Jason Gulya is Professor of English and Media Communications at Berkeley College, where he is also chair of the AI and Academic Integrity Committee and member of the AI Task Force.

I am a professor of English and Media Communications at Berkeley College and, like many other faculty, I teach a mix of onsite, online, and hybrid courses. This past semester, I taught four online courses that were fully asynchronous with no mandatory Zooms, designed so that students could log in and complete their coursework at any time. 

I begin with that information for two reasons. First, I know that many other faculty are in the same boat, teaching in a wide variety of modalities. Second, I don’t think there are nearly enough resources for online asynchronous teaching in the age of AI. Online professors often don’t have the benefit of what Tricia Bertram Gallant and David Rettinger (two experts on academic integrity) call “close observation.” I cannot—as Tom Kaspers did for his philosophy students at the University of Chicago—have my online students write with me in a physical classroom. Whenever I search for information about adapting my courses to this new technology, I am met with strategies that demand synchronicity: students and professors sharing a space in real time, so that we can test out (together) when to pick up an AI program and when to do things ourselves. 

Synchronicity is the fastest, surest way to teach process because I can see things unfold in real time, with little to no filter. And yet, in my online courses, it’s a strategy that is mostly off-limits. 

A couple of years ago, this realization gave me a bit of a panic attack. At the time, I was teaching five online courses at once, and the strategies I came across for teaching process and encouraging productive struggle were almost exclusively designed for synchronous courses. I still have a little panic in me, since I acknowledge that (a) building trust into the course, (b) designing for student agency, and (c) showing the value of process go only so far. Those are big moves (and, I think, the right moves). But sometimes I need a smaller strategy, something designed to encourage productive struggle. 

That is why I started to experiment with what I call “pressure bots,” custom chatbots that interact and have conversations with students. These are chatbots that I create to put pressure on my students—or, if you prefer, to selectively increase friction—so that they need to slow down a bit.

The strategy is not perfect, by any means. But it has helped me to reclaim the value of productive struggle even when I cannot observe that struggle, to make it more difficult for my students to offload their thinking to an AI program, and to provide me with a document (a chat transcript) that paints a picture of their thought process.  

How I Use Pressure Bots

I rarely use pressure bots as stand-alone activities. After all, my goal is to create an AI-aware exercise that has students (a) learning the course material, and (b) improving their knowledge of how AI programs work and the possibilities and limitations of incorporating them into their learning process. A healthy skepticism toward AI needs to be embedded in each activity. 

For that reason, the lead-in to the pressure bot exercise and the follow-up are just as important as the chatbot activity itself. 

For the lead-in, I provide students with a reading or video to engage with and work through on their own. I am a big fan of collaborative readings, which I orchestrate in Google Docs or Microsoft 365. I give students a specific time frame (usually about an hour, self-timed) and challenge them to mark up the passage as much as they can, sharing their impressions, observations, and questions on the document itself. Though it’s not AI-proof by any means (is anything truly AI-proof anymore?), the activity does put up a barrier. It is much harder to offload marginal comments to an AI program than to a discussion board or even a quiz. 

Another option is short think-aloud videos, where students film themselves analyzing a passage or video in real time. (Students can even share a video and hit pause when they want to slow down and talk about something.) Think-aloud protocols have a long history in Writing Studies, and they are valuable ways for students to (a) show how they think and learn, and (b) look at themselves from a distance. 

I emphasize for students why it is so important to work through their own ideas rather than cognitively offloading the task to an AI program. I also tell them that diving into the reading or video themselves is going to make the next stage much easier. 

For the next stage, I provide students with a pressure bot. For my writing and research course, I give them a debate bot. Students provide the argument for their research paper, and the bot pushes against it. They need to use their previous readings and critical thinking skills to push back against the pressure bot, challenging it on its own assumptions and defending their position. In my Art of Film course, I provide a pressure bot for the movie Wall-E. After watching the film, students engage with a chatbot that pushes against their analyses of particular scenes (Click here to check out the bot yourself!). For example, the chatbot might ask the student to slow down and explain an idea more clearly. Or it might ask why it needs to analyze the movie instead of simply enjoying it as a form of entertainment. 

Using these pressure bots flips my students’ usual use of AI programs on its head. My students are used to going to AI for answers. This activity switches the dynamic, so that the chatbot is coming to them for answers and putting pressure on them as they pivot, rethink their assumptions, or go back to the text to find new evidence. 

For follow-up: Over the years, I have experimented with some follow-ups for the pressure bot exercise: 

• Annotating the Transcript – Students put the entire transcript into a Word document and write comments on it, with observations of how the chat went. 
• Self-Assessment – I ask the pressure bot to assess students at the end of the exercise. Then, students engage with that assessment and walk through whether they think it’s accurate. Typically, this means providing the chatbot with some basic instructions on the kinds of skills it should assess and how it should assess them. For example, I may ask the chatbot to rate students on their ability to engage critically with the chatbot’s questions, on a scale from 1 to 5. 
• Reflection – Students work the exercise into a metacognitive reflection, where they unpack what the experience was like and whether it helped them learn. For instance, I may ask them if using the chatbot helps or hurts their intellectual ownership of the idea generated. Or I may ask them if they would do anything differently, if they needed to reengage with the chatbot. 

As mentioned earlier, students engaging with the follow-up activities is just as important as experiencing the chat itself. This follow-up analysis takes students away from the chat window and encourages them to think more critically about what the chatbot may have missed and how they understood their own role in the creative process. 

For grading: With these assignments, I often provide students with either a “Complete” or a “Try Again” in place of traditional grades. More often than not, I ask students to either try again or talk to me individually about the follow-up activities, partially because (a) reflecting critically on the chatbot’s outputs is difficult, and (b) these kinds of follow-up activities are where we really build AI-awareness. 

When creating pressure bots for my course, I often reserve them for some of the most complex, load-bearing readings. If I have an early text that helps understand later material, I am likely to provide students with a pressure bot that pushes them to either think through their ideas about the text carefully or, perhaps even more crucially, go back to the text itself. In my courses, I have learned that some of our readings are more prone to general summaries than others. In that case, I find that pressure bots are great ways to get students to go back to the text, find specific examples and quotes, and nail down how that material helps push back against the chatbot.  

How I Create Pressure Bots

I use a tool called Playlab to create the pressure bots. The program allows me to design the custom chatbots and share them with students, so that they can use them without limits and without even having an account. Playlab is currently free to use. (The one caveat: as of right now, instructors need to sit for an hour-long session on the program before getting access to it. In my experience, it is a small price to pay.) 

The most important part of the setup is the bot instructions. And for these, I use the acronym GRIT: Goal, Role, Instructions, and Tone. It’s a quick reminder to myself that the goal is to encourage students to engage in productive struggle. It often takes me about 15–30 minutes to write and play with the initial prompt before running it 2–3 times and making tweaks (for a total of 30–45 minutes).  

If you click here, you’ll be able to see the full prompt for the Wall-E pressure bot, along with some of my annotations that walk through the prompt’s structure and how you might adapt it for your own purposes.

Why I Do It

I am not a fan of requiring too much AI use in my courses, especially given that, in accordance with the AI-aware framework put out by Annette Vee, Marc Watkins, and Derek Bruff, I am far more interested in whether students can articulate their own understanding and approaches to this increasingly ambient technology than in whether they can simply use and create with it.  

But I do hope my students allow me to assign a few activities like this, where they engage with an AI program in a slightly different way than they might be used to. If they decide to engage in this activity and move on—because they see it as ethically questionable or because they felt that it didn’t allow them to engage in productive struggle—I completely understand and do what I can do to honor that position. 

In fact, I would count it as a win. 

Process Statement

Here are the steps I used to create this blog post. 

1. Wrote out the first three sections (a 1 on the AI Assessment Scale, first developed by Leon Furze, Mike Perkins, and Jasper Roe, meaning no AI) – I did not need help getting started and had a relatively good map in my head. I wrote rapidly. 

2. Used Claude to get suggestions on fixing up the grammar and condensing (a 3 on the AI Assessment Scale, meaning moderate use of AI) – This part was more collaborative and less of a solo act. I needed to clean up some of the grammar and syntax quickly. I asked Claude to suggest some changes. Most of them were logical. I did not run into any trouble with the suggestions for condensing sentences.  

3. Used Claude to generate a possible ending (a 5 on the AI Assessment Scale, meaning full-on AI) – I was a bit stuck on how to finish things. I wanted to experiment with using the program to generate an ending, because I was curious how it would complete my thought process. Its attempt was horrible, so I deleted it. 

4. Wrote my own version of the ending (a 1 on the AI Assessment Scale) – After deleting Claude’s attempt, I took a step back and thought about what was missing. It completely missed the link to the AI-aware framework. So, I framed my ending around that. 

MEET THE AUTHOR

Jason Gulya is Professor of English and Media Communications at Berkeley College, where he is also chair of the AI and Academic Integrity Committee and member of the AI Task Force. He recently co-wrote a book on metacognition and reflection in the age of AI and is currently completing another book on process-focused teaching (under contract with University of Oklahoma Press).