Tristan B Taylor
University of Saskatchewan
This two-part activity asks students to interrogate assumptions about poetry, broadly construed, based on their preconceived notions and prior knowledge of poetry. In the first part of the activity and assessment, students are presented with a poem composed by generative AI and asked to define how the text is a poem. The instructor should guide the students in a discussion about what constitutes poetry. In the first part of the activity, students should not be made aware that the poem under discussion is composed by generative AI. The instructor should encourage students to close read the text. In the second part of the activity at the end of the term, students are presented with the same poem and asked to discuss the poetic mechanisms employed by the generative AI. The instructor poses the following questions: How does generative AI conceptualize poetry? Can generative AI compose poetry for an audience? Is generative AI capable of producing poetry? Through this activity, students gain a greater critical understanding of poetry, audience, and the rhetoric and mechanics of poetry, as well as AI literacy. I have taught this exercise successfully in two separate courses, and students have responded positively on both occasions.
Learning Goals
Original Assignment Context: first-year undergraduate English course on poetry and composition
Materials Needed: A poem produced by an LLM on this or a similar prompt: “Write me a poem.” Additionally, slides or handouts should be used so students can access the poem.
Time Frame: Approximately two classes. Part One should be completed near the beginning of the term; Part Two should be completed in a review session near the end of the term.
Overview
I have taught first-year undergraduate English courses for five years, and students’ reactions to reading poetry are nearly universally negative. When asked why, they often refer to form, diction, and metaphors as features of consternation as opposed to delight. Their perception of poetry is often circumscribed by the notion that it is subjective and can mean anything. Thus, when students are asked to provide close readings of poetry, they often stretch beyond what is reasonable. Umberto Eco posits that “the world of literature is a universe in which it is possible to establish whether a reader has a sense of reality or is the victim of his own hallucinations,” adding that “there is a dangerous critical heresy, typical of our time, according to which we can do anything we like with a work of literature, reading into it whatever our most uncontrolled impulses dictate to us” (Eco 7). He concludes that “literary works encourage freedom of interpretation, because they offer us a discourse that has many layers of reading” (4). Students’ approaches to poetry are evidence of this type of hallucination that Eco discusses. This is not the fault of the student, of course, but a pedagogical misunderstanding of poetry, teaching poetry from a singularly interpretive perspective in lieu of a mechanical perspective.
Alongside the poems in the course, students are required to read Poetic Designs by Stephen Adams, which begins with a metaphor of hockey. Adams hypothesizes that any viewer can watch and enjoy a game of hockey, but the true pleasure of watching a sport like hockey comes from understanding the rules. Like with poetry, only after an understanding of the mechanics is established can students effectively engage with the material. Herein lies the rub: when students are presented with a poem and asked to close read it, they have no discourse other than interpretation to inform their understanding, so they cannot comment on the poetic mechanics deliberately employed by the poet to make meaning. In other words, students can read the poem and maybe even enjoy it, but they will not understand it nor understand why they enjoyed it. Further, while a student may be able to interpret or understand one poem, they may not, without comprehension of the mechanics involved, go on to understand other poetry.
The purpose of teaching poetry through the lens of its mechanics is to initially divorce interpretation from function. This is why an LLM like ChatGPT is an effective pedagogical tool. An Aristotelian view of literature demands an understanding that there is an inextricable link between author, audience, and message. This triad leads to an interesting case study for LLMs, where the author lacks the agency of a human voice. Instead of deliberate rhetorical choices, LLMs rely on probability for composition. Thus, poetry composed by LLMs lacks a key component: the author. If we can strip away the author, we can strip away any notion of intention, consequently limiting an emphasis on interpretation. Reading poetry composed by LLMs is therefore an effective approach to understanding poetic mechanics, as the meaning behind the poetry is by nature vacuous. While there is a danger to employing a tool like ChatGPT or other LLM in a classroom environment, suggesting to students its ability to compose text that appears intentional, the purpose of an exercise like this demonstrates that the meaning behind text generated by LLMs lacks the rhetorical aspect of authorship and thus meaning.
While LLMs like ChatGPT have become ubiquitous in higher education, much to the concern of their English instructors, they still prove to be useful pedagogical tools to provide a platform for evaluating texts, especially given the probabilistic generation of text. Because they produce text not intentionally but through probability, we can, in many ways, divorce mechanics from interpretation. Furthermore, we can begin to ask questions about the assumptions of LLMs and how they are trained. Thus, LLMs like ChatGPT can interrogate assumptions about poetry in ways that Shakespeare, Dickenson, or Agard cannot.
Adams begins Poetic Designs with the following observation:
Some teachers evidently think it possible to evoke this experience on the spot, bypassing the snarls of prosodic terminology and understanding and instead attempt to convey a personal enthusiasm for a poem through a mixture of oohs and ahs, nods and winks, and personal charisma. The student may be temporarily persuaded into some kind of experience of the poem, but is left with little or nothing to help in understanding the next poem, for [they remain] ignorant of how effects have been achieved (Adams 1).
One way of introducing poetic mechanics is through AI-generated poetry, which lacks the rhetoric of actual poetry. Students’ first instinct when exposed to an AI-generated poem will be to evaluate it and provide some moral judgment on it: either they like it, or they do not. Infrequently, students will be able to articulate why. This should be noted by the instructor and brought up again in the discussion.
The instructor needs access to a LLM like ChatGPT to create the sample poem, as well as a means of distributing the poem to students. A recommended reading to go alongside this exercise and lesson plan is Chapter One of Poetic Designs by Stephen Adams, which provides a basic vocabulary that new students of poetry would likely already be familiar with.
The primary objective of Part One is to have students identify common features of poetry. Students should be able to identify features like rhyme, stanzas, line length, and metaphors. The secondary objective of Part One is to have students connect these features to the AI-generated poem. The third objective is students should be able to connect these abstract features to real-world examples.
In Part Two, near the end of the course, students should be presented with the same AI-generated poem from Part One and identify the mechanical features employed by generative-AI. The learning objective of Part Two is to challenge the students to consider what assumptions LLM and those who develop LLM make about poetry.
Session One
Brainstorming (15 minutes)
Before beginning this activity, have students brainstorm as many features of poetry as they can in a short amount of time. Create a list that you can consult later in the class. Go over each feature or concept and define the key terms for the class.
Reading AI-generated poetry (35 minutes)
Introduce the poem and ask students to identify features in the AI-generated poem, drawing on either the previously generated list or any new features they can think of.
After reading through the poem, ask the students to provide a close reading. Ask them to consider what the poem looks like and sounds like and connect these observations to the previous developed list. Finally, ask them to interpret the poem. What does the poem mean? What is the poet trying to tell the audience? For whom is this poem written? Lead a class discussion inviting as many interpretations as you can.
Identify the poem as composed by AI and ask the students if this changes their interpretation of the poem. Ask the students why knowledge of the poet changes the interpretation of the poem, and how it challenges our assumptions of poetry and its mechanics.
Session Two
Brainstorming (15 minutes)
Near the end of the term, after students have developed a sophisticated critical vocabulary of poetry features and have had the opportunity to provide close readings of diverse types of poetry, reintroduce the poem from Session One. Have students familiarize themselves with the poem by completing a close reading of the poem and identifying the poetic mechanics of the poem. Remind the students that generative-AI composed the poem. Lead the class in a discussion where they identify the assumptions that LLMs and LLMs developers make about poetry. Ask students to consider the following question: According to LLMs like ChatGPT, what is a poem? By the end of the exercise and discussion, students should be able to articulate the formal features of poetry.
Writing AI-generated poetry (35 minutes)
After a short class discussion, ask the follow-up question: can LLMs write poetry? Divide the class into two groups where they debate the subject, with one side taking the positive position that LLMs can write poetry, and the other side taking the negative position that LLMs cannot write poetry.
By the end of the debate, students will be able to articulate the relationship between authorship and audience as well as form and meaning-making.
The early conception of this activity emerged from a conversation I had with Michael Sargent at the 2023 meeting of the European Society for Textual Scholarship, held in Canterbury, UK. We discussed the prevalence of generative AI in the classroom and how it might be used to illustrate its inadequacies in producing poetry.
Adams, Stephen J. Poetic Designs: An Introduction to Meters, Verse Forms, and Figures of Speech. (Peterborough: Broadview Press, 2003).
Eco, Umberto. “On Some Functions of Literature,” in On Literature, trans. Martin McLaughlin (London: Harcourt 2004).