A&S T&L Community of Practice: Red-Teaming Generative AI – New Strategies in Undergraduate Education

Presented by Teaching & Learning, Office of the Dean, Faculty of Arts & Science

Presenter: Siobhan O’Flynn, Director, Canadian Studies Program, University College; Assistant Professor, Teaching-Stream; Undergraduate Instructor, Department of English

Session Description:
The advent of generative AI tools has significantly altered the educational landscape. Over the past year, I developed a series of modules and activities to enhance students’ understanding of the limitations and potential risks associated with predictive text technologies. This presentation will outline the three modules that foster critical engagement with AI.

Module #1, Facticity vs. Concoction, highlights the dangers of assuming the reliability of AI-generated text, drawing on current best practices and cautionary insights from the field. Module #2, Guardrails and Algorithmic Bias explores the inherent biases in generative AI systems, offering strategies for testing AI guardrails through activities such as bilingual testing and red-teaming exercises. Module #3, The Delusion of Sentience, prompts students to consider the ethical concerns surrounding AI chatbots and the potential delusional perceptions of sentience, informed by Joseph Weizenbaum’s concept of the “Eliza effect.”

The session will provide participants with critical frameworks and adaptable activities and prompts, encouraging deeper reflection on AI’s role in educational contexts.

Visit our site for more upcoming events and Teaching & Learning-related content. If you have any questions, email us at teachinglearning.artsci@utoronto.ca.