What is the nature of service? This question carries with it a number of assumptions and raises numerous secondary questions. I take it as a given that we have all experienced interactions that we found lacking in "service" and that most of us have experienced (if all too rarely) instances of service that left us more than satisfied. Yet with this common experience, why is it that we have a very fuzzy understanding of the factors that made the former situations undesirable and the latter exemplary. In other words, what is it that makes some people and some actions "helpful" and others not? Through this research program, I want to better understand those factors and thereby build better computational assistants -- and ultimately be a better servant to others myself.
To study helpfulness, we create opportunities for interactions between artificial agents within simulated environments and measure problem solving success and efficiency. One of several environments we have used, MÆDEN, provides a constrained grid-world in which we can pose tasks of varying difficulty. We have implemented an agent architecture, GARCIA, that supports the instantiation of agents with distinct skill sets by selecting a particular combination of skills from a menu. Thus, we can easily create and test agents with a wide variety of competencies. Likewise, we can combine agents with disparate and complementary skills. Our preliminary GARCIA agents have one goal and ask other agents for help as needed to accomplish that goal. With extensions to GARCIA (supported by corresponding changes to the MÆDEN simulator), agents simultaneously pursue multiple goals and must switch focus between them as deemed appropriate. We attempt to quantify the helpfulness of the service recieved by an agent under various conditions that include task difficulty, respective competencies of the primary and helper agents, priority of the helpers' competing tasks, as well as other factors.