Declarative Adaptive Agents

This is a report on the aims of and the latest results from my Ph.D. research work. The aim of the work is to create a framework which allows a declarative representation of programs that interact with external processes and can adapt according to generalisations made from that interaction. Such systems are also described as adaptive agents. A new logic programming language, Eel, has been developed especially for this work. The language has a declarative interpretation of process interaction based on abduction of partially ordered sets of communication events. An Eel interpreter which simulates concurrency and process communication , has been implemented in Prolog. The Eel language has been used to implement an agent with a meta-level 'perceive-act' cycle. The agent interacts with concurrently running objects and extracts knowledge about the functionality of these objects in form of example communication events. From these examples, the agent is to induce the general behaviour of the external objects in order to successfully manipulate them. I am currently working to create a suitable learning algorithm for inducing the general properties of external processes from communication events, and to nd out how to use the synthesised programs to satisfy given goals. 1 Motivation Artiicial Intelligence has left the approach of trying to implement intelligence in programs and has with some successes started to look to learning as a way to achieve intelligent behaviour 1]. Due to the lack of a declarative framework in which to handle communication with external processes, these new successes in AI are all based on non-declarative learning methods. Declarative machine learning methods are superior to non-declarative ones in providing a declarative representation of the knowledge gained. Such declarative representations of knowledge are possible to use as programs and they provide a new clarity which helps the understanding of adaptive programs.