I'm teaching an introductory C course next semester with more than 100 students. Modern AI tools can solve many beginner programming assignments, so I'm considering how to adjust the course without creating an unreasonable workload for students or teaching assistants. Possible approaches include designing assignments around ASCII art or other inputs that AI struggles with, asking students not to use AI, reducing the weight of assignment correctness, or requiring timed work on university computers. However, tightly controlled lab sessions would be inconvenient and expensive for students who already have their own computers, and grading large numbers of paper assignments is impractical. The final exam is also standardized across all sections, so I cannot change it. From a student's perspective, which approach would be fairest and most useful? Are there scalable alternatives that encourage students to learn the fundamentals rather than simply submit AI-generated code?
5 Answers
Paper-based or restricted-computer assessments are still useful for the core concepts, even if they feel old-fashioned. Students don’t need perfect syntax on paper, but they should be able to trace code, design an algorithm, identify pointer and memory bugs, and explain what a program will do. For programming labs, an automated grader can handle normal test cases while hidden tests and a few targeted edge cases check quality. You could also include intentionally flawed code for students to debug. I wouldn’t rely on ASCII art or prompt tricks as the main defense, since AI tools will likely improve at those tasks and the assignments can become artificial.
Another reasonable option is to teach responsible AI use rather than pretending it can be eliminated. Demonstrate how to use it for error messages, documentation, test generation, and alternative explanations, while prohibiting it from writing an entire solution that the student cannot explain. Then assess the skills that matter: predicting behavior, testing edge cases, reviewing generated code, and making small changes without assistance. The important distinction is not whether students ever consult a tool, but whether they develop enough understanding to notice when its answer is wrong.
I would explicitly tell students that using AI to complete the assignments is equivalent to skipping the learning process, then make the connection to the assessments very clear. If the course teaches foundational C, students will eventually face exams or later courses where they must reason about code without assistance. Homework should be framed as preparation rather than the only place where they earn credit. A warning alone won’t stop everyone, but students who want to learn will understand why the practice matters.
The most practical solution is to move the high-stakes assessment into supervised settings and keep homework relatively low-weight. Homework can provide practice, but use short in-class quizzes or coding exercises to check whether students actually understand pointers, memory, control flow, recursion, and basic algorithms. Ten-minute quizzes with a few conceptual questions can scale surprisingly well and also reveal who is relying on generated solutions. I would avoid strict timed homework deadlines, since they can punish students with jobs or other responsibilities and may encourage more copying.
Don’t spend too much effort trying to make assignments AI-proof. Instead, make students explain and modify their own work. For example, have them submit a program and then complete a short follow-up task in class: change the input format, handle a new edge case, predict the output of a code fragment, or explain how memory changes as the program runs. AI can help with debugging and explanations, but students still need to demonstrate that they understand the resulting code. A brief code walkthrough or peer-review exercise could work better than trying to detect AI use directly.
This also gives students a healthier way to use AI. They can ask it to explain an error or suggest tests, but they still have to write, inspect, and defend the important parts themselves.
That makes sense. Fewer, larger assignments combined with brief in-class checks could test understanding without requiring the teaching assistants to grade hundreds of handwritten submissions.