Writing Excellence in the AI Era: Fostering Academic Writing Skills With Supportive Feedback
Exciting times.
I’m a believer in the essay—an uncontroversial opinion up until a few years ago, when Large Language Models became readily available through easy-to-use prompt interfaces, bringing the essay mill to the desktop. My timeline is a bit longer than that. Essays have structured humanistic inquiry for hundreds of years. Learning to write them, students learn to think and write clearly. They model objectivity; they construct an argument. They learn the meta-skills of working with feedback: how to integrate it, but also how to offer feedback to others. They learn to apply a requested structure—page length, bibliographic formatting. These are vital job and life skills.
Anyone who teaches is having AI problems now, in this moment when large language models can write not-particularly-great papers for our classes. Such students, in Ted Chiang’s instantly iconic phrase, are missing the point of what we do: “Using ChatGPT to complete assignments is like bringing a forklift into the weight room; you will never improve your cognitive fitness that way.” (1) This AI moment offers us an opportunity to clarify what we assign essays for. Indeed, AI offers many opportunities. To continue with Chiang’s metaphor, although we shouldn’t be using AI to write essays for us, we should think about how they might train us. Chiang writes, I believe correctly, that AI will never make art. But it’s going to be very helpful teaching the technical basics of artmaking, an endlessly adaptive tutor with unlimited patience. Here LLMs’ tendency to reproduce the average of a situation is an advantage: we learn the basics—the averaged baseline of our chosen technique—so that we can excel within them.
Meaningful writing instruction is iterative: achieved through cycles of feedback, the integration of that feedback, and revision. Academics are distinguished by how long they’ve spent in these loops—the loop for an undergraduate essay is shorter than that for a doctoral thesis, but the process is similar. Someone looks at your writing, someone points out some issues; you work on those issues, send the writing to someone else, and the process continues.
Theisfy’s AI model offers endless patience, and is always available. You can give it a version of your paper, receive feedback, and run it through again as many times as you feel like. This work complements our existing processes of offering writing feedback. Students can ask Thesify for advice in advance of going to see their professors for writing feedback. It can clear the low-hanging fruit—issues with use of evidence, with thesis organisation—leaving them free to talk about the really interesting stuff: the ideas these papers are helping clarify and shape.
A few things about Thesify really startle me:
Advice on working with evidence. A lot of writing feedback is noting what to evidence further—saying “this is interesting, I’d like to hear more about this.” The version of this program I’ve been working with can point to exact moments in the text you might think about evidencing more. Developing a feel for what to evidence and how to do so is an important part of developing as a writer—but it’s also something even professional academics recommend to other professional academics. Clarifying evidence isn’t something we solve as writers; it’s something we know to be ever-mindful of, and to ask others to help us with.
Advice on tying papers together: The program can advise you on whether or not evidence connects back to the main thesis. Sustaining a focused argument through a mass of evidence is one of the most difficult things to master in academic writing—again, something you never exactly solve, but instead something you endlessly invite feedback to correct and improve. Thesify can offer that sort of feedback.
Specificity of feedback: Thesify can also point directly to the parts of a paper that need work; students are taken directly from advice to where in the paper they may apply it. This type of guidance in feedback is extraordinarily time-consuming to give as a reader, and to input as a writer; the AI’s endless patience helps here.
Tone of feedback: Tone is one of the hardest things to get right in offering feedback. I appreciate Thesify’s positive tone: that it identifies things writers are doing correctly along with things they need to improve. We’re vulnerable when we write; a patient and helpful tone can help us see writing as on ongoing technique-based process of self-discover, rather than a deficiency to be corrected or a learning gap to fill.
I believe that someone using this program will be able, through patience and hard work, to use it to improve their writing and thinking. This isn’t a turnkey solution or an online essay mill—indeed, this software can’t write a paper for anyone. To return to our gym metaphor, Thesify is like a physiotherapist, pointing to subtle issues and offering exercises to correct them. It’s there to help the grind of self-improvement become the joy of self-expression through mastering the still-vital skill of writing the essay.
(1) Ted Chiang, “Why A.I. Isn’t Going to Make Art,” The New Yorker, 31 Aug 2024.