7th International Workshop on Engineering Societies in an Agent World (ESAW)
September 6-8, 2006

RIACS Sr. Scientist, Maarten Sierhuis, is invited to give a keynote lecture at the 7th International Workshop on Engineering Societies in an Agent World (ESAW), that will be held at University College Dublin (UCD), Ireland. The title and abstract of his keynote lecture is available below.

Title:    "It’s not just goals all the way down" - "It’s activities all the way down"

Abstract:   Representing how people do work can be done at many different levels. In the knowledge engineering and AI world, people’s work has been described in terms of their problem-solving expertise. There, the theory is that we can model people’s problem-solving behaviour by representing this behaviour in a computational model that is able to duplicate some of it. From this, goal-directed reasoning was born. In goal-directed reasoning a goal is an end-state that represents some objective to be reached. Production rules enable us to represent the qualitative relationships between a goal and its causal antecedents. Using backward-chaining inferences, a goal can be reached by starting with the initial goal and reasoning backwards over the rules until the truth of the antecedents of the original goal is found.

The agent community, especially the BDI community, has adopted goal-directed reasoning as the solution for modelling all agent behaviour. Goal-directed reasoning is only one form of reasoning that can be applied to representing agent behaviour, and not a particularly useful one for representing action-oriented behaviour.

The agent-oriented programming paradigm is very useful for engineering societies that correspond with the natural world. Similar to the object-oriented paradigm, the agent-oriented paradigm enables us to represent real-world phenomena with conceptual entities that have a direct correspondence to people and artefacts in the world. However, the agent-oriented paradigm brings another benefit, namely the ability to represent entities that have (or seem to have) autonomous action-oriented and social behaviour.

I will put forward the bold claim that if we want to represent autonomous action-oriented and social behaviour in agents, goal-directed behaviour is not necessarily the most useful way to do this. Our theory of modelling human work practice is based on the notion that most human behaviour in the real world is activity-based. An activity is a situated state of being and doing actions. It has a well-defined beginning and end and takes time. An activity is not necessarily driven by goals, because not every activity is a problem to be solved. Activities are situated, therefore, if we are able to describe the situation in which an activity is to be performed, we can represent when, where and how an agent should do the activity.

In this talk, I will first discuss our theory of modelling human work practice, and will then describe our activity-based modelling paradigm as a paradigm for modelling autonomous and social agents. I will present our Brahms environment, and explain the representational capabilities of the language, by using examples from my research at NASA Ames Research Center.