CREATE Research on Artificial Intelligence

Advances human-centered artificial intelligence for the design, operation, and resilience of complex built environments, with a particular focus on emergency preparedness and response. She develops AI-driven sensing, simulation, and decision-support systems that integrate human behavior, building performance, and environmental conditions to better anticipate and manage high-risk scenarios such as active threats, evacuations, and extreme events. This work combines virtual reality, agent-based modeling, and data-driven methods to evaluate how design and operational decisions influence safety, health, and human outcomes under uncertainty. By grounding AI in real human experience and spatial context, this research aims to inform evidence-based design, training, and policy strategies that enhance resilience and reduce risk in buildings and communities (Liu, Becerik-Gerber, et al., 2025).

Focuses on advancing the state of the art in combinatorial optimization techniques for solving real-world large-scale problems, particularly ones that arise in sustainability areas such as biodiversity conservation planning and urban planning. Her work is at the intersection of discrete optimization and machine learning. One key area is designing machine-learning-driven combinatorial optimization algorithms, by leveraging the plethora of data generated by solving distributions of real-world optimization problems (Huang, Piansky, Dilkina, & Molzahn, 2025; Pynadath, Dilkina, Jeong, John, et al., 2023).

Has developed and tested algorithms to optimize resource allocation, minimizing the threat, vulnerability, and expected consequences of attacks by adaptive adversaries (Cui & John, 2014; Pita, John, Maheswaran, Tambe, et al., 2012; Yang, Kiekintveld, Ordonez, Tambe, & John, 2013). These algorithms utilize a Stackelberg Security Game framework and account for the adversary’s risk attitude, uncertainty, and conflicting objectives. In addition, the algorithms are designed to relax the usual assumption of a rational adversary and account for heuristics and biases that have been empirically validated in behavioral game experiments. Example events modeled include recovery from a biological attack in a major U.S. city (Pynadath, Rosoff, & John, 2016) and public evacuation response during a hurricane warning (Pynadath, Dilkina, Jeong, John, et al., 2023). These agent-based models are informed by empirical behavioral data collected from surveys and immersive virtual simulation experiments.

With his research team, has provided quantitative insights into the optimal adaptation protocols for resilience in interconnected systems, such as civil infrastructure. His work considers physics, system connectivity, and behavioral constraints (Savla, Shamma, & Dahleh, 2020). He has also synthesized, and empirically investigated the efficacy of, algorithms that utilize information asymmetry and the power of peer influence to nudge individual decisions towards social good (Zhu & Savla, 2022). These works collectively provide foundations for an AI system to enable complex systems to adapt, from individual-level to systems-level, towards a variety of disruptions.

Over the past three years has focused on two key areas of safety-critical Generative AI settings for critical infrastructure: (1) understanding infrastructure risk, resilience, and AI connections, and (2) developing robust and secure AI agents based on Reinforcement learning (RL) involving sequential decision making in critical cyber and cyber-physical system settings. While initial progress is promising, more work remains to fully realize the potential for advanced levels of autonomy via AI systems within real-world critical infrastructure operations. (Du, Chatterjee, et al., 2023; Dutta, Chatterjee, et al., 2023; Mukherjee, Chatterjee, et al., 2026; Rahman, Shuvo, Chatterjee, et al., 2025; Rose, Prager, Chen, & Chatterjee, 2017; Thekdi, Tatar, Santos, & Chatterjee, 2023; Thekdi, Tatar, Santos, & Chatterjee, 2024)

Has been studying issues related to AI safety and human/AI collaboration. With regard to the former, funding from UW and the National Science Foundation supported work on the relative strengths and weaknesses of human and machine learning. This project built a platform that allowed humans and AI systems to play a set of rule-learning games on an equal basis. Subsequently, Pulick et al. (2022; 2023) presented results of experiments conducted on this platform, expanding for example on the challenges posed by rule generality for human learners, since humans appear to rely on assumptions about possible simple rules, which may complicate learning of rules that don’t fit those schema. With regard to AI safety, Gutfraind and Bier (2023; 2025) highlighted a number of strategies that could be used to help improve the safety and security of large learning models.

Focuses on advancing data-driven and decision-centric methods for homeland security and disaster management. His prior research applied machine learning to complex security problems, including predicting unauthorized immigration flows using nonparametric models that integrate economic, demographic, climatic, and border security factors, and developing AI frameworks to monitor and classify misinformation on social media during crisis events, enabling more effective situational awareness and resource allocation (Al Aziz, Ahmed, & Zhuang, 2024; Hunt, Agarwal, & Zhuang, 2022). Building on this foundation, his ongoing and future research aims to integrate game theory, multi-criteria decision analysis (MCDA), and Agentic AI to support adaptive, explainable, and strategically informed resource allocation in homeland security and disaster response, particularly in complex, uncertain, and high-risk environments such as the Arctic.

Research Professor of Public Policy and of Civil and Environmental Engineering, and CREATE Director Emeritus, recently served on a panel at the Symposium on Artificial Intelligence and Public Safety sponsored by the Israeli Consulate of Los Angeles. Professor Rose commented on potential contributions of AI to analyzing the economic consequences of and resilience to man-made and natural disasters. Professor Rose’s research team utilized machine learning in developing the reduced form decision-support system known as the Economic Consequence Analysis Tool (E-CAT) (Rose et al., 2017). He is currently working on a Defense Advanced Research Projects Agency (DARPA) project, in conjunction with Noah Dormady, Associate Professor in the John Glenn College of Public Affairs, integrating resilience (Dormady, Rose, et al., 2022) into agent-based models of critical supply chain resilience on the basis of empirical research production function modeling on a project sponsored by DARPA.