This project investigated strategic decision making among networked agents whose behavioral choices are interdependent. The project has led to several breakthroughs in the field of computational game theory. Because of its interdisciplinary nature, the project has made connections with political science and other areas of the social sciences and engineering. Overall, the award supported a collaborative project that has successfully interwoven fundamental research with undergraduate and graduate education, enriching both.

In terms of specific scientific advances, the core of this research investigated a game-theoretic model of multiagent behavior called an "influence game" where the agents are connected by a network. The network captures the interdependencies among the agents' behavioral choices (e.g., a network of influence among senators). Notable contributions of the project on influence games include (a) modeling more than two behavioral choices as well as a continuum of behavioral choices; (b) modeling how a network of influence changes dynamically over time and designing machine learning algorithms to learn how the agents influence one another; (c) modeling the underlying behavioral context (e.g., the type of bills, sponsors, etc.) that may impact a network of influence and designing context-sensitive machine learning algorithms; and (d) designing a variety of novel algorithms for computing game-theoretic solutions. In addition to creating new knowledge on the influence game model, the collaborative project has made significant contributions on the other fronts, oftentimes applying influence games to other areas. For instance, the project began to explore applications to the study of social network interaction, and high-level behavior and decision making using a very large dataset of large-population information about individual players and groups collected from an online team-player video game. It also began to explore extensions to domains such as autonomous driving. Those explorations already highlighted the potential opportunities, limitations, and challenges likely to remain in extending the techniques that the project produced to those domains. As another example of broader impact to other areas of science, given that the project established a connection with mathematical models in biology and the physical sciences (i.e., compartmental systems), the project has the potential to more broadly impact the future design and development of improved mathematical and computational models for complex systems in those areas.

In terms of broader societal impact, the algorithms and tools developed in this project may have many potential real-world applications. For example, we can study the following questions in the context of congressional voting: Given the nature and language of a legislative bill, how will senators vote? How do bill types impact the extent of polarization in Congress? Who are the most influential senators for achieving a desired voting outcome? The research may also be applied to studying how a network of influence plays a role in the propagation of information and behaviors.

In terms of specific impact on education, having brought together for intellectual exchange through research undergraduate students at Bowdoin College, the collaborating institution, with undergraduate and graduate students at the Dearborn Campus of the University of Michigan (UM-Dearborn), the collaborative project has been immensely successful in its objective of collectively enriching the education and training on advanced research of students at both of the collaborating institutions, including (a) aiding the training of underrepresented students in CS, (b) the incorporation of research results into undergraduate courses at Bowdoin; and (c) the creation of lecture notes and video lectures from workshops between Bowdoin and UM-Dearborn and other visits organized by the respective collaborating principal investigators (PIs).

In terms of professional development of human resources and other broader impact on institutional resources, the project has aided the training on advanced research of nearly 30 undergraduate students at Bowdoin; some have gone on to apply for graduate studies and received prestigious research fellowships, with one student winning an award at a selective conference. Specific to UM-Dearborn, the project has trained a graduate student who is about to complete his Ph.D. in computer and information science, thus supporting a fledgling program created about 7 years ago, and an undergraduate student who has already joined the local labor force working in the general application areas of artificial intelligence, a core technical area of this project, and who has pursued plans to continue Ph.D.-level graduate studies.  Moving forward, potential for improving the quality and quantity of institutional resources at UM-Dearborn in the near future and sustaining it in the long term. The products of the work carried out during the project supported by the award provide the PI a significant foundation in which to continue to systematically build a sustainable research program and consistently seek and procure funding to support larger projects at UM-Dearborn in the future.

Last Modified: 01/19/2025
Modified by: Luis E Ortiz