AI Policy for an AI Course

In today’s episode of sociotechnical ouroboros, I’m developing an AI fair-use policy for a course in…developing AI (Introduction to AI). My goal is to convince students that you have to understand the fundamentals of a tool before wholeheartedly embracing it. Here, I think understanding AI as a technology means not only “how it works” but who it works for. I’m concerned, too, with the idea of user and non-user; how does an individual use of AI affect others? Yes, that might be the grouchy professor reading AI glop (me) but what about the neighbors of the new data center? What about the authors and artists whose work was used, likely without permission, to develop models, which are proposed as a means of putting them out of a job?

Have a thought? I’d love to hear it (arothschild at bard dot edu). If you adapt this policy for your own courses, I’d love to hear about that, too!

Annabel Rothschild. 2025. “AI Policy for an AI Course.”

—————

CMSC 251 (Intro to AI) AI Policy:

The irony of discouraging use of generative AI in a course that teaches the computational minutiae and theoretical foundation of exactly that is not lost on me. However, it remains the case that in order to build AI systems, we have know how those systems operate in the first place; to acknowledge their utility requires first understanding how they augment our learning environment. Consider Anthropic’s job ads for AI engineers, a position which requires a signed declaration that the candidate did not use any generative AI in the preparation of their application. This is not to mention the ecological harms that come from generative AI, a topic we will return to throughout this course, nor the ethics of how these models were trained [1, 2] (and on whose data). We will also make extensive discussion of the implications of AI adoption and why some communities appear more concerned than others [1, 2].

As a result, generative AI is generally prohibited without prior authorization, with the following rationale:

• Writing assignments in this course are offered as a place for you to grapple with complex sociotechnical ideas and issues. I am much less concerned with your spelling and grammar, beyond the request that, when applicable, you run “classic” spellcheck (or review your writing at a copywriting level) to enhance essential readability (e.g., correct “teh” to “the”). “Classic” spellcheck refers to built-in editors in programs such as Microsoft Word or Google Docs, as opposed to more invasive tools like Grammarly. Classic spellcheck, as we will discuss in this course, can be reasonably categorized as a form of AI, but I ask you to avoid systems that would alter your tone or verbiage. I’m interested in hearing your thoughts as you develop them, not as they “should” sound, written in some monotone, formal, and frankly boring “voice from nowhere” (a concept we will explore in this course).
• Proactive demonstrations of computational assignment completion without any AI assistance is actually good programming practice, though maybe not for reasons you would expect. Documenting your work early and often is a step many of us forget, but it is one of the most critical to building long-lasting, cooperatively-designed software and systems. To document your thinking process and catalog the state of your code is actually an invaluable tool, particularly when building durable tools and systems. Further, this practice should encourage you to work on your assignments early and often. At this point in your CS career, you may have already discovered that bugs are rarely “squashed” with hours of single-minded focus; it is when you step away for a few hours and return to your program that you suddenly see the source of the original error. Subsequently, pursuant to the respective formats of different assignments, you will be required to turn in a “portfolio” of your progress to a final submission.   
• Today’s generative AI tools are usually owned and operated by private corporations and entities. While these programs may be free or low-cost now, what is the likelihood they will be in the future? Learning this material without the assistance of AI will ensure that down the road, whether or not you have access to assistive AI, you will understand the fundamental concepts and be able to continually build your own advanced understandings of the subject.

If you find what feels to be an interesting or novel use case for generative AI related to this course that feels worth these trade-offs, you are welcome to discuss it with me. Note that I will very rarely approve the use of any closed-source or proprietary models except for exploratory purposes. Any use of generative or assistive AI besides classic spellcheck will require a completed AI attribution worksheet: https://research.ibm.com/blog/AI-attribution-toolkit, including portions of assignments that require specific, authorized generative AI interactions as a comparative learning activity.

Should I have reason to suspect that unauthorized assistive or generative AI was used in the completion of your assignment, I will ask you explain your work and thought process in rigorous detail. I do not use AI recognition tools or “detectors” for all of the reasons listed above, as well as the fact that I am highly dubious of their efficacy.