DOI: 10.37514/DBH-J.2024.12.1.08 Double Helix, Vol 12 (2024)
AI and Writing. Sidney I. Dobrin. Broadview Press, 2023. 144 pages.
Adam Katz Quinnipiac University
I like to think that most of the assignments I give my students are AI-proof, meaning that students won’t try to complete them using AI, or doing so will be obvious and a failure to address them. But there are still some assignments—for example, small ones following up on some questions raised in class—that are a bit “leakier.” If I don’t detect their leakiness in advance, it becomes clear as soon as I start reading student responses that look suspiciously fake in the way we, as writing instructors, have come to recognize AI-generated work. Such situations raise all kinds of questions about our relationship with our students, but what I want to note here is that I’m not sure whether I give a more or less leaky assignment to see what the AIs might be able to do with it or what the students might be able to do with the AI. And I am, of course, aware that the AI-proof assignments I’m designing will very likely no longer be AI-proof one or two or three new large language models down the road, in which case defending a space for thinking independent of AI is a rearguard action. My thinking, then, is more an inquiry into what it would mean to automate one or another of the skills or practices I associate with literacy, productively focusing attention on how that skill or practice is composed, and therefore how it might be iterated and transferred to other reading and writing situations, with the endpoint of rendering the skill or practice irrelevant because it is transferable to a machine. We must somehow, then, learn to think within this paradox that what enables us to study human practice with increasing resolution is the very thing leading to the disappearance of that practice—and this might simply be a compressed and accelerated continuation of the history of human-technology relations. Some sense of this paradox is what I find missing in Sidney Dobrin’s very useful— maybe too useful—AI and Writing. Dobrin’s approach to AI in this short book, designed for classroom use, is comprehensive, pragmatic, theoretically and ethically informed, businesslike and optimistic. He locates AI in the history of writing, addressing and contextualizing questions AI has raised regarding academic integrity, intellectual property, and bias, as he takes advantage of the uneasiness around AI to demystify these categories. He provides suggestions regarding assignment design and ways students can be introduced to various uses of AI. Each chapter begins with a list of “Learning Objectives,” with some things to think about before reading the chapter, and ends with “Questions for Discussion,” often including “provocative” ones that could serve to spark a class discussion (many of which I could imagine students using an AI to answer). Near the end, he takes readers, presumably teachers and students together, through a discussion of the economic transformations expected from generative AI (GenAI), along with the implications for career opportunities for students. Dobrin is taking the stance—and I’m not sure what other stance would be justifiable—that we should simply accept the existence of AI, while acknowledging the understandable qualms about it, and get on with the business of helping our students learn how to make the most of it: what skills will equip them to master AI and what mastering AI
Double Helix, Vol 12 (2024)
will in turn enable them to do. So, for example, in a discussion of Skills that Benefit from GenAI, Dobrin mentions “problem solving and critical thinking”:
Strong problem-solving and critical thinking skills are necessary in order to tackle the complex challenges of any workplace. You should be able to analyze problems, devise creative solutions, and evaluate the performance of competing models. GenAI can be useful here too, as it’s an ideal tool for testing hypotheses and running simulations. When prompted appropriately, many GenAI programs are also quite effective in offering innovative solutions to problems or sober and well-considered pro-and-con lists. (p. 99)
The boundary separating what the human is doing and what the GenAI program is doing is rather blurry here. Other than prompting appropriately, the problem-solving and critical thinking seem to be done by the GenAI. Presumably, the user will also assess the results provided by the GenAI—out of however many simulations generated, one will be “better” than the others, meaning that we have a new “skill set” here that may not be exactly what we are accustomed to calling “problem-solving and critical thinking”: prompting and selecting from among the results of the prompt and then iterating this practice over and over again. Insofar as we come to think in potential “prompts,” the GenAI is no longer a simple “tool” that we are standing outside of and “using.” A precursor to this cognitive scenario (if “cognitive” is the right word) might be the tendency, developed as Google became the universally used interface with the internet, to think in search terms, which must be further refined with each new set of results. The system is using us as much as we are using it. The implication is that the emergent human “vocation” is to train AIs, as in the “supervised training” necessary for a computer program to, for example, consistently and accurately identify images of cats. Supervised training is one of the more notorious aspects of AI use, as Dobrin points out, quoting Sasha Luccioni:
Essentially, once a model has been trained on a large quantity of unlabeled data . . . humans are then asked to interact with the model, coming up with prompts . . ., and provide their own answers or evaluate answers provided by the model. This data is used to continue training the model, which is then again tested by humans, ad nauseum, until the model is deemed good enough to be released into the world . . . . But that success has a dirty secret behind it: To keep the costs of AI low, the people providing this “human feedback” are underpaid, overexploited workers. (pp. 112–113)
The question, then, is whether this essential labor, carried out by underpaid and overexploited workers can be “provocative,” as the most interesting and rewarding activity ever; that is, if it’s really essential, then it should become completely integrated into all economic, political and cultural institutions and at various levels of complexity, difficulty and urgency: anyone can tell cat from non-cat, but very few people will be able to judge whether, say, a weather model is off enough so that major climactic events might be missed, or economic projections on which billions of dollars and hundreds of thousands of individuals’ economic futures might depend is over-weighing one patch of data—or, more precisely, what sort of search should be undertaken, or prompt designed, to answer such questions. So, those
Double Helix, Vol 12 (2024)
capable of machine training at the higher levels will be much in demand, and such careers lucrative and, as intellectually demanding and socially important, very satisfying. We would all “always already” be interacting with databases, and part of that interaction would be examining and revising their composition. Whatever one might think of such a future, it’s hard to see how our present teaching tools—our Learning Outcomes and Rubrics—are adequate to educating the very different kinds of human who will populate it. But we will— and should—continue to see books like Dobrin’s, and eventually they will prompt the AIs in ways that elicit the necessary information on how we should best habituate ourselves to their omnipresence.