My AI Background and Usage
I am in a graduate certificate, Digital Humanities in the Age of AI, so I work with AI often. I am neither an AI zealot nor someone who thinks that not using AI makes me somehow pure. I use AI. I use AI a lot, in fact. I should point out that I am not using AI to write this statement—as tempting as that might be since I have a lot of other tasks to get done–but I am demonstrating a process where I am actually mechanically writing out this addendum and then having Claude Sonnet 4.6 provide feedback on my work. Claude is providing feedback and even revision suggestions, but I am not implementing them because I disagree with some of the wording and phrasing since it does not sound like me.
I believe in the possibility of leveraging a Large Language Model (LLM) for productive work in a world where we are constantly taking on too many tasks because that is just how life works now. The point of AI should be to free up some of our time to do things that matter in life: for instance, I enjoy going to the skate park with my friends, playing mandolin in traditional Irish music sessions, and hanging out with my girlfriend and our cat. But I would never use AI to think for me.
A Practical Case
This statement has many latent ins and outs and technicalities concerning what that means to any individual. So, I want to highlight what that means to me. I typically have at least one book to read during my week for classes; I usually have a book to read per week for reasons outside of my classes. Would it make sense for me to just have Claude read the text for me and return uninspired information? No. Because I am not receiving authentic knowledge. What I mean by authentic knowledge is the recognition of how the author’s writing relates to my experiences, worldview, and values.
An LLM cannot accomplish authentic thinking for me because it is essentially, what computer scientist Nick Montfort stated in a 2025 Electronic Literature Organization workshop, “a bullshit machine.” What this means is that LLMs–especially large-scale commercial ones like Anthropic’s Claude, Google’s Gemini, and OpenAI’s ChatGPT–are trained on massive datasets that predict statistically probable outputs that might resemble a “reasonably best effort”; in essence, think of when LLMs “drift” or “hallucinate.”
So what I want to point out is the difference between deference and material shaping. Deference would be me just prompting Claude to “write me an AI policy for my ENC 1102 course.” From there, even if I did not simply cut-and-paste but edited Claude’s output, I am deferring to Claude’s judgement and expertise rather than my own; in other words, this is arguably plagiarism.
On the other hand, let’s say I need to explain Bruno Latour’s Actor Network Theory for my Texts and Technology in History course. I have one week to meaningfully interact with this text and I do not have a background in Actor Network Theory. My first step might be to go to either Claude or DuckDuckGo’s AI Search Assist and simply write, “Can you provide me an overview of Bruno Latour’s Actor Network Theory?”.
From there, I can read through and have a general idea of what Latour is laying out in his theory, and then I will be able to better understand the text because I have more of a background than if I went into it with no information or background. This is similar to the sorts of overviews (or reviews) on materials that instructors provide in lecture. This sort of operation enables me to take more meaningful notes on what patterns I recognize and key questions I may have regarding Latour’s text. Then, when I approach Claude to help synthesize some of my notes into that “half-outline-half-draft” then I know my thinking is the input and that Claude is doing the work of collaborating with me much like how dictation works, but as if there were a built in editor function.
Translation and Access
When a student communicates in English as a second language, they might not have the easiest time brainstorming (see Santos Netos 2012). This tends to be the case in so many instances of second language acquisition. Simply put, a major barrier for anyone communicating in a language that is not their native is that they may experience ideas that are not easily translatable (if at all!) between native and secondary language, or they may not immediately comprehend some of the cultural weight of certain references or idioms.
Likewise, I generally encourage students to first outline in their first language. Then, should they like, they can input that writing into an LLM like Claude to translate that writing to English (in respect to our case here, in the composition classroom). I want to be frank in saying this is not cheating; it is prosthesis. As Claude puts it in our collaborative development of this overview: “Students who use AI for translation are using a prosthetic access tool to participate in a linguistic context that was not designed with them in mind. The concern about AI in this course is cognitive replacement and not linguistic access.” I include this here because I do not think I can put it better at the time of writing this, but Claude is still citing from my ideas on a chapter proposal for “Authentic Thinking with AI in the Classroom”: “Failing to distinguish AI as prosthetic access from cognitive replacement is not merely a pedagogical error but a continuation of deficit thinking dressed in new technological anxiety. This is not a problem AI created but a structural condition of which it reminds” (Ritchey III 2026). Claude has, more than anything, summarized my scholarly language and made it more accessible.
Please do let me know if this is your situation so we may discuss equitable AI usage in the classroom.
Even if the issue is not communicating in English as a second language, many native English communicators still have issue with things like outlining or drafting; I am one of them! A lot of folks have an easier time speaking through their ideas rather than writing them down. If this seems like the case for you, or you want to experiment with this mode of composition, then I recommend Handy. From their website:
I built Handy because I broke a finger, was put into a cast and as a result my hand was out of commision. I tried some of the existing speech-to-text apps but none were open source and extensible. So Handy was made to fill this gap.
It’s probably the most simple speech to text app, it’s only function is to put whatever you say into a text box. Press and hold a keyboard shortcut, speak, and release. Your words appear wherever you were typing. It runs completely offline using Whisper, works across platforms, and doesn’t require subscriptions or cloud services.
One reason I enjoy Handy so much (and use it to leave my commentary on your assignments!) is that it is privacy-focused. It leverages a local-running instance of OpenAI’s instance in such a way that your voice, words and ideas, are not sent to train an AI model. There is no extraction; there is no cloud mediation.
So feel free to speak through your stream of consciousness and see how you may shape your text! This is legitimate process work that many writers, including ones who do not have the same motor functions as others, use in composition. I want to bring this up, too, because a lot of students seem to defer to AI when they feel like they are up against the wall on a deadline; I think you might find it removes some of the pressure to use AI.
The Model We Will Use
I want us all to use Duck.AI for the AI work we accomplish in this course. Duck.AI is privacy-oriented browser and search engine DuckDuckGo’s AI instance that uses many different models and acts as a mediator between you and LLMs like ChatGPT, Claude, and Gemini so that your data is not used to train their models. I want you to use Duck.AI so you may experiment with different AI models to see how they differ in application. Also, Duck.AI has a great, simple export feature that allows me to evaluate how you are using AI for course work. Much like I have demonstrated in this document, I value process transparency. Let’s work on that together.
Unauthorized AI Use
Unauthorized AI use–be it using AI without permission and/or disclosure, that does not use Duck.AI as evidenced by a lack of documentation, or using AI to replace cognitive labor–will result in an automatic 20% deduction from the maximum points possible for any given assignment. I will also use ChatGPT (which is in my opinion the worst AI model) to grade your assignment.
This is not a “gotcha” clause. This is about us respecting each other’s time and efforts in a way that allows room for discourse. I want to understand the “how and why” of your AI usage so I can consider contextual information that might be absent from my teaching approach. I am more interested in growth and process development than I am in outright penalty. Still, you will be expected to either prepare a discussion during an office hours appointment in a way that is not just in defense, or to accept the grade and whatever feedback ChatGPT generates for me.
Finally
Everyone continues to evolve their understanding of AI’s potential and limitations. I have been wrong about LLM functionality and use cases in the past and probably will be again. This policy is a living, breathing document that will continue to undergo revision. However, the principles undergirding it are unwavering: I want this course to help develop your skills when it comes to thinking and writing. Any tool that helps to serve that goal is worthy of serious consideration, and so is the ability to understand how to properly use any tool: can we effectively hammer with a saw or saw with a hammer? Can we use a fork to eat soup with minimal difficulty?