Topic: Whose AI Dream? In search of the aspiration in data annotation
When: November 21 2024, 2-3pm GMT online (register on Eventbrite)
This talk explores the critical importance of annotator perspectives—encompassing their diverse demographics, cultural backgrounds, and lived experiences—in building responsible AI/ML. Challenging the perception of data annotation as simple and standardized, Ding’s research delves into the complexities of annotator viewpoints and work practices, examining how these diverse perspectives impact data quality. Through interviews, ethnography, and mixed methods, this work uncovers a disconnect between acknowledging the importance of diversity and actively incorporating it into dataset production. This is illustrated by examining the annotation of dialogue safety in chatbots, where defining “safety” is inherently subjective and influenced by cultural norms. Moving beyond “gold labels” as absolute truth, this talk proposes alternative methods for interpreting data that embrace annotator disagreement and incorporate qualitative assessments to build more robust and responsible AI models.
Bio: Ding Wang is an HCI researcher at Google Research, currently working with the Technology, AI, Society and Culture team. Her research explores the intersection of HCI and AI, specifically the labor involved in data production and its impact on AI systems. Prior to joining Google, Ding completed her postdoc research at Microsoft Research India, where her projects focused on the future of work and healthcare. She received her PhD from the HighWire Centre for Doctoral Training at Lancaster University. Her doctoral thesis offers a critical/alternative view on how smart cities should be designed, developed, and evaluated.
All welcome! Sign up on Eventbrite.