Over on the ACOUP blog, University of North Carolina Historian Bret Devereaux has produced a useful critique of ChatGPT as a tool for writing college essays. In it, he first presents his definition of an essay, the one that guides his classroom assignments:
Writing an essay thus involves a number of steps, of which communication is merely the last. Ideally, the essay writer has first observed their subject, then drawn some sort of analytical conclusion about that subject, then organized their evidence in a way that expresses the logical connections between various pieces of evidence, before finally communicating that to a reader in a way that is clear and persuasive.
…
ChatGPT is entirely incapable of the first two steps (though it may appear to do either of them) and incompetent at the third; it’s capabilities are entirely on the last step (and even there generally inferior to a well-trained human writer at present).
The reason that some people are saying that ChatGPT is a threat to the college essay is that they misunderstand the purpose.
[There is] a consistent problem in how we teach students, which is that we rarely explain our pedagogy (our ‘teaching strategy’) to the students. That tends to leave many assignments feeling arbitrary even when teachers have in fact put a great deal of thought into why they are assigning what they are and what skills they are supposed to train.
…
[The] most important thing I am trying to train is not the form of the essay nor its content, but the basic skills of having a thought and putting it in a box that we outlined earlier…. What we are practicing then is how to have good thoughts, put them in good boxes and then effectively hand that box to someone else. … at no point in this process do I actually want the essays.
The reason I’m writing this blog…well…essay, is that Devereaux’s approach is similar to how our Senior Project worked back when I was teaching Management Information Systems. For their final class in the MIS major, students were broken into teams and given a Request For Proposal for some sort of computer system. They had to be small problems, because they had to be completed and documented within the eight week quarter. So, it might be an appointment scheduling system, or an inventory management system or the like. Students would first design the system (and turn in a response to the RFP), then build (and document) it, and then, finally, deliver the completed project, complete with user manual and budget documentation. It was great fun, for me, at least, but I had to keep pushing back against feature creep on the part of the students.
They were always going out and finding plug-in code that would produce snazzy calendars or pretty inventory graphics, even if there was no requirement for them in the RFP. I had to keep telling them that I didn’t want an inventory system, and I certainly didn’t want three different inventory systems by three different teams. What I wanted was for them to have the experience of team development of a project, to a schedule and with certain deliverables. From all reports, ChatGPT is not capable of generating workable code, and it certainly isn’t capable of producing a response to an RFP, a user manual, or a detailed budget.