TransparencyAccountabilityTrustworthinesscs.HC
More than Model Documentation: Uncovering Teachers' Bespoke Information Needs for Informed Classroom Integration of ChatGPT
Mei Tan, Hariharan Subramonyam
September 25, 2023
International Conference on Human Factors in Computing Systems
40 citations
3 influential
Key Takeaways
-
Teachers lack practical guidance for classroom use: The study found that 22 middle and high school teachers faced "significant information gaps" when trying to assess ChatGPT's capabilities for specific learning tasks, revealing that standard AI documentation like model cards don't adequately address educators' real-world needs.
-
Current AI transparency tools aren't designed for educators: Existing model documentation was created primarily for technical audiences and fails to bridge the gap between "pedagogical and technical knowledge" that teachers need to make informed decisions about integrating ChatGPT into their classrooms.
-
Teachers struggle to serve diverse learners: A key finding was that teachers lacked clarity on "ensuring its fit with the needs of diverse learners," suggesting that current ChatGPT documentation doesn't help educators understand how the tool might affect students with different learning needs or backgrounds.
-
Interactive documentation could empower teachers: Rather than static model cards, the researchers propose "a framework for interactive model documentation" that would give teachers hands-on ways to explore ChatGPT's capabilities and limitations for their specific educational contexts.
-
AI tools entered schools without proper preparation: Unlike typical educational technology adoption, ChatGPT arrived in classrooms without "comprehensive training, documentation, and vetting," leaving teachers to independently assess risks and benefits—highlighting a gap in responsible AI deployment practices.