William G. Kennedy  

Ph.D., Captain, USN (Ret.)  



Research Interests

Cognition and behavior at the individual, small group, and societal levels; building computational cognitive models of social cognition and social interactions; Theory of Mind; trust; cognitive robotics; human-robot interactions; robotic autonomy; and long- term learning using ACT-R, Soar, NetLogo, and MASON.

Current Research

Developing a Reduced-Order Model of the evacuation after a no-notice event. An Agent-Based Model of disaster evacuation with every person implemented in the model as an agent is computationally expensive. We developed a simplified, reduced-order, "zero-order" approximation to be computationally efficient. A Cellular Automata is a form of an Agent-Base Model where agents only respond to their environment and do not interact with other agents. Applying a Poisson probability distribution based on distance from the event and time delay on notification of the event is being investigated. This applied research is being supported by the Defense Threat Reduction Agency.

Common Aging Trajectories and Caregiver Stress (2015-present): This project is knowledge mining effort to uncover the common patterns in older adults’ developing difficulties with “activities of daily life.” How people age in terms of ADLs is currently unknown although data has been collected from individuals for over a decade. The sequencing and timing of the development of difficulties with ADLs drives the demands and stresses of all caregivers, kin, professionals, and institutions. Knowledge is needed to credibly model common aging trajectories. This work has developed a mode of caregiver stress, conference papers and presentations, a journal article, and a book chapter.

Developing an Integrated Cognitive Architecture (2010-present): The primary research-focused cognitive architecture, ACT-R, represents only the rational origins of human behavior. It does not address other forms of cognition, specifically, physiological, emotional, and social. This research has been investigating the development of an integrated cognitive architecture combining these three “beyond rational” forms of cognition into both a research-focused architecture and a practical architecture for large-scale computational social science. So far, this effort has produced presentations at ACT-R workshops, ICCM and CogSci posters, a book chapter, and white papers. Teaming with social and neuro psychologists, efforts are now focused on obtaining funding.

Advancing the Understanding of Human Trust (2010-present): Developing computational cognitive models of human intuition and trust of other agents, both human or advisory systems. This work began in collaboration with the Air Force Research Laboratory and the George Mason University's Center of Excellence in Neuroergonomics, Technology, and Cognition (CENTEC). Products include a paper on implicit learning and the organization of a symposium on 'beyond rational' cognitive modeling at the International Conference on Cognitive Modeling (ICCM 2012) and a paper at ICCM 2010 on the interaction between the fast and slow cognitive processes. Recent activities include a collaborative effort with AFRL to model intuitive decision-making and Krasnow fMRI experiments on trust (NSF supported). Current activities include building a cognitive model of behavioral trust in advice from human, avatar, and computer advisors with individual differences to replicate human subject data.

Another project focuses on trust in an advisor with the advisors ranging from other humans to obviously automated systems. The approach is to implement a theory as a computational cognitive model and give it the same stimuli use in the human study to see if we can replicate the individuals' behaviors. This work supports the Air Force Research Laboratory and George Mason University's CENTEC, (Center of Excellence in Neuroergonomics, Technology, and Cognition).

Applying Agent-Based Modeling to Policy in Social Welfare (2014-present): Developing models to evaluate policy options for reducing social worker stress. Integrating a Systems Dynamics model of stress within agent-based models of individuals providing in-home care to older adults. The work has produced a paper presented at BRiMS-2015, a presentation based on abstract presented at GSA-2015, a journal article on the methodology, and a book chapter.

Completed Research Projects:

Characterizing the reaction of a mega-city to a nuclear WMD event as the Principal Investigator (PI) (2016-2020): Originally a 3-year basic research grant, this work included Andrew Crooks as Co-PI, four graduate research assistants, and several volunteer undergrads and high school students. The objective was to develop an understanding of a population’s reaction to a nuclear WMD event through implementing computational models of verbal theories of individual behaviors and exercising them in a computer simulation of a mega-city area. The model shows how the population NYC might react to a nuclear event in the first hour, day, week, and month. In 2018, we won a 2-year, funded extension to improve model’s fidelity. Research continued with refinement and publishing of the geographically specified population with home, work, and schools as the basis of social networks that are updated with survivors’ emergent social networks.

Advancing the specification of autonomous systems (as PI) (2016-2017): Design of an advanced engineering language (as PI): Led Mason work with LMI on a NASA research contract (one year) “Improving the Systems Engineering Complex and Increasingly Autonomous Systems through the Development of Advanced Engineering Language, Symbols, and Visualizations”. The research created a new language and visualization methods to create coherent and shared, explicit and implicit, mental models for the specification, design, development, management, and use of autonomous systems. The work produced two technical reports.

NSF Project on Social Impacts of Climate Change (2015-2016): Facilitated the final stages of this 3-year project. The team developed agent-based models of the population's migration in Canada using four different sources of weather data sources of weather data (800AD-1300AD, last 150 years, and optimistic and pessimistic projections for the next 100 years).

Human Supervision of the Military's Semi-Autonomous Robots (2015): At the Warfighter Human System Integration Laboratory starting in the summer of 2015, contributed to policy discussions on use of autonomous machines and man-machine systems. Developed an expert system monitoring and advising the experimenter on changing the cognitive workload for the Naval Research Lab's experimental operator interface (SCOUT) for controlling multiple unmanned aerial vehicles (UAVs). This work resulted in a paper submitted to the AAAI Fall Symposium on Self-Confidence in Autonomous Systems.

Enhancing the agent cognition in computational social science (2013-2014): Working with Johns Hopkins University, modeled human social cognition and decision-making concerning migration based on economic, violence, political issues, and social influences. This agent-based simulation project was done for the Air Force Research Lab and featured survey data integration and synthesis for statistically-driven decision trees used by millions of agents. Paper presented at BRiMS-2015.

Mason-Yale MURI (Multidisciplinary University Research Initiative) enhancing the agent cognition in computational social science (2008-2013): Successfully led a 5-year, multi-disciplinary team developing agent-based models of societal conflict, humanitarian assistance, and disaster relief in East Africa. Developed cognitively plausible models of household decision-making for pastoralists and agriculturists using a rule-based cognitive architecture. This Mason-Yale MURI (Multidiscipline University Research Initiative) was sponsored by ONR and produced many publications and presentations as the lead author.

Developed and applied experimental data on human judgments of tasks taking "just a moment" (2008-2011). Measured student performing a task involving walking to develop a model of how long it takes to perform a task involving walking. Then had students evaluate videos of someone leaving to perform the same task as to whether the person took too long, too short, or about the right amount of time. The results were incorporated into a robot that knew the difference between the actual performance and human perception of the time to perform the task (humans underestimate times starting at about 1 minute in length). Journal article published in 2011.

"Like-Me" Simulations at a Robotic Theory of Mind (2005-2008). Modeled human and chimpanzee social behavior based on reasoning about another agent's decision-making using a "like-me" simulation. Journal article published in 2009. Developed a cognitive model that maintains a model of other team members' decision-making to improve its own and the team's performance. Paper published and recognized as best in track at AAAI 2008. Built on the Naval Research Laboratory's experience in static spatial perspective taking, added the ability to model another agent's movement so that a robot would covertly approach the other mobile agent. Paper published and recognized as best in topic track at AAAI 2007.

Research sponsored by:

  • Defense Threat Reduction Agency (DTRA);
  • National Aeronautical and Space Administration (NASA) via Logistics Management Inc. (LMI);
  • Center for Social Complexity (CSC), Krasnow Institute for Advanced Study, GMU;
  • Office of Naval Research (ONR);
  • Center of Excellence in Neuroergonomics, Technology, and Cognition (CENTEC), GMU;
  • Air Force Office of Scientific Research (AFOSR) and Air Force Research Laboratory (AFRL);
  • Joint IED Defeat Organization (JIEDDO);
  • National Science Foundation (CO-Pi) (NSF);
  • Naval Research Laboratory (NRL); and the
  • National Academy of Science/National Research Council (NRC).