This talk explores the "AI-Agency Paradox," examining how the rapid adoption of conversational AI is driven by a powerful, yet potentially deceptive, subjective perception of capacity. Drawing from a cross-cultural ethnographic study and nationally representative surveys, we demonstrate that daily users frequently link cumulative AI usage to perceptions of increased personal agency. These perceived agency gains play a critical role in driving sustained usage, often outweighing traditional drivers of human-computer interaction such as reliability, accuracy, and consistency. We observe these perceived agency "boosts" across five distinct dimensions: cognitive, social, affective, instrumental, and structural, with an emphasis from existing AI products on instrumental productivity gains. However, these immediate gains can coexist with concerns about long-term disempowerment, including cognitive outsourcing and dependency. To mitigate these risks, we argue that centering all dimensions of human agency should be a primary design principle for AI technologies, requiring models and products to serve users' long-term aspirations rather than short-term desires. To translate this from theory to action, we propose a new framework for measuring "structural agency" to ensure AI development and use can support substantial and sustained human empowerment.
Note: This talk features unpublished (shortly forthcoming) work by interdisciplinary researchers at Google Jigsaw, Google Tech & Society, University of Southern California, and Harvard University.
Speaker
Beth GoldbergHead of Research & Development at Jigsaw