not only replicate, but often amplify racial and ethnic bias.
Elsewhere, the rules are more obvious, and thus, more like a traditional computer program. These rules are often sound patterns like alliteration, rhyme, and puns. For instance, this exchange seems to have been generated by several formal rules working in concert:
MUSE: How like a thing, how like a paragon
YOUTH: how like a think, how like an epicure
MUSE: how like a stink, how like a pedicure
YOUTH: how like bacteria that thrives in the footbath
MUSE: how like a strand of flesh-eating staph13
The most salient formula in play is anaphora. There is also a metrical pattern—iambic pentameter, with trochaic substitutions—at least in the first three lines, and a rhyme pattern. In fact, if we insert line breaks at the caesura, we would have the ABABACC of ottava rima. The transformations of the rhyme words follow some other formulae: the Gs become Ks, then a letter gets added (or added and transposed) to the beginning of the word. Word choice is a function of these formula: “Epicure,” for example, seems to have been chosen to mediate between the sounds of “paragon” and “pedicure,” an aural stepping stone rather than a meaningful contribution to the conversation. It is as if there is a multi-step algorithm guiding the verse.
This is true, to some extent, whenever a poet uses a received form. Counted meter and set rhyme condition and sometimes inspire what can be said next, so that the form takes on a kind of agency in the composition process. Susan Stewart calls this phenomenon “lyric possession” and gives examples of poems by Keats, Hardy, and Bishop that are “haunted” by specific formal antecedents.14 This is, I think, what McSweeney means by “posthumous poetics”: text dictated by linguistic rules rather than human agency. Her choice of the word “poetics” rather than “speech” reminds us that verse is always a kind of game—or even a kind of automation.
Near the end of the play, the would-be captain of the ship, Saint-Exupery, commands would-be hijacker Muse to “Sing, muse.”15 With this command—a pun on the way his name looks, not sounds—Saint-Exupery enacts the convention at the beginning of every epic. When Muse demurs, Saint-Exupery settles for a compromise: “Then prattle.” Muse launches in on another sprawling, haphazard series of puns, figurative substitutions, and mistranslations, starting with Wallace Stevens’s “The Emperor of Ice-Cream” and ending with a different emperor, Napoleon. “Prattle,” then, is the play’s name for this kind of linguistic play. To prattle is not to narrate a coherent story on an epic scale, but to follow patterns of sight and sound, and take literally figures of speech and etymological roots. It is to take dictation from the material properties of language rather than a god.
This sounds a bit like the procedural poetry that was popular among the Language poets in the US and OuLiPo in France. But this automatomorphism is faux proceduralism: the passages sound automated, but they do not actually adhere to a formula or the output of a machine-learning algorithm. It is McSweeney who writes the dead youth’s lines. Language writers like Bernstein and MacLow would call this “cheating”: presenting text as the result of a chance operation when it has actually been doctored by the writer.16
And yet, McSweeney downplays her own authorship, going as far as to cede the title of “author” to Lacks. Lacks asserts herself first as the play’s mother (“My body is gravity / rocking this ship / in the belly of the play / the play rides inside of me”), then as playwright (“From now on I am the author of this play”), then eventually as programmer (“I write code. / I distribute copies. I propagate my line.”).17 This is part of the analogy McSweeney constructs between Lacks’s vulnerability as a poor, Black woman in 1950s America and the precarity we all experience in the age of the internet. Our words can get away from us, surviving—sometimes in garbled form—after our deaths, just like the cell line Lacks “authored.” For this reason, Dead Youth could be read as an instance of the trend Natalia Cecire has identified in American experimental writing, whereby white writers “appropriat[e] a minority position” by virtue of their “oppositionality.”18 But Dead Youth does not erase Lacks or her suffering. In fact, McSweeney tries to honor her, portraying her with agency in life and death. Ultimately, McSweeney’s treatment of Lacks is ambivalent, at once instrumentalization and tribute, as it confers on her a new kind of immortality, as a figure for poetry itself .
This is part of the cluster Poetic Voice and Materiality. Read the other posts here.
- Melanie Mitchell, “The ELIZA Effect,” interview by Roman Mars, 99% Invisible, 10 December 2019, https://99percentinvisible.org/episode/the-eliza-effect/transcript/.
- Jonathan Culler, Theory of the Lyric (Cambridge: Harvard University Press, 2015), 35.
- Joyelle McSweeney, Dead Youth (New York: Litmus Press, 2014), 16.
- McSweeney, “Justice Absconditus, or Why I Write Verse Drama,” Fanzine, 8 October, 2017, http://thefanzine.com/justice-absconditus-or-why-i-write-verse-plays/.
- Paul de Man, “Hypogram and Inscription: Michael Riffaterre’s Poetics of Reading,” Diacritics 11, no. 4 (1981): 34.
- Paul de Man, “Autobiography as De-Facement,” in The Rhetoric of Romanticism (New York, Columbia, 1984), 81.
- Ibid., 76.
- McSweeney, Dead Youth, 17.
- Ibid., 18.
- Ibid., 31.
- Ibid., 32.
- Ibid., 32.
- Ibid., 71.
- Susan Stewart, Poetry and the Fate of the Senses (University of Chicago, 2002): 124, 132.
- McSweeney, Dead Youth, 70.
- Bernstein and Andrews, The L=A=N=G=U=A=G=E Book (Carbondale: Southern Illinois University Press, 1984): 26.
- McSweeney, Dead Youth, 13, 14, 86.
- Natalia Cecire, Experimental (Baltimore: Johns Hopkins Press, 2009), 35.