Use of Generative AI in CSE 333
Official Policy
This course uses the Allen School "AI as a Reference Tool" policy.
In short, AI may help with general questions and ungraded work, but
not graded assignments or assessments.
See the link for the full description and guidelines, including
permitted and prohibited actions.
Other related information can be found in the
Academic Conduct section of the
syllabus.
If you decide to use generative AI in this course (e.g., ChatGPT, Codex, Claude, Bard, UW Purple), make sure that it is in line within course policy. If you are unsure about what may constitute inappropriate use of AI tools, please don't hesitate to reach out to the course staff about the situation!
Generative AI and Learning
Article: What Every UW Student Should Know About AI
Written by UW's Vice Provost for AI (and CSE faculty member) Noah Smith.
Video: Generative AI and Education
One perspective on generative AI and education. The section on "The science of learning" (starting at 11:18) is also a good refresher on good habits and practices for learning in general.
Risks of Generative AI
- Accuracy: If you are using generative AI tools for learning then you should always double-check the content. For example, if you are assigned to write a program that uses a specific algorithm, AI tools may generate a solution that arrives at the correct answer but does not use the required algorithm. If you use generative AI to assist in the creation of assessed content then you are responsible for the accuracy and correctness of the work that you submit.
- Quality: Content generated may be of poor quality, and generic in nature. Code may have security flaws and may contain bugs. It is important that you understand how any generated code works and you evaluate the quality of the content.
- Learning: Generative AI can be a powerful productivity tool for users who are already familiar with the topic of the generated content because they can evaluate and revise the content as appropriate. Tasks assigned by your teachers are designed to help you learn, and relying on AI tools to complete tasks denies you the opportunity to learn, and to receive accurate feedback on your learning.
- Over-reliance: Using AI tools to do your work for you may achieve the short-term goal of assignment completion, but consistent over-reliance on AI tools may prevent you from being prepared for later examinations, subsequent coursework, or future job opportunities.
- Motivation: Some students may experience lack of motivation for tasks that generative AI can complete. It is important to understand that you need to master simple tasks (which generative AI can complete) before you can solve more complex problems (which generative AI cannot complete). Stay motivated!
Impact on others
There are many consequences to inappropriate usage of AI tools. Some of these consequences may be unintended, and could potentially harm others. For example:
- Other students: You could expose other students to harm by preventing their learning or including content in a group assignment that violates academic integrity.
- Faculty: Enforcing academic integrity standards requires time and energy, and is emotionally draining to teachers and administrators. The increase from the use of AI tools has been substantial; we would much rather spend our time and energy in creating better educational experiences for our students.
- Institutional: Including code from AI tools that you do not understand could expose the university to loss of reputation or even financial harm through lawsuits.
Generative AI and Coding
CSE 333 is an introductory course about building software systems and software that interacts with hardware systems, using the C and C++ programming languages. These are things that generative AI models and agents can do quite well at this point. And yet, we are sticking with the "AI as a Reference Tool" policy. Why?
Foundational Programming Capability
Even in a world where most code generation is done by AI models and agents, the models are not yet good enough to be trusted to create large, complex systems that don't have correctness, robustness, or security issues without significant direction and iteration from the user (using them on introductory problems give a false sense of their capabilities). Many software engineers currently spend a lot of their time understanding and checking code produced by generative AI, but most of them already have the requisite foundational programming capabilities to do so (e.g., graduated from a pre-genAI CS program). You have no shot at debugging generated code if you aren't already deeply familiar with the languages involved. We aim to build your familiarity with C and C++ code in this class.
Working on a Development Team
The importance of knowing how to work as part of a development team before adding an AI agent into the mix. [More coming soon]
Costs: Data, Ethics, and Surveillance
The classroom should not be a surveillance environment. Reliance on generative AI tools is reliance on private corporations that don't have your best interests at heart — no guarantees about pricing model and availability. Don't want this to act as a hidden "course fee." Generative AI tools are build on the stolen works of numerous people and the exploited labor (and trauma) of many data labelers. [More coming soon]
Environmental Impact
The energy demands and environmental impact. [More coming soon]