“ChatGPT doesn’t go away if we ignore it,” said Kristina Reardon, lecturer in English and director of Amherst’s Intensive Writing Program. Indeed, the reason she was speaking in Frost Library on Feb. 20 was to help faculty, staff and students grapple with the advent of the language-based OpenAI program, which has attracted more than 100 million users since November 2022. ChatGPT can generate grammatically correct and confident-sounding passages of text on most topics—which means it has potentially radical implications for the future of schoolwork and written communication in general.
So the College’s Academic Technology Services and Center for Teaching and Learning convened a panel to discuss it. Reardon was one of the panelists; the others were Associate Professor of English Chris Grobe and Professor of Computer Science Lee Spector, with facilitators Jaya Kannan, director of technology for curriculum and research, and Riley Caldwell-O’Keefe, director of the Center for Teaching and Learning. Here are a few questions they covered.
“We don’t actually know exactly how ChatGPT works, because it’s not completely public,” said Spector. “But we do know the basic idea behind it and the basic way the algorithm works.” It is roughly analogous to a smartphone’s auto-complete function, able to predict, based on the huge amount of data it’s consumed, the most likely or expected next word to come. The writing it produces is coherent but not necessarily factually accurate, and the software can’t evaluate meaning. However, said Spector, “I think we can expect it to be doing a lot more very rapidly. A lot of our thinking about how we use this, how we should use it, what it has to do with education: We shouldn’t get too fixated on the current state of the art, because, although it has glaring gaps—including [having] no idea what it’s saying or what truth is—I don’t think that’s permanent.”
“What ChatGPT seems particularly good at is producing clean sentences with minimal grammatical errors in a standard type of English,” said Reardon. “If I think about the goals for teaching writing, clean and fluent English has only been one piece of the puzzle in terms of writing instruction and students’ and writers’ investments in both the product and process of writing itself. ChatGPT makes me think about why I’m assigning certain types of writing, what my goals are for student learning and how best to facilitate those goals through assignments and classroom discussion.” Yes, there’s now a machine that can produce grammatically clean English. So Reardon thinks about what she can push her students to do that ChatGPT can’t do.
Grobe said he had his students try to get ChatGPT to write a first draft. “It was a way of measuring their knowledge,” he said. “I think their first reaction was they were totally overawed by how confident it sounded and how good the sentences felt, but when they sat with it long enough they realized that they knew things that contradicted what was said.” He described this “beneficial effect of having something to push off against, something that by contrast reveals to you what you know and what you are feeling moved to argue. In that sense, it doesn’t really matter whether ChatGPT is unoriginal or original; it’s not what you know and what you mean to say, and therefore it becomes useful as a starting place for writing.”
Illustration by Ard Su