User understanding of cognitive processes in simulation: A tool for exploring and modifying

Abstract

Agent based simulations often model humans and increasingly it is necessary to do this at an appropriate level of complexity. It has been suggested that the Belief Desire Intention (BDI) paradigm is suitable for modeling the cognitive processes of agents representing (some of) the humans in an agent based modeling simulation. This approach models agents as having goals, and reacting to events, with high level plans, or plan types, that are gradually refined as situations unfold. This is an intuitive approach for modeling human cognitive processes. However, it is important that users can understand, verify and even contribute to the model being used. We describe a tool that can be used to explore, understand and modify, the BDI model of an agent's cognitive processes within a simulation. The tool is interactive and allows users to explore options available (and not available) at a particular agent decision point. INTRODUCTIONDeveloping an Agent Based Model for social simulation requires the modeler to capture people's behaviors. In Padgham et al. (2011) it was argued that simple rules, typical of many agent based modeling platforms such as Repast (North et al. 2006) are inadequate, or at least inconvenient, for modeling the decision making of humans, who often operate using abstract plans over multiple time steps. This work described the integration of the JACK Belief Desire Intention (BDI) agent platform (Winikoff 2005) with Repast, to allow for easier modeling of human decision making processes. Also, it is becoming widely accepted that to make social simulations effective, with respect to their particular intended use, then end users, stakeholders and/or domain experts, need to be involved in the model specification, design, testing and use (see e.g. Ramanath and Gilbert (2004)). Who should be involved and how depends on the intended use of the simulation, which may range from education to social research exploration to decision support. This paper examines how -in the context of using the BDI paradigm to specify agents within an ABM simulation -to give users the possibility to understand the cognitive processes of an agent in the simulation, interact with these processes during a simulation, and also to help specify what those cognitive processes might be.Agent based simulations can be used for a range of different purposes around gaining greater understanding of complex situations. We have done some work (and developed a prototype simulation) around evacuation in response to a bushfire (i.e., forest fire or wildfire). In exploring with stakeholders how the tool we have developed may be further refined and used, one aspect which stands out is the need to explore and potentially interact with a representation of the cognitive processes (plans, goals and decisions) of the agent. If the simulation was to be used in community awareness building, we have been told it will be necessary for an individual to identify "their" representative agent in the simulation visualization, and also

Agent programming in the cognitive era

“…Perhaps the key contribution of the BDI-based approach, at least compared to agents where programs are learned, is the intelligibility of the resulting agent behaviour by end users and other stakeholders, see, e.g., [26,88,108,128]. This has been termed Explainable AI (XAI) [69], i.e., the development of autonomous systems capable of explaining their decisions and actions to human users.…”