A Red-Light AI Policy for an Engineering & Society Course
Joshua Earle is an assistant professor of engineering and society at UVA. He takes a firm red-light stance toward AI in his courses, as shown through this syllabus statement from his capstone course.
AI Policy:
Generative AI and Large Language Models are both technically impressive, and seemingly quite helpful in completing large writing assignments. That said, the use of generative AI is not allowed for any tasks in this course, including, but not limited to: summarizing texts, transcribing notes, brainstorming topic or QFE ideas, or producing assignment text (even if only for a preliminary draft).
Reading and Writing is thinking, and building up your thinking and writing skills is a key part of this course, and using generative AI skips these important steps. It's like bringing a forklift to the gym to help you lift the weights. The movement of the weights is not the point of the gym, the point is the change in yourself produced by doing it the hard way. Generative AI has been shown to diminish learning, reduce skills in users, lead to cognitive decline, and is generally counter to the goals of the course.
Generative AI has other ethical problems as well:
- Generative AI systems are environmentally destructive. Generative AI is extremely power-intensive, and uses significant amounts of water for cooling. Just a few ChatGPT prompts can waste 20oz of water, and AI data centers are currently endangering the water sources of the communities in which they reside. Pollution from data centers has also been a large issue in those same communities, many majority-Black. The power use of AI data centers outstrips that of entire countries, and has prompted the increased use of fossil fuel and nuclear power.
- Generative AI systems are trained on text used without consent or attribution to the authors. This is theft and plagiarism, and companies like OpenAI are currently getting sued by many of the people who hold the rights to much of the training data. They have admitted that their business model cannot survive licensing the data, and are trying to undo copyright laws in order to make their models profitable. Using the output of such a tool, even if the text is not an exact copy, is an act of plagiarism, and thus a violation of the Honor Code of UVA.
Furthermore, why would anyone want to read something no one could be bothered to write? Writing is about communication and connection with others. I am interested in connecting with you, in hearing your voice, your thoughts, your insights. The statistical average of everyone in the training data could never do what you can do.
