AI Policies for an AI Course

Where This Course Comes From

This course has roots that go back well before generative AI existed. It descends from a course that David Birnbaum developed at Pitt, and many of its exercises, designed long before anyone could ask a chatbot for help, remain essential.

But the changes AI has brought to coding since then are profound. Most professional programmers now work with AI assistance. Some of what the original course asked of students, such as debugging long, complex programs by hand, is now something few working coders do unaided.

Thus, AI introduces exciting new possibilities for a course such as this, while also introducing a genuine pedagogical puzzle. You cannot direct a tool well, or tell when it has gone wrong, unless you understand the logic of what you are asking it to do. But the entry-level work through which people used to acquire that understanding, writing your first function and hunting down your own typo, is precisely the work AI now does for you. Economists have a name for this: the missing junior loop. Automate the apprentice’s tasks and you cut the pipeline that produces the people who can judge the results (Catalini, Hui, and Wu 2026). Part of our task for the semester is to try to close that loop.

In the words of Ian Malcom in Michael Crichton’s Jurassic Park:

You know what’s wrong with scientific power? It’s a form of inherited wealth. Most kinds of power require a substantial sacrifice by whoever wants the power. There is an apprenticeship, a discipline lasting many years. Whatever it is you seek, you have to put in the time, the practice, the effort. You must give up a lot to get it. It has to be very important to you. And once you have attained it, it is your power. It can’t be given away: it resides in you. It is literally the result of your discipline. But scientific power is like inherited wealth: attained without discipline. You read what others have done, and you take the next step. There is no discipline, no mastery. There is no humility before nature.

He meant it as an indictment of scientists who take the next step without the discipline that earned the previous one. So, as we wield this new, unearned technology, let’s be careful not to let the dinosaurs out…

Contextual AI Usage

This course’s stance toward AI is therefore contextual. In many parts of this course you are expected to use AI, and to use it well. In other parts we will not use it at all. That is not because AI is good or bad. It is because understanding certain things without it will let you use it much better.

If you can (or cannot) use AI on a given assignment or exercise, that is by design. The default is that AI is allowed. When it is not, you will know: either the exercise is handwritten, so AI is out of the picture anyway, or we will say plainly that we are asking you not to use it. (Pitt’s Teaching Center describes a range of approaches to generative AI in courses; this course uses several of them, assignment by assignment.)

Pen and Paper

Counterintuitively, then, this course includes short in-class exercises done by hand, without a computer, let alone AI, on core concepts in coding and logic: What is a function? How does a computer think about paths? What is a for loop? The goal is to recognize each concept when you meet it, and, yes, to ask AI to implement it for you in other contexts.

A quite reasonable reaction might be: “I will never have to do this in Real Life, so why on earth are we being tormented this way?” The calculator rendered long division obsolete, and yet you still memorized your times tables: the calculator changed the purpose of that work, making it a means to mathematical reasoning rather than an end in itself. Coding is no different. If you want to tell AI what good code should do, rather than the other way around, you need to spend some time alone with pen, paper, and your own intellect. There is research behind this: among novice programmers, the ability to trace what a piece of code does, and to explain it in plain English, goes hand in hand with the ability to write it (Lister, Fidge, and Teague 2009). When AI does the writing, tracing and explaining are the skills you have left, and they are what tell you whether the writing is any good.

Responsibility

When you submit a piece of code, or use it in your project, you are responsible for it, even if AI did most of the actual coding. You answer for what it does, what it gets wrong, and what it breaks. Pitt’s own standard for AI use makes the same point: you are accountable for anything you produce with AI, and for verifying its accuracy.

In this course, that means you should be able to read the code you use, understand it, and explain it: what each part does, why it is there, and how you know it works.

Does that mean sight-reading every bit of syntax? Not necessarily. Ironically, AI is very good at helping you decipher code, and we will teach you to do that better; and the stronger your independent checks on a result, the less rides on your reading of every nut and bolt. But you do need to understand the code you use at the level of its logic. Call it coding literacy.

Saying What You Did

There is no separate paperwork for AI use in this course. There are two habits instead.

In your code. Where AI wrote or substantially reworked something you hand in, say so in a comment, along with what you checked and what you changed. Every coding assignment requires real comments anyway (see Course Format and Materials); this is part of what they are for.

In your journal. Your Worm Journal is where you work out what you are learning: by hand, in ink, with no AI in the loop. When AI was involved, the questions worth sitting with are the ones that are easiest to skip in the moment. What were you actually trying to do? Where did the model turn out to be wrong, and how did you find out? What did you change, and why? What do you still not understand?

Together they show us your understanding and your learning, not just your output, which is why output is not what this course grades (see (Un)Grading).

Tools

Everything AI-related in this course can be done with tools that cost you nothing:

• GitHub Copilot in VS Code, for working directly with code. Copilot Free comes with VS Code, and verified students get the more capable Copilot Student plan at no cost through GitHub Education (GitHub Copilot plans). Verification can take a while, so apply in the first week (see Getting Set Up).
• Claude for Education, Pitt’s university-provided Claude account, for a chat window and a second model to check the first against.

What you feed into these tools, especially copyrighted texts and archival images, raises questions of its own, covered on the course’s Copyright page.

How These Principles Apply to the Instructor

I hold myself to the same standards.

• Grading and feedback: I will never, ever use AI for grading, or for feedback on your work. Not as a first pass, not as a suggestion, not at all. The same goes for the TAs.
• Your work: I will never put student work into any tool, AI or otherwise, that Pitt has not approved for use with University data.
• Course materials: I use AI in building this course, including some example code and some of the AI output you will be asked to critique. When the origin of something matters to an exercise, I will tell you.
• Everywhere else: I use AI to design websites like this one, and constantly in my own research and work.