CLIP Talk - Yutong Xie
CLIP Talk - Yutong Xie
AI Behavioral Science: Building a mutual understanding between humans and AI
Abstract: As artificial intelligence becomes woven into daily life, it increasingly shapes how people work, think, and relate to one another. Yet, fundamental questions remain: Can AI serve as a trustworthy and collaborative partner? Can it accurately interpret human intentions? And how will human-AI interactions shape the future of our society? In this talk, I will try to address these questions with AI behavioral science, and introduce three interrelated lines of our work. First, we assess AI behavior using methods from behavioral science, engaging AI chatbots in economic games to measure their personality traits and decision patterns. Second, we leverage AI to interpret human behavior: with large language models, we extract “behavioral codes” that reveal the motivations underlying human actions. Finally, we propose behavioral foundation models, which aim to provide a shared scientific representation of behavior across humans and AI systems. Together, these efforts move us toward a mutual understanding between humans and AI, laying the groundwork for a more trustworthy, collaborative, and sustainable human-AI ecosystem.
Bio
Bio: Yutong Xie is a Ph.D. candidate and Barbour Scholar at the University of Michigan School of Information, advised by Prof. Qiaozhu Mei. Her research spans AI behavioral science, AI for science, and AI for creativity, with publications in top-tier venues like PNAS, ICLR, NeurIPS, AAAI, KDD, WWW, and NAACL. Yutong actively engages in the academic community, co-organizing workshops on AI behavioral science and graph learning, and serves as a regular reviewer for conferences such as NeurIPS, ICML, AAAI, KDD, and WWW. Her research has been recognized with prestigious awards including the Rising Stars in EECS, University of Michigan Barbour Scholarship, Gary M. Olson Outstanding Ph.D. Student Award, and D. E. Shaw Research Graduate and Postdoctoral Women’s Fellowship. Her work is supported by fundings from NSF, LG Research, and Rackham Graduate School. She also collaborates with industry partners like Moblab, Niantic, and ByteDance. Prior to her doctoral studies at Michigan, Yutong earned her Bachelor’s degree from Shanghai Jiao Tong University as a member of the ACM Honors Class.