Foreseeing, Recognizing, and Influencing Possible FuturesUsing Multi-Agent Planning Algorithms

Future-oriented thinking is challenging because of the huge number of possible futures that may exist and the complex relationships between the events that lead to them. Intelligence analysts are tasked with monitoring global events and notifying decision makers of impending attacks or other dangerous activity. They have to account for the beliefs and intentions of the agents involved and how previous events cause future events. This is not a one-time activity, but an ongoing process where each new observation changes the landscape of what futures are possible and impossible. In short, the analyst's task is to foresee possible futures, recognize critical events, and identify how possible futures can be achieved or avoided. This project will investigate how multi-agent planning algorithms can serve as creativity support tools for intelligence analysts. These algorithms provide a formal, generative model of how agents act according to their beliefs and intentions. We hypothesize that these models will be useful in two different ways: as a generative tools to understand possible futures and as an analytic tool to make sense of observed events. When used as a generative tool, a model of how agents behave can generate examples of possible futures, answer questions about what is possible, and explain how different futures come about. They can also recognize which events are critical to those futures (e.g. event x is necessary for this future to happen). When used as an analytic tool, a model of how agents behave can be used to fill in missing information for a partial sequence of observed events. When we know an agent's beliefs and intentions, we can predict their behavior, so this should also work in the other direction—when we observe an agent's behavior, those same models can be used to infer what beliefs and intentions would have led to that behavior.