By Evan Thomas

What does it sound like to sound educated yet know nothing? In a 17^th-century comedy by Molière, Le Bourgeois Gentilhomme (“The Middle-Class Aristocrat”), a rich cloth merchant tries to imitate aristocratic education and speech. He takes philosophy classes and learns that his normal expressions “require a little lengthening” – he must learn how to stretch heartfelt statements (“your lovely eyes make me die of love”) into aristocratic contortions (“Of love to die make me, beautiful marchioness, your beautiful eyes”; “Your lovely eyes, of love make me, beautiful marchioness, die”; “Die, your lovely eyes, beautiful marchioness, of love make me”; “Me make your lovely eyes die, beautiful marchioness, of love”). The joke is on him, as his rhetoric tutor cruelly exploits his easy admiration for excessive, voluminous, amplitudinous, prolix, verbose, copious speech.

The example of the Bourgeois Gentilhomme is echoed in a new development in AI. Recently, OpenAI released ChatGPT, a large-language model (“LLM”) AI that appears to have tremendous facility at composing passable long-form texts. As an educator in higher ed, I don’t think that writing pedagogies are remotely ready yet for the instructional challenges posed by this technology. The main concerns that academics have had about AI and collegiate writing have to do with academic integrity. These are important concerns and addressing them will probably have massive relevance in the years to come.

However, not all academics are especially concerned by the threat posed by AI language models. First, some academics express confidence that their domain-specific knowledge is too inscrutable for a machine to understand. Second, others suggest that the strength of their bonds with their students would make it impossible for their students to make an unnoticed switch to a different voice. Whether the first or second case is true, whether some content or character is indelible, there are finer, more constructive applications of LLMs to writing in higher ed.

For my part, I would approach AI in much the same way that I approach the rhetorical concept of “copiousness.” Copia, which can be translated as “plentifulness,” can be artificially trained. Classical rhetoric teaches figures of amplification, as in Erasmus’ 206-chapter guidebook to amplification De Copia. But without a worthy idea or a lick of sense, an education in copiousness can leave a student sounding like the Bourgeois Gentilhomme. Likewise, the internet is filling up with examples of ChatGPT saying a lot while saying little. It is easy to draw ChatGPT into nonsense, absurdities, circular arguments, and so on. ChatGPT, like the Bourgeois Gentilhomme, needs to learn about the rhetorical canons of style and decorum.

What I’m proposing is to regain a clarity of purpose about what words are for. Words have been invented by humans for humans. Humans need words because we need each other. Composition class is not valuable because we have a critical shortage of words on pages. Quite to the contrary, composition class is valuable because students of higher ed must be prepared to enter into positions in society where they will be charged with observing, analyzing, developing, debating, revising, and communicating ideas – and writing is their best tool for all of the above.

If anything, human effectiveness in rhetoric can be found in the opposite of copiousness, namely brevity. Researchers at MIT have recently attempted “A Minimal Turing Test.” The Turing Test is a popular litmus test for language-based artificial intelligences, where a human and an AI are each challenged to persuade a judge of their own humanity. John McCoy and Tomer Ullman, at the MIT Department of Brain and Cognitive Sciences, devised a variation on the Turing Test where both sides are only allowed one word to make the case for their own humanity. In other words, if you had to persuade another human of your own humanity with a single word, what word would be most effective? What McCoy and Ullman found in a study of 936 respondents was that although affecting words like “love” or “compassion” worked fairly well, the most effective response for humans to identify other humans was “poop.” Humans know humor when they see it: “poop” is funny, and there’s not a chance that a robot would submit “poop” to a Turing Test.

What this example demonstrates is that persuasion depends on awareness of audience, opportunity, and a playful understanding of fit. Furthermore, I would suggest that the art and craft of persuasion will depend more on these concerns as LLMs continue to advance. Ancient rhetors believed that no composition was absolutely original – we all rely on others for the words that shape our speech, as well as the commonplaces and metaphors by which we think, and so on. What defines a communicative act as authentic may have less to do with the selection of any particular set of words and more to do with its relevance to an audience or its responsiveness to a given situation.

In practical terms, I’m proposing to bring ChatGPT into the classroom, to invite students to generate meaningless text with the LLM, and then to invite students to refine the dross with reference to rhetorical principles. To my mind there is no fundamental difference about who produces walls of meaningless text, whether it’s an LLM or a mad-lib. In fact, the Dadaists proposed writing poetry by drawing words randomly out of a hat and many early AI poetry programs did the same. I would ask students to mass-generate a huge sum of words – maybe 10,000 words for the first assignment – and then evaluate them. Out of 10,000 words, are any of them relevant? Are any more fit for this audience or that? Which ones are in an admirable style? After several rounds of selection and analysis, I would invite students to begin revising the LLM’s text to better fit their own ideas. I think this might deepen students’ appreciation for the thinking skills of openness, responsiveness, and play that are available in the best composition classrooms.

These skills are of a specific type: flexible and human, not mechanical. The term “rhetoric” was coined in an exchange as part of Socrates’ call for excellence in communication, not cheap tricks. Socrates famously confronted a famous rhetor and legendary bullshitter named Gorgias about the nature of rhetoric. Gorgias gained his fame as a kind of political fixer: someone who would provide his clients with the right words to say, based on a handful of gimmicks and tricks that Gorgias would re-use from situation to situation. Socrates argued in contrast with this a person worthy of persuading others would treat writing not as a techne (a knack, to be crafted for definite, deliberate effects) but as an arete (an excellence, a natural avenue for human development). Over the course of the millennia since Socrates, many of the techne of ancient Athens have become industrialized and human efforts have moved to more refined applications: weaving, for example, has now been replaced with industrial looms for at large and handicraft for personal and social expression at home. If we now treated ChatGPT as another industrial machine like a mechanized loom, we might look for ways that students could develop their own human excellence in writing at large and at home.