Agent deliberation via forward and backward chaining in linear logic

BDI Deliberation cycle, linear logic, forward chaining, Lygon Agent solutions to programming problems are often based on the Belief-Desire-Intention (BDI) paradigm [12]. Beliefs represent what the agent believes to be the current state of the world. Desires specify the proactive behaviour of the agent, in that the agent works to make these true. Often desires can be mutually exclusive or contradictory, requiring the agent to select from among them, and so BDI implementations often use goals, which can be thought of as desires with some restrictions on them (such as requiring goals to be consistent, feasible and not yet achieved). There can be several types of goals, including achievement goals, whcih are dropped once they have been achieved, and maintenance goals, which are continually monitored, even when currenlty true. Intentions are plans of action that the agent has committed to to achieve its current goals. Often there are many ways to achieve a set of goals that the agent is working on, implying the need for a mechanism to choose between them. Implementations of BDI systems are usually based around an observe-think-act cycle, in which an agent will observe the current environment, which may have changed since the last observation, determine which goals it should be pursuing and what plans should be used to achieve them, and choose a particular action to perform. Note that while the number of actions performed in the act phase is not specified, it is intended to be relatively small, so that the agent will be able to detect changes in the environment (which is only done in the observe phase) and respond to them within an appropriate amount of time. Hence a fundamental feature of BDI systems is the manner in which they provide both proactive (or goal-directed) and reactive behaviour. In this paper, we consider how we may adapt existing logical inference techniques to implement a BDI architecture. Using logic as a basis for the architecture will mean that we can develop methods for formal analysis of agent systems via logical inference, as well