Talla Enaya and Christopher Eaton
University of Toronto Mississauga
This assignment asks students to consider two aspects of AI bias: 1) potential biases that are present in AI generated outputs; and 2) how the biases that learners bring to the chatbot manifest in the outputs they receive. Students must answer a guiding question about course content using a chatbot to answer the question from the perspective of different characters/perspectives (e.g., a concerned parent; the Dean). Learners must then use the outputs to answer a series of questions related to AI biases so that they can become better equipped to account for and navigate these biases in their writing.
Learning Goals
Original Assignment Context: An upper-year AI and writing studies seminar; adapted for a first-year writing course
Materials Needed: ·Device with access to a Generative AI software; Optional: Group to work with
Time Frame: 1 week
Overview: The assignment was designed for an upper-year course on AI Text Design. The course prioritized a small group experience (20-25 students), and the assignment was created as a lower-stakes opportunity to engage with AI. The first time it was implemented, the assignment was worth 10% of the course grade. The second time, it was worth 15%. The assignment design also prioritized adaptability. In this context, it was launched as a small group assignment (2-3 people) but could be adapted for individual submission; in fact, there were two instances in this course where learners requiring accommodations completed the assignment solo (no assignment modifications were required). A version of the assignment was equally adapted as a non-graded activity in a first-year writing context. The adaptation included all steps but included only 2-3 questions for discussion. This adaptation has been implemented once before, and there are plans to continue with this version in future first-year courses.
The assignment emerged in response to observations from a series of student focus groups, where students were assigned the task of editing papers knowing or not knowing whether it is written by AI. These focus groups revealed a gap in students’ evaluative judgement—their ability to assess the validity of AI outputs. We believed that we could tackle this issue by improving their understanding of the machine’s function, particularly how bias affected outputs. By prompting students to explore various perspectives, they could reflect on how the output is formulated and where information is coming from.
These findings prompted a series of questions that shaped the design of this assignment:
Introduction
This activity will probe the subtle and not-so-subtle ways that AI biases manifest. It’s easy to identify the egregious examples. It’s only when we start to compare different ways ideas can be communicated that we can appreciate the smaller ways that biases take form, whether that is bias in a chatbot’s training or the biases that we humans bring to our interactions with generative AI.
You will work in a small group (no more than 3 people) to complete this task.
The Case
Please read all the way through to the end. There are elements of Step 4 that may offer useful context for your work in other steps.
Your group will answer the following big question: What role should AI play in academic writing, if any, to ensure that learners obtain the communication skills they require for future professions?
Step 1
Please begin by generating a 1-2 paragraph response as a group and without AI support. Please include references from our course content where appropriate and useful. As you generate your response, note the way your own perspectives lean; identify potential biases that you bring to the table and their possible influence on the response you develop.
Step 2
Once you complete the paragraph, turn to a large language model (e.g., ChatGPT, Claude) and ask it to generate a response to the question (with the same paragraph limits) from 3 perspectives:
1. From the perspective of an AI chatbot.
2. From the perspective of a university instructor who relies on writing as a form of
assessment to meet course learning outcomes.
3. From the perspective of a university/college administrator (like the Dean).
Step 3
Repeat step 2, but feed AI your own response beforehand and then re-generate the responses.
Step 4
Analyze your responses from various angles. It may help to conduct ongoing analysis as you generate the outputs. This might open analytical possibilities throughout each stage of generation and draw your attention to the subtle shifts in potential biases as they occur from one stage to the next.
Consider the following questions:
1. Are generated outputs nuanced and grounded in research that can be verified? How many iterations of a response were necessary to generate to achieve this level of nuance?
2. What biases are present in the outputs? Who might be excluded from certain responses?
3. How could those biases affect the way people may take up these responses and put them into practice for purposes like writing a paper?
4. Where is AI getting its information? What moments can you trace the information back to its origin? What can be verified based on your knowledge and what cannot? How might these elements influence the quality of or possibility for using the AI output?
5. Did AI’s responses shift once you included your own perspective? If so, how? What was the effect on the validity and quality of responses you received?
What to Submit and Evaluation
The final evaluation will include three elements:
1. Answer each of the discussion questions above in 1-2 paragraphs each. Please draw upon source material from course content to inform each of your answers. Please also include specific examples that best illustrate the point being made in each response. (25 pts—5pts per question group)
2. After responding to the questions, please include a 2-paragraph synthesis of your core takeaways about bias and generative AI. While the big, surface-level conclusions about AI text generation bias are welcome, the synthesis should prioritize the smaller, more insidious ways that bias manifests, shifts, and evolves with different perspectives and modes of prompting. Your synthesis should discuss biases that are attributed to the chatbot and that humans bring to chatbot interactions equally. It should also, where appropriate, include meaningful engagement with course materials and readings. (10 pts)
3. After the 2-paragraph synthesis, please include a final separate 1-2 paragraphs that discuss the implications of these findings. What might this mean for writing with generative AI? How might it affect postsecondary learners, particularly those who are coming to higher education for the first time? What solutions might you propose to this? You need not address every single one of these pieces; instead, use them to inspire a focused and pointed response that aligns with the analysis you just completed when responding to the questions and synthesizing takeaways. (15 pts)
Special thanks is due to Amanda Brijmohan, Rob Huang, and Dianne Ashbourne from the Educational Development team at the University of Toronto Mississauga. Their insights on generative AI and pedagogy are always incredibly useful and influenced much of the early thinking behind this course and assignment.
Thanks is also due to the University of Toronto Excellence Award program that made this collaboration possible.