Scary Tools for the Toolbox: Observations from the 2024 AI-Themed MNWE

The 2024 MNWE Conference provided the perfect opportunity to reflect on a full academic year’s worth of focused concern for the impact of AI on college writing and for strategies around coping with it. I presented on the utility of the AI toolbox and its drawbacks at the 2024 Lake Superior Summit on the Teaching of Writing and English as a Second Language in February—a conference that has a somewhat different vibe.

First, my general attitude towards AI is skeptical and intensely critical. However, my impulse with this, as with any new technology, is to assess its potential as a useful tool and to try to determine what it is good at or what it is useless at. In its current form ChatGPT 3.5 and 4.0, since those are the versions I have experimented with, have some very limited potential utility, some moderately useful functions, and a great number of time-wasting, reputation-destroying traps for the inexperienced or unwary.

The mood at MNWE was different—in some ways more engaged and curious, but in some ways far more cynical or even paranoid. Some presenters were actively researching the kinds of questions I’m raising in this essay. Others were dismissive, suspicious, or wary of the technology. Some related that they had pivoted to hand-written-only assignments, and some refused even the idea of trying the tech themselves out of an (over-) abundance of caution. Possibly because of data scraping or identity theft—the concern was very genuine even if the feared result seemed a bit nebulous.

My colleagues at the Lake Superior Summit had a much more positive or optimistic view of the technology than I was prepared for. To the point that there were several discussions where I felt it necessary to actively raise concerns about proposed approaches. One presentation went so far as to suggest the chatbot as a useful starting point for research—which it categorically is NOT.

A large language model (LLM) can produce text based on a given prompt—the more specific the prompt the more targeted the produced text. However, as multiple anecdotes and several in-depth articles have explained, an LLM is not equipped to verify or fact-check because it was never designed to evaluate its training data. If asked for source suggestions on a project, it is equally likely to produce citations on outdated, unreliable, or completely fabricated references. At the same time, its citations will look entirely plausible because of what it was designed to do well—produce polished-looking text that adheres to a specific template and prompt. The question then becomes, who is responsible for fact-checking or verifying the veracity of materials and the time required to do that? The student? The instructor? Who is held responsible for errors, mistakes, or other issues? Clearly anyone but the developers of the technology (pending litigation notwithstanding).

Given this situation, how do we instruct students in this evolving environment?

First, it is clear that we need to do more experimentation ourselves with this technology. Try the kinds of searches or prompts that a student in our courses might try.[1] Do some adjustments to the prompt. Analyze the results considering a few factors: How much tweaking was needed before the chatbot could approximate a useful response? How close was the response to what could be an acceptable or passable assignment submission? How obvious was it that the prose was bot-generated? What kinds of telltales remained? How many were acceptable versus unacceptable?

Beyond giving the toolset a try, what implications do your results have for your assignments? Does the assignment invite genuine, personal reflection? Is the student asked to engage with the subject and share their own perspective or relate their direct experience on some level? If a chatbot can produce an acceptable response, perhaps consider adjusting the assignment.

What advice do you take away for your students? Might they use a bot to brainstorm around a particular topic? To generate talking points? To switch their own style from one register to another (code switching)? To see what the standard format for a particular kind of document might be? Determining potential use cases and recommending some of them can demystify LLM-based AI for your students and demonstrate your familiarity and confidence.

Discussion of the tech and its potential with my own students and with a recently graduated family member yielded some fascinating results. First, many students are still unaware of these tools—remember that not all of our students have equal access to or experience of technology. Second, many students value the idea of time-saving toolsets and are fully ready to use them if they can be relied upon. That said, they generally want to turn in quality work, and if the tools are not reliable (or if we can adequately articulate where they are or are not to be relied upon). They are no more eager to waste their time, money, and energy than we are.

The bottom line of these somewhat scattered observations is this: tools as such are morally neutral and innovation is cyclical. While it is clearly important to engage with and understand these developments, there is no need for fear. What is needed is communication and trust. Tell your students what your expectations and prohibitions are, explain why, and trust that the enormous ongoing investment they are making in their education will motivate them to diligently endeavor to learn what we have to teach.