Simulating Human Agents: Papers from the AAAI Fall Symposium
Michael Freed, Chair
Simulated human agents are a key software component in many kinds of applications including, e.g., simulation-based training, games and other forms of interactive entertainment, and simulation-based tools for analyzing human-machine system designs. Creating sufficiently powerful and realistic human agents presents several challenges. To get the agent to behave capably in dynamic, time-pressured and otherwise demanding application environments requires adapting state-of-the-art AI techniques. Making the human model accurate or believable requires identifying and incorporating relevant human performance data. Finally, reusable, well-documented software architectures are needed to reduce the time and expertise needed to construct new human agent simulations. The symposium will address practical questions about the incorporation of existing AI and human performance modeling technologies into applications such as those listed above. Questions to be addressed during the symposium include: (1) What AI technologies are most relevant for simulating human behavior? How should these be improved or adapted? (2) What aspects of existing human modeling architectures are most/least helpful for building new applications? How can they be improved to become more useful to applications developers? (3) Which aspects of human behavior are most worth capturing in a human modeling architecture, generally or for a given application area? (4) What relevant scientific findings are "ready" to be incorporated into general- purpose human simulation tools? How should one best go about filling in the gaps where appropriate scientific findings do not yet exist?