CAREER: Learning to Avoid Tragedies: Heterogeneous Multi-Agent Learning in Dynamic Environments

Society is rapidly evolving at the whims of cultural shifts and technological advancement, leading to systems that are driven by decision-makers with various types of capabilities, behaviors, and incentives. This emergent heterogeneity is now a defining feature that governs the modern world, yet its impact is poorly understood at a scientific level. How will societal outcomes unfold when everyone has an abundance of information and AI tools at their fingertips? Will these outcomes be for the better? This CAREER project seeks to shed light on how and when decision-making at a large scale can avoid bad, or tragic, outcomes. The intellectual merit of the project includes the development of theoretical frameworks that are well-equipped to analyze the benefits and impacts of AI tools to the society. The broader impacts of the project include generating new approaches for solving collective action problems that arise in the society, such as the management of common resources and epidemics. Moreover, this project will develop a new course on agent-based simulations that will be made accessible to all students on campus. It will also include K-12 outreach activities as well as the involvement of undergraduate researchers. The project focuses on studying multi-agent systems driven by large numbers of agents that interact with dynamic environments and learn to make decisions over time. The proposed research will build and validate new mathematical frameworks at the intersection of evolutionary game theory and control systems to address the challenges associated with predicting and influencing collective behavior. This effort represents fundamental extensions to existing analyses, which primarily focus on dynamics where agents obey simple learning rules. The project represents a unified effort to understand the impact that a wider range of more complex agent learning has on societal systems, such as the utilization of common resources and epidemics. A persisting question in this project is to identify whether agents with higher decision-making capabilities necessarily lead to better societal outcomes. Complementary to this effort, the project also seeks to derive new classes of influence mechanisms that can steer the dynamics to desired system outcomes, where the populations are comprised of agents that follow a variety of learning behaviors. The findings from the project’s investigations will broaden the domain of problems that can be addressed, as well as provide practitioners with insights for solving pressing societal problems. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria. NSF Award ID: 2541011 | Program: 01002627DB NSF RESEARCH & RELATED ACTIVIT | Principal Investigator: Keith Paarporn | Institution: University of Colorado at Colorado Springs, COLORADO SPRINGS, CO | Award Amount: $527,459 View on NSF Award Search: https://www.nsf.gov/awardsearch/show-award/?AWD_ID=2541011 View on Research.gov: https://www.research.gov/awardapi-service/v1/awards/2541011.html