Context-Mediated Behavior

Context-Sensitive Reasoning for Intelligent Agents

Context is a critical factor in behaving appropriately. Unfortunately, most agents at best represent and reason about context implicitly. For example, context may be present in the preconditions of plan steps or the antecedents of rules. However, there is little or no attention paid to context per se, that is, as an entity in its own right, and so such agents cannot easily recognize, reason about how to behave in, or learn about their context.

We have developed a style of reasoning called context-mediated behavior (CMB) that explicitly represents an agent’s context to foster context-sensitive reasoning and behavior. In our approach, an agent’s situation consists of all features, seen and unseen, of the agent, the environment, other agents, etc., including the agent’s goals and other knowledge. A context is a class of situations that has meaning in terms of how the agent should behave or make decisions.

CMB, which grew out of the schema-based reasoning used in Orca, represents contexts as contextual schemas, or c-schemas. C-schemas not only contain descriptive knowledge about the context, but as important contain prescriptive knowledge about how the agent should behave when in the context. A context manager (ConMan) continually finds c-schemas, representing known contexts, that the current situation is an instance of. These are merged to create a coherent picture of the current context, which in turn provides the agent with predictions about unseen features of the situation, knowledge about how to interpret sensor and other information in the context, knowledge about how to recognize and respond to unanticipated events, knowledge about how to choose which goal(s) to work on in the context, suggestions for how to achieve those goals appropriately, and settings for behavioral parameters (“standing orders”) appropriate for the situation.

The CMB project overlaps heavily with much of what our laboratory does, most especially with the Orca and CoDA projects.

We are also part of an international, interdisciplinary research community focused on modeling and using context in artificial intelligence and other fields. This community is centered around the CONTEXT conference series.

Details about CMB can be found in our publications page, including soon a book chapter describing the approach.

Contact person: Roy M. Turner