How can we use AI to lose the grunt work but keep the teaching?

Like most things, I have mixed and complicated feelings about AI. On the one hand, I am pretty sure the world would be better if it had never been invented. I teach writing. Writing is hard, and many students (and people in general for that matter) would just rather not. Any tool that makes it easier for students to write less is not going to be a win for all of the amazing things that writing does for learning, development, and the human mind.

Nevertheless, I am a pragmatic person. One of the things I love about my field (Rhetoric) is the concept of phronesis or "practical wisdom." Phronesis is different from sophia in that the former requires us to be able to use knowledge to know what to do in specific situations, while the latter concerns itself with theory for its own sake and uncovering universal truths. So my affinity for phronesis demands that I accept the existence of these tools and try to figure out how to use them effectively and ethically.

So in this post, I will try to capture my current line of thinking about Generative AI, and also describe three ways that I use these tools to make our work a little easier.

What worries me about AI in teaching

Taking a pragmatic starting point does not mean that we should ignore concerns we have about AI in general and Generative AI (GenAI) specifically. Like many in academia, I have a great number of concerns about what tools like ChatGPT and its ilk are doing to our students' learning conditions and our own working conditions. GenAI is problematic in a multitude of ways. For example, in terms of our teaching:

• Writing less will hurt learning or critical thinking.
• GenAI makes it easier for students to write less and harder for instructors to recognize that they are writing less.
• GenAI can disrupt the relationships we build with our students.

Beyond these teaching-specific concerns, GenAI raises several individual, social, and community health concerns. Here are just a few:

• GenAI is relentlessly positive.
• GenAI can foster delusional thinking.
• GenAI is trained in unethical ways.
• GenAI is bad for the environment.
• GenAI data centers are bad for communities.

All of these concerns are valid, and I'm sure there are a multitude of others that I haven't captured here. Even so, I believe that teachers can look for ways to use AI tools effectively and ethically while also critiquing these tools and demanding change from the companies that run them.

We should not let AI interfere with our relationships with students

I doubt if a college teacher saying that AI should not be allowed to teach will come as much of a surprise. Like many in my profession, I have been filled with existential dread as more and more tools seem to be appearing which substitute some form of AI use for authentic student-teacher interaction.

I remember when my 8th grader came home from school and told me that his teacher told them to use SchoolAI to get feedback on a writing project. Isn't that the teacher's job? I thought.

In April of 2024, I went to the Conference on College Communication and Composition in Spokane, WA. ChatGPT had been out for a little over a year, and many of the panels were about Generative AI. One presenter helped me put GenAI in a helpful perspective for teaching and learning: Anything that disrupts our relationship with our students will be bad for their learning. So the question is, what does GenAI do for that relationship? Naturally, the answer depends on how we use the tool, and on how we view technology and tools themselves.

Some people see tools and technology as completely neutral. This is the logic behind "Guns don't kick people, people kill people." Personally, this doesn't make sense to me: Some tools are clearly designed to do specific tasks, and will not be able to do other tasks. Technological determinism offers an opposing view. Although this term is complex and understood in various ways, I find it helpful to think of technological determinism as the view that once a decision or value is baked into a technology, humans have no agency over how that technology is used.

I've always found the middle road perspective to be more useful. (Did I mention that I'm a pragmatist?) One such perspective comes from philosopher of technology Andrew Feenberg's book Transforming Technology. Feenberg sees technology as "an 'ambivalent' process of development suspended between different possibilities."¹ In other words, while it is true that technologies have specific values and perspectives built into them, human beings have a great deal of agency over how we use those technologies. The end is not predetermined.

That said, I also think that the values baked into a given technology limit the range of possible outcomes I can expect while using it. As an obvious example, a hammer is great at pounding nails, but terrible at cutting boards. Similarly, the people building Generative AI tools are embedding a set of values that make some outcomes easy to achieve and others much harder. Understanding these tools and the range of outcomes they make possible will be a crucial part of being able to function as a member of the modern world. Helping teach this to our students is becoming an ethical requirement of teaching in most disciplines.

It's our job to teach effective and ethical AI use to our students

While writing this post, I have vacillated between two worries: The first is that this position will provoke the ire of other teachers and academics. The second is that this position is patently obvious. I suppose I am writing to the people between those two positions, namely, those among us who have major concerns with AI but still feel a need to understand it and help our students understand it and use it ethically. I saw a comment on reddit/r/professors recently that claimed "There is no ethical use for AI." But such an absolutist perspective is hard for me to swallow.

I think absolutist perspectives like this one stem from two key concerns:

1. AI is built on stolen intellectual property.
2. AI is making climate change worse.

I know there is a growing body of thinking and writing dealing with these issues, but I'm not going to investigate that in this post.

I am no philosopher, but surely there must be some range of uses that are ethical, these two objections notwithstanding. If there isn't, then ethical purists will need to sell their cars and stop taking commercial flights, and (in many cases) start looking for jobs and apartments in Europe. I guess what I mean is that we are always making the best ethical decisions we can in extremely flawed situations. That's the nature of adulthood, right?

The fact is, we live in a world where these technologies are ubiquitous. Like many people, I would rather we didn't. I'd also rather I didn't have a cell phone in my pocket and that I didn't have to try to convince my kids that social media is like cigarettes, but here I am, occasionally checking reddit on my phone and looking for opportunities to promote this blog!

Of course, as a professor with plenty of autonomy and job protections, I could refuse to engage with AI and suffer very little in terms of personal consequences. But my students will not have the same privilege. When they graduate, they'll be dumped into a world that is haphazardly adopting a technology it doesn't understand. Their employers will expect them to help make sense of it. We need to do what we can to help them meet that challenge.

Three ways I use AI to make teaching easier

I want to make one thing as clear as possible: I think we should avoid, as much as possible, tools that claim to create teaching materials for us. My hedging here is not just a personality flaw. The fact is that many if not all uses of AI will involve some level of curation, selection, and presentation. Each of us needs to be clear with ourselves about how much intellectual work we want to offload to a large language model. If the model is doing something that saves us time, then it is taking actions on our teaching content. My own preference is to try to minimize the amount of "thinking" that the model does while maximizing the amount of energy and time it saves me. Here are a few uses I have found that save tons of time while not intruding too much on the intellectual choices I make as a teacher:

• Slide preparation: I will mark up a chapter or a paper that we are reading for a class and leave notes in the margins about what I want to include in the slide deck. I then feed a scan of the marked-up reading into Claude and give it specific instructions on how to construct the slides. Claude creates a Markdown file, which I then edit, add speaker's notes to, and convert to PowerPoint and Word. I use the PowerPoint file in class and share the Word version later on Canvas. This saves me hours of fiddling with PowerPoint, and keeps my slide content in a more accessible format.
• Finding examples: I will provide a description of the learning goals I have for a given class and then ask Claude to find examples of writing that help illustrate related concepts. Claude can search the web, find the examples, and make sure they will help illustrate the concepts I want to teach on a given day. I might use one example for a demonstration and then put another into a Markdown file with guiding questions drawn from the course materials that help students actively apply the concepts we are working on in that unit. This saves hours of fruitless internet searches, finding examples that almost work but don't quite illustrate the concept that I'm trying to teach on a given day.
• Topic extraction: This is the use that led to the creation of Sheetbend. I had several years' worth of lecture topics all bound up in Word files. I built a system that scans those files and tries to break them into individual topics. It stores those topics as individual Markdown files and provides an interface for organizing them into different arrangements and output formats. This literally saves me hours of digging through files on my desktop, trying to find the one class where I presented a specific topic in a certain way.

Clearly, each of these approaches offloads a certain amount of decision making to an LLM, and not everyone will be comfortable with that. For me these kinds of uses are a reasonable trade-off: I save time and energy, create more consistent teaching materials, and retain complete control over the pedagogical decisions and authorial voice that inform those materials.

Footnotes

1. Feenberg, A. (2002). Transforming technology: A critical theory revisited. Oxford University Press. ↩