ChatGPT for Teachers: The Risk of Off-Loading Pedagogical Reasoning

Creative Commons photo of a Chicago hot-dog vendor (2003) taken from https://en.wikipedia.org/wiki/Chicago-style_hot_dog

I have concerns about the accelerating roll-out of generative AI systems in education. Many of these concerns relate to uncritical student use of AI tools, but I’m also skeptical about the implications of these tools for teachers.

Open AI’s Release of ChatGPT for Teachers

This week, Open AI released ChatGPT for Teachers. According to Open AI, ChatGPT for Teachers is built upon ChatGPT 5.1 (the version available publicly through paid subscription), but specifically with teachers and school leaders in mind. It’s designed to work seamlessly with tools and software that teachers already use, facilitate collaboration with colleagues, and provide examples from other teachers. Supposedly, it’s built with teacher-specific tasks in mind, and is data privacy compliant with US FERPA rules for student data. I must confess upfront that I have not used it. It’s currently only available to verified K-12 teachers in the US (Open AI has rolled it out free for US-based K-12 teachers through June 2027). I will note that my paid subscription to ChatGPT has given me access to version 5.1 since it was released on Nov. 12 (and to version 5.0 before that). I will also note that I recently spent some time playing with Khanmigo’s AI teaching assistant for teachers, which is built upon the 4.0 version of the ChatGPT LLM system. 

Cognitive Off-Loading

My primary concern about AI tools for teachers relates to the concept of cognitive off-loading.  Cognitive off-loading is not a new concept. It refers to any time that humans distribute cognitive tasks (memory, imagination, reasoning, etc.) into the environment so as to free up mental capacity for other cognitive work. Everytime we travel as a family, I will say at some point to my wife some version of… “help me to remember that I just put our house keys in this pocket of this suitcase.” Of course, that’s a faulty approach to cognitive off-loading, because often, by the end of our travels, we’ve both forgotten where I put those keys. As another example, I currently have an unsent email in my email “scheduled” box, set to send to my daughter in May, reminding her (and me) of the summer camp supplies that we stored in a box in the attic at her grandparents’ house at the end of last summer. Let me provide one more example that will be familiar to many. When my wife rattles off a list of things she wants me to buy at the supermarket, rather than trying to keep that list straight in my memory, I type the list into the Notes app on my phone. 

In each of these instances, rather than trying to mentally hold the information myself, which requires significant cognitive effort, I off-load the information into the environment through language (verbally or in writing). This frees up mental space to think about other things (like listening to a podcast while I drive to the supermarket).  The examples I’ve provided all relate to memory off-loading, but we do this for other cognitive processes as well.  We write out an outline to an argument, not just to remember it, but to help ease the reasoning load.  We draw a sketch or a diagram to assist with imagination load. We use written mathematical symbols and notations to help us complete mental calculations.  We use maps or GPS tools to support our spatial reasoning.

Proponents of generative AI regularly point to the possibilities for cognitive off-loading as a reason for their enthusiasm. Humans, they say, will be able to off-load significant cognitive work to generative AI tools, thus freeing up capacity for higher-level and expanded thinking. Generative AI tools can efficiently handle lots of mundane mental tasks, opening up more time and energy for creativity.  They will also be able to handle increasingly more complex and sophisticated tasks, thus potentially leveraging human cognition beyond its current capabilities. All of this sounds enticing, but what about when the cognitive process itself is important for humans, and not just a means to an end?  What happens when we cognitively off-load to the point that we erode something essential to our humanity?  What happens when off-loading a process to an AI degrades the outcome for humans, because the process itself was the point?

A Chicago-Style Hot Dog Analogy

Back when I lived in Chicago, I heard an episode of the radio program This American Life that told a story of the Vienna Beef company, producers of the all-beef sausages of the iconic Chicago-style hot dog.  As the story goes, in the early 1970s, Vienna Beef built a new factory on the city’s north-side, relocating from its former Maxwell Street location. The factory was state-of-the art for the era, focused on rational efficiency. Vienna Beef went about producing the same sausages with the same ingredients, but found that the product coming out of the new factory wasn’t as good. The color was gone, the taste was different, it didn’t “snap” in the same way. The company could not understand why, until one day a bunch of employees were sharing memories of the old days at the former factory, and they started telling stories about a guy named Irving. It turns out that Irving’s job at the old factory was to carry the uncooked sausages from one side of the factory, meandering all the way through the plant, to the smokehouse on the opposite end, a trip, given Irving’s tendency to shoot the breeze with everyone along the way, that could take 30 minutes. During this journey through the plant, Irving would pass the hanging pastrami, the warm boiler rooms, and the corned beef tanks before finally getting to the smokehouse.  It turns out that this process gave the uncooked sausages time to warm and absorb the factory. Cutting out this inefficient process at the new factory inadvertently diminished the final product. Sometimes there’s something about the process itself that defines the essence of a thing, even when that process is slow and inefficient.

PCK and Pedagogical Reasoning

I’m currently in the midst of my final capstone research for my Ed.D. I’m conducting case study research with some teachers as they engage with a model of teacher professional learning rooted in the theoretical framework of Pedagogical Content Knowledge (PCK). Lee Shulman articulated the concept of PCK in the late 1980s. He was interested in the nature of teacher professional knowledge, considering the paradigms at the time, he was looking for a third way. For Shulman, teachers were more than just people with some knowledge in a subject-area; at the same time, teaching wasn’t just the application of prescribed pedagogical methods. Shulman argued that teachers must engage in pedagogical reasoning in order to transform their own knowledge of the subject matter into pedagogical forms that make it comprehensible to the students. Shulman believed that this process of transformation was a cognitive act, one that defined the profession of teaching.

Shulman was pushing back against two different trends in the history of research on teaching. Older approaches assumed that teaching was just a process of transmitting information from the teacher to the students; the teacher merely needed to have knowledge of the subject matter.  By the time of Shulman’s work in the 1980s, behaviorist approaches to teaching had been dominant for several decades. Shulman referred to this as process-product research; its focus was on determining the most effective pedagogical inputs for teachers that would ensure the desired outcomes in students. Shulman was dissatisfied with both of these approaches and argued that teaching required teachers to engage in a cognitive process where they draw on their knowledge of the subject matter and on what they know of pedagogical methods, together with their knowledge of the their students in the classroom (their prior knowledge, their pre-conceptions, their cultures, interests, abilities, etc.), in order to select and create the explanations, metaphors, models, demonstrations, exercises, experiences, and assessments that will guide the specific students in that classroom to understand that subject matter.

For Shulman, this pedagogical reasoning, facilitated by teachers’ PCK, was the essence of the teaching craft.  Much of this pedagogical reasoning takes place as the teacher is planning, but it’s also part of teacher reflection on recent teaching and learning in order to plan what comes next. It requires the teacher to think about what worked and what didn’t.  It requires the teacher to assess student understanding and analyze where it’s still partial or misconceived. It requires the teacher to consider the next instructional moves that will strengthen understanding. Teacher pedagogical reasoning is cognitively demanding, but, for Shulman, we can’t transform subject matter for our students – we can’t teach – without engaging in this cognitive process.  Off-loading this process will only diminish teaching and learning. 

The Process is the Point

It’s often difficult to identify what makes a good teacher when you see them in action in the classroom. They seem to act intuitively with a gut-driven sense of what works, what doesn’t, the correct pace and the right question; they even seem to have almost psychic insight into what the students are thinking. But the intuition of an effective teacher emerges because of the painstaking cognitive reasoning work the teacher has put in planning for and reflecting on how to best teach a particular concept to specific students.

I’m concerned that teachers will use ChatGPT – or Khanmigo or whatever other AI tools come out – to generate unit and lesson plans. They will plug in a few prompts about the curriculum, the standards and the students, and ChatGPT will spit out a very viable lesson plan, saving teachers the time and energy-demanding cognitive effort.  But, if this happens, I think it will be a cognitive off-loading mistake. I think it will be an example of off-loading a cognitive process that is actually essential to the teacher and the teaching craft; it would be missing the point that the process is the point. It would be like producing sausages in a fancy new factory, only to find the final product to be limp, pale, and flavorless.

References

Shulman, L. S. (1986). Those who understand: Knowledge growth in teaching. Educational Researcher, 15(2), 4–14.

Shulman, L. S. (1987). Knowledge and teaching: Foundations for a new reform. Harvard Educational Review, 57(1), 1–22.