Preterition

What’s frustrated me about my discipline’s response to the AI panic has been the steadfast refusal to define its terms and to subject it to proper analysis: slice it up into its parts and see what it does. Humanists seem to have read Bender et al.’s stochastic parrots article with its grab-bag of complaints and taken its form as the proper model of AI critique. There is scant differentiation or specification: compositionists take “AI” as a rhetorical hyperobject and try to come at it via the usual routes of hermeneutic interpretation. Many compositionists seem to imagine AI use as sitting in front of a chat prompt and having the model generate prose on a topic of their choice: the essay-writing model. With rare exceptions—I find John Gallaher’s work on LLMs admirable—the field consigns the bot to invention without defining invention’s terms and then complains that it scores poorly on what they see as their home turf.

The Arxiv-published experiments aimed at humanistic questions are largely uninformed, because they were written by engineers. Critiquing what an engineer built because it doesn’t carry non-engineer values is like critiquing a poem because it doesn’t work as a calculator. Typing prose into a chatbot window does not exhaust the possibilities of technological exchange.

Composition’s stance ignores the enormous volume of non-invention-related, non-chat applications of what ML technologies might do for the other rhetorical canons of delivery, memory, organization, and style. The field tries to anthropomorphize the technology into the shape of its humanist values and often misses the point: recent composition-oriented scholarship in The Elsevier Journal that Shall Not Be Named surveys the headlines of Arxiv articles for what they say models can’t do and hilariously ignores the phenomenon the articles were investigating. There have been some salvos directed at machine style and what probabilistic models do with the shape of prose, but I have yet to see disciplinary critiques that go beyond the hermeneutic stance of imagined rhetorical unreliability or that don’t rely on variables external to the question. We can’t imagine machine experiments with meaning, so we dismiss the possibility of constructing experiments.

Stop complaining about meaning and construct the experiments. Meaning belongs to humans: isolate the variable.