Stearns Center Recommendations:
Strategies for Teaching Well When Students Have Access to Artificial Intelligence (AI) Text Composition Tools
Overview
Recent advances in AI tools have resulted in broad worldwide access to “chatbots” and other online sites that can produce extended pieces of writing, some at the level of mid-range college student competencies, in response to a very wide range of topics or questions. While this may feel like a present-day version of the 1983 movie, WarGames, the reality is more nuanced and the process of adapting to new opportunities will take a while. This technology does have potential impacts on how we teach and how students learn, and those effects will continue to evolve. Your students currently have access to these tools, and the few who don’t know about them yet will know soon.
It's important to view AI text generators as tools that can be used for a range of purposes, including activities that support learning as well as interfere with learning. You might think, and ask students to think, about how similar tools have changed the work of your field and thus the skills and abilities that educated professionals need, over the past decades:
- Multifunction calculators
- Spelling and grammar checkers
- Website generators
- Statistical analysis software
- 2D and 3D design software
- High-powered imaging tools
The arguments in your field for "doing something by hand" vs. using a tool that enhances some individual abilities -- both in a learning situation and in a professional context -- are likely to be relevant to conversations about AI text generators.
If you would like to participate in a Stearns Center-sponsored discussion of AI Text Generation and Student Learning in February, please take our two-minute survey to help us plan.
General Suggestions: Short Term
For Spring 2023 courses, you should take some basic steps. Additional resources and examples follow this list.
- Learn: Familiarize yourself with the kinds of text that AI tools can generate.
[Note: We recommend not using language that suggests that AI tools “compose” or “write” texts; they are algorithms assembling words and producing text in using a prescribed data set.]
-
- See a brief explanation of how AI text-generation works here or here.
- See examples of AI-produced text in response to a prompt here and here (content disclaimer: the second link is a public social media thread with examples).
- Try an AI text-generation tool out for yourself here or here.
- Read about the limitations of “AI Text Detection” tools here.
- Consider some of the implications for teaching: There’s a great short IHE article here.
- Provide clarity:
- Prepare and make time to communicate directly with your students about the ethics of using AI text generators for assignments in your class. In addition to concerns about “cheating,” you should address the learning value that you intend for students to gain from completing your assignments without AI assistance. (See some sample syllabus statements below.)
- Articulate your policies for addressing issues of academic integrity with regard to students who turn in work that has been produced by a machine.
- Be skeptical of easy solutions: Note that there are some technologies existing and more on the way that will claim to be able to "detect" writing from an AI text generator.
- These may provide some feedback to instructors, but they are unlikely to be anywhere near a foolproof tool. (See this video of a faculty member experimenting with a flawed detector.)
- Chasing after "cheating" students, particularly as technologies continue to develop additional capabilities, is a losing game. Stearns Center continues to recommend that faculty invest their time now in alterations to their courses, messaging, assignments, and support structures to encourage and enable students to engage in meaningful assessments of their learning.
General suggestions: In the next several months
- Begin to explore what "competency" will mean in your field/profession: In 2025 or 2030, as AI tools continue to expand, what do you imagine will separate "average" professionals in your field from leaders in the field? What will experts be expected to have memorized, to do "by hand" without tools, to do quickly in one try? What kinds of complex problems will they need to address using all the tools available? Mason students should be learning (and being assessed on) high-value skills.
- Consider adapting some of your assignments to increase the value of tasks where students do the work without AI assistance and to limit the opportunity for AI assistance to meaningfully contribute to student success. (See suggestions below.)
- Consider adopting activities that include AI text generation as a sample, starting point, and/or opportunity for critique-and-revision
- Avoid actions that over-prioritize "catching/preventing cheating" while placing undue stresses on equity, accessibility, and innovative thinking:
- Requiring handwritten submissions places additional burdens on students, especially those with some physical disabilities
- Adopting intensive surveillance tools or processes -- such as requiring all steps of writing to be done under human and technological supervision -- is likely to produce few gains while limiting students' thinking and introducing additional biases and inequities into the learning process
- Planning out systems to "recognize" "original" writing, either through eyeballing and comparing in-class samples to take-home samples by the same student, or by using an external text-evaluation tool, are likely to cause more problems than solutions, since all of these approaches are subject to bias and error that may reduce equity without significant gains in integrity
Sample suggested syllabus policy language
Brief: Any text generated by an artificial intelligence (AI) text-generation tool (such as ChatGPT) is not accepted in this class as “the student’s own work,” and so will be considered similarly to text published on paper or online or text composed or significantly edited/altered by another person. The use of such text without proper attribution is a violation of academic integrity.
Extended: We have multiple writing assignments in this class. Because the act of composing a response in your own words actually increases your learning, it is important that you complete the task yourself, rather than rely on an artificial intelligence (AI) tool. Completing these writing assignments yourself will help strengthen your performance in this class on later assignments and activities, as well as help you develop professionally and succeed in your career goals. You should also be aware that AI text generation tools may present incorrect information, biased responses, and incomplete analyses; thus they are not yet prepared to produce text that meets the standards of this course. If you do choose to experiment with AI text generation, you are expected to indicate your usage of it and give credit for text that has been generated by AI. Use of AI-generated text without proper attribution is a violation of academic integrity.
Suggested assignment adjustments that support student engagement and learning:
- Increase the specificity of your question or task: For example, rather than an open-ended writing prompt (“compare X and Y”), ask for specific information or examples (ask students to expand on one point discussed in class, create a scenario and ask for specific resources/sources, or provide an extended reading and have students critique/analyze it)
- Require the application of key concepts to specific local cases, problems, or contexts: For example, provide students with a reading and ask them to analyze and expand on the topics or issues in order to focus the students’ thoughts and responses on that paper, example, or case study.
- Require inclusion, citation, and direct response to specific course texts: For example, prompt students to cite the textbook or another reading in class, in addition to X number of additional sources that they find in Y database in your field.
- Require drafts and revisions. Multiple, sequential stages of a larger assignment can be useful to help students scaffold their understanding as well as help them produce their own content. For example, credit/noncredit activities can be integrated for topic development (and you can provide time in class on this or with paired peer activities). Low-stakes activities can be useful for writing out outlines of papers (for grading, these can be completion/noncompletion or similar). Peer reviews and iterative development, particularly when combined with a short (one-paragraph) reflection on what changes were made as a result of a first submission or peer review help students continue to frame and develop their learning.
- Shift grading criteria, especially for formative writing such as discussion boards or drafts, to place additional value on how students respond to and/or integrate ideas or feedback from their peers or from you. For example, integrating more specifics in discussion prompts (see above), then having students make peer comments and respond to each other, will help continue the conversation and allow students to develop their thoughts on the topic more fully.
Upcoming “big questions” for post-secondary educators
- What is "original" student work, and when do we most want it, and for what (educational/growth/ethical) purposes?
- What will students need to do to be “good writers” as educated professionals, as AI text generators continues to gain facility in generating accurate and relevant text?
- How are current AI text generators in a "free" mode likely to progress toward a for-profit approach, and how will that affect students and young professionals?
- How can AI text generators be used to support inclusion and equity in higher education?
- How can responsible writers incorporate the use of AI text generators going forward? For instance, would giving citations and building further or editing the content be sufficient? Are there any “baseline” uses for composing early drafts?
- How can faculty shift their use and assessment of student writing to continue to encourage students to use writing for learning, communication, and documentation of their thinking?
- What kinds of privacy and accessibility concerns might be raised concerning AI text generators?
- What are the ways that bias and various “-isms” (for example, racism, sexism) appear in AI generated texts? How do educators consider the potentials of hate speech?
If you would like to participate in a Stearns Center-sponsored discussion of AI Text Generation and Student Learning in February, please take our two-minute survey to help us plan.
Additional resource pages for faculty:
- Stanford faculty weigh in on ChatGPT’s shake-up in education. Stanford Graduate School of Education. December 20, 2022. https://ed.stanford.edu/news/stanford-faculty-weigh-new-ai-chatbot-s-shake-learning-and-teaching
- Five chats to help you understand ChatGPT. The Atlantic. December 8, 2022. https://www.theatlantic.com/technology/archive/2022/12/openai-chatgpt-chatbot-messages/672411/
- Robo-writers: the rise and risks of language-generating AI. Nature. March 3, 2021. https://www.nature.com/articles/d41586-021-00530-0
- AI Will Augment, Not Replace. Inside Higher Ed. December 14, 2022 https://www.insidehighered.com/blogs/just-visiting/guest-post-ai-will-augment-not-replace
- AI and the Future of Undergraduate Writing. Chronicle of Higher Education. December 13, 2022 https://www.chronicle.com/article/ai-and-the-future-of-undergraduate-writing
- Hidden biases and societal risks from People of Color in Tech, Christian Ilube, Dec. 13, 2022
- “On the Dangers of Stochastic Parrots” by Emily M. Bender, Timnit Gebru, et al, FAccT ’21: Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, March 2021. Association for Computing Machinery, doi: 10.1145/3442188.
- A blog post summarizing and discussing the above essay derived from a Critical AI @ Rutgers workshop on the essay: summarizes key arguments, reprises discussion, and includes links to video-recorded presentations by digital humanist Katherine Bode (ANU) and computer scientist and NLP researcher Matthew Stone (Rutgers).
Additional resources will be added to this page as they become available.
Related Resources
- Ericson: Writing a Goal-Centered Syllabus
Active Learning
Multilingual Students