3Behavioral Science of AI
I research the intersection of AI and behavioral science — treating AI systems as objects of behavioral study, and building the instruments and infrastructure needed to study them at scale.
● Data Mining Lab, KAIST · Ph.D. expected Jun 2027
I am a Ph.D. student in Artificial Intelligence at KAIST, advised by Kijung Shin. I work on modeling and controlling AI behavior, drawing on psychology, psychiatry, network science, causal inference, and mechanistic interpretability. My dissertation makes a case for machine psychopathology: that computations resembling mental disorder can emerge in AI systems, and that they can be measured and intervened upon.
Three lines of inquiry, one question: how do systems behave, and can we steer them?
I research the intersection of AI and behavioral science — treating AI systems as objects of behavioral study, and building the instruments and infrastructure needed to study them at scale.
I aim to advance understanding, inference, and control of relations within complex networks. My focus has been on utilizing graph neural networks to learn complex functions on networks.
Psychopathology can be seen as a complex network. With better understanding and control over complex networks, I am interested in enhancing its understanding and treatment.
4 papers that best represent the direction of the research.
Soo Yong Lee, Hyunjin Hwang, Taekwan Kim, Yuyeong Kim, Kyuri Park, Jaemin Yoo, Denny Borsboom, Kijung Shin
The first mechanistic evidence that computations of psychopathology — mental disorder — may have emerged in LLMs. We computationally adapt the network theory of psychopathology and test it across 12 LLMs using mechanistic interpretability and behavioral simulation.
Soo Yong Lee*, Jongha Lee*, Jaewan Chun, Hyunjin Hwang, Fanchen Bu, Ziv Ben-Zion, Taekwan Kim, Denny Borsboom, Jaemin Yoo, Kijung Shin
AEROBAT automates the behavioral research pipeline on AI agents: hypothesis generation, controlled experiment design, simulation, behavioral assessment, analysis, and reporting.
Soo Yong Lee, Sunwoo Kim, Fanchen Bu, Jaemin Yoo, Jiliang Tang, Kijung Shin
How the joint distribution of node features over graph topology — not homophily alone — mediates whether graph convolution helps or hurts.
Soo Yong Lee, Fanchen Bu, Jaemin Yoo, Kijung Shin
Why graph attention degrades as networks get deeper — and remedies that keep attention informative at depth.
20 papers and preprints, grouped by research area and newest first.
* denotes equal contribution
Soo Yong Lee*, Jongha Lee*, Jaewan Chun, Hyunjin Hwang, Fanchen Bu, Ziv Ben-Zion, Taekwan Kim, Denny Borsboom, Jaemin Yoo, Kijung Shin
Geon Lee*, Fanchen Bu*, Soo Yong Lee*, Sunwoo Kim, Kijung Shin
Soo Yong Lee, Hyunjin Hwang, Taekwan Kim, Yuyeong Kim, Kyuri Park, Jaemin Yoo, Denny Borsboom, Kijung Shin
Shinhwan Kang, Soo Yong Lee, Jaewon Kim, Kijung Shin, Buru Jang
Sunwoo Kim, Soo Yong Lee, Kyungho Kim, Hyunjin Hwang, Jaemin Yoo, Kijung Shin
Sunwoo Kim, Soo Yong Lee, Jaemin Yoo, Kijung Shin
Hyeonsoo Jo, Hyunjin Hwang, Fanchen Bu, Soo Yong Lee, Chanyoung Park, Kijung Shin
Taehyung Yu, Soo Yong Lee, Hyunjin Hwang, Kijung Shin
Sunwoo Kim, Soo Yong Lee, Fanchen Bu, Shinhwan Kang, Kyungho Kim, Jaemin Yoo, Kijung Shin
Sunwoo Kim*, Soo Yong Lee*, Yue Gao, Alessia Antelmi, Mirko Polato, Kijung Shin
Fanchen Bu, Hyeonsoo Jo, Soo Yong Lee, Sungsoo Ahn, Kijung Shin
Soo Yong Lee, Sunwoo Kim, Fanchen Bu, Jaemin Yoo, Jiliang Tang, Kijung Shin
Geon Lee, Soo Yong Lee, Kijung Shin
Sunwoo Kim, Shinhwan Kang, Fanchen Bu, Soo Yong Lee, Jaemin Yoo, Kijung Shin
Soo Yong Lee*, Juwon Kim*, Kiwoong Park, Dong Kuk Ryu, Sangheun Shim, Kijung Shin
Soo Yong Lee, Fanchen Bu, Jaemin Yoo, Kijung Shin
Minah Kim, Jungha Lee, Soo Yong Lee, Minji Ha, Inkyung Park, Jiseon Jang, Moonyoung Jang, Sunghyun Park, Jun Soo Kwon
Sarah Pospos, Ilanit Tal, Alana Iglewicz, Isabel G. Newton, Ming Tai-Seale, Nancy Downs, Pamela Jong, Daniel Lee, Judy E. Davidson, Soo Yong Lee, Caryn Kseniya Rubanovich, Emily V. Ho, Courtney Sanchez, Sidney Zisook
Janet R. McClure, Caroline A. Macera, Ming Ji, Caroline M. Nievergelt, Soo Yong Lee, Josh Kayman, Sidney Zisook
Matthew J. Worley, Melodie Isgro, Jaimee L. Heffner, Soo Yong Lee, Belinda E. Daniel, Robert M. Anthenelli
Invited talks and a tutorial series run across four venues.
From clinical psychiatry research to graph learning and AI behavior.
Psychology to artificial intelligence, with the thesis line intact.
Awards, review distinctions, and undergraduate recognition.
Domains, methods, and languages carried across the work.
Best reached by email.