KAIST ISE x KAIST Urban AI Symposium

Modern intelligent systems, such as service or assistive robots and autonomous vehicles, increasingly operate in human-shared environments. Human behavior in these settings is highly dynamic and uncertain, arising from complex social interactions. As a result, building reliable robot autonomy that can safely operate in human-centric environments requires generalizable interaction modeling and principled uncertainty quantification. Moreover, for humans to understand, accept, and trust robotic technologies, robots must be human-centered, capable of understanding human preferences, generating personalized behaviors, and providing meaningful explanations of their reasoning, predictions, and actions to developers, users, and stakeholders. In this talk, I will first introduce an uncertainty-aware interactive decision making framework with dynamic relational reasoning for safe and robust human-robot interaction. This framework infers explainable relations among entities and generalizes effectively to unseen, out-of-distribution scenarios. I will then demonstrate how robots can infer human intent and internal states and align their behavior with human preferences to enable effective and personalized interaction.

Dr. Jiachen Li will join the Georgia Institute of Technology (Georgia Tech) as a Coca-Cola Foundation Early Career Professor and an Assistant Professor, where he directs the Trustworthy Autonomous Systems Laboratory. He received his Ph.D. from the University of California, Berkeley, and was subsequently a Postdoctoral Scholar at Stanford University. Dr. Li was recognized as an RSS Robotics Pioneer and an ASME DSCD Rising Star. He serves as Co-Chair of the IEEE RAS Technical Committee on Robot Learning and as an Associate Editor for more than ten leading robotics and AI journals and conferences.

Building robust decision making systems for is challenging because decisions must be made based on imperfect information about the environment and with uncertainty about how the environment will evolve. In addition, these systems must carefully balance safety with other considerations, such as operational efficiency. Typically, the space of edge cases is vast, placing a large burden on human designers to anticipate problem scenarios and develop ways to resolve them. This talk discusses major challenges associated with ensuring computational tractability and establishing trust that our systems will behave correctly when deployed in the real world. We will outline some methodologies, including some involving large language models, for addressing these challenges and point to some research applications that can serve as inspiration for building safer systems. 

Mykel Kochenderfer is an Associate Professor of Aeronautics and Astronautics at Stanford University. He is the director of the Stanford Intelligent Systems Laboratory (SISL), conducting research on advanced algorithms and analytical methods for the design of robust decision making systems. His research contributed to what became the international standard for aircraft collision avoidance. Prof. Kochenderfer is a co-director of the Center for AI Safety and senior fellow of the Stanford Institute for Human-Centered Artificial Intelligence (HAI). He is editor-in-chief of the Journal of Artificial Intelligence Research. He is an author of the textbooks Decision Making under Uncertainty: Theory and Application (MIT Press, 2015), Algorithms for Optimization (MIT Press, 2019), Algorithms for Decision Making (MIT Press, 2022), and Algorithms for Validation (MIT Press, 2026). 

As LLM agents begin to autonomously carry out a wide range of tasks, a pressing question emerges: can they be entrusted with an organization's core decisions? Decision-making in complex industrial systems such as logistics, manufacturing, and supply chains largely reduces to combinatorial optimization—a domain in which an LLM cannot verify whether its own answer is correct. It can confidently produce plausible but wrong solutions, and in operational settings such failures translate directly into cost and risk.

This seminar proposes Operational Intelligence as an approach to this problem. Rather than delegating every task to LLMs or machine learning, its essence lies in assigning each part to what handles it best—classical algorithms for optimization, machine learning for prediction, and LLMs for intent interpretation, formulation, and orchestration—and enforcing this division of labor so it operates reliably. Central to this is the OCP (Optimization Context Protocol) Server, which gives the LLM the context to invoke optimality-guaranteeing solvers accurately, keeping it where it is strong and delegating where it is weak. Through cases of application to real industrial problems, the talk shows how the collaboration of algorithms and LLMs leads to decision-making systems that can be trusted. 

Professor Jinkyoo Park graduated from Stanford University with a degree in Civil and Environmental Engineering in 2016 and has been a faculty member in the Department of Industrial and Systems Engineering at KAIST since that year. He is also the CEO of Omelet, a startup building Operational Intelligence.