Motivated Declarative Agents in Multiagent Domains: Open Issues

Despite the renewed interest in the declarative nature of goals in autonomous agents, a gap between declarative agent theory and practical architectures still exists. This can be attributed to the way in which architectures of autonomous agents were initially designed to rely on a pre-defined plan library that includes all possible agent behaviours encoded in procedural plans. In such architectures, the agent knows only that when a certain condition holds it should execute a plan in its entirety to achieve an implicit goal, and if some step of this plan fails, that implicit goal is viewed as impossible. An agent operating under this model has limited flexibility and no actual autonomy since its only concern is to carry out plans regardless of their implications to higher-level objectives. In this report we argue that agent autonomy is closely linked to the declarative nature of goals and the agent ability to reason about the importance of goal achievement as well as the implications of plan and goal failure. We review work on motivations and argue that a motivations model can be used to direct autonomous agent behaviour and drive the agent to adopt or to drop goals. We also review work on multiagent interactions aiming to investigate how social behaviour may be initiated by an agent following declarative goals. The literature thus reviewed allows us to point out a series of issues relating to the construction of practical architectures of declarative agents.