HAI Seminar: Intersectional Biases in Generative Language Models and Their Psychosocial Impacts

The seminar provides a compelling and well-structured presentation of original research on a timely and important topic. The speakers effectively combine personal narrative with rigorous methodology, making the research accessible while maintaining scientific credibility. The use of a concrete example (the ChatGPT story) immediately engages the audience and illustrates the subtle ways bias can manifest in AI-generated text. The research design is robust: collecting 500,000 stories from multiple models and domains allows for statistical analysis of patterns, and the inclusion of power dynamics as a variable adds depth. The connection to stereotype threat is well-founded, drawing on established psychological literature to argue that these biases have tangible psychosocial impacts on students. However, the presentation is a seminar, not a peer-reviewed publication, so some methodological details are glossed over, such as the exact taxonomy of names and the statistical methods used. The speakers also acknowledge the limitations of using names as proxies for race, which is a reasonable approach but not without criticism. The discussion of implications for education is thoughtful, but could be expanded with concrete recommendations for educators and policymakers. The title accurately reflects the content, and the talk does not overpromise. Overall, the seminar is a valuable contribution to the discourse on AI ethics and bias, offering both empirical evidence and a human-centered perspective.

The title accurately reflects the content, focusing on intersectional biases in generative language models and their psychosocial impacts.

8/10

The seminar presents original research with a clear methodology (500,000 stories, multiple models, controlled prompts) and references established concepts like stereotype threat. However, the presentation is a talk, not a peer-reviewed paper, and some details are not fully elaborated.

Markers derived by PSI from the transcript: the creator did not define chapters.

• 00:00Introduction and land acknowledgment
• 05:00Personal motivation: student's anxiety about AI
• 10:00Example of biased story from ChatGPT
• 15:00Audience discussion on the story's portrayal
• 20:00Research methodology: 500,000 stories, multiple models
• 30:00Analysis of race and gender indicators
• 40:00Discussion of stereotype threat and psychosocial impacts
• 50:00Implications for education and future research
• 1:00:00Q&A session

• Stereotype threat — Mentioned as a psychological phenomenon affecting academic performance
• US Census Proposal 2023 — Used for taxonomy of races including MENA

• Stereotype threat — Supports the claim that stereotypes can impair academic performance.