An Insight to Current Agent Architectures that Restricts Agents from Adapting to Dynamic Environments

Kumari Wickramasinghe, Damminda Alahakoon


Intelligent agent technology, the result of behavior oriented AI, can be considered as a step towards the next stage of computational intelligence. Methodologies to facilitate autonomous decision making capabilities of agents in dynamic and continuous environments have been studied heavily during the last two decades and as a result numerous agent architectures have been implemented. But still it remains as a task of the programmer to explicitly specify the states that may encounter in an environment. This paper is an insight to current agent architectures in identifying what features are lacking in them that prohibits agents from adapting to environment changes. We have identified abduction, learning and evolution as the requirements for autonomous actions in dynamic environments. As a result an adaptive agent architecture is proposed in this paper. Also the paper presents a conceptual layered framework inspired by natural science of human behavior, psychology and brain science, which provides a foundation for the adaptive agent architecture. The novel feature of the conceptual architecture is that it enables the agent to evolve, such that decisions can be made through learning and abduction processes rather than using a discrete set of pre defined actions or plans.

Citation Info :

In Conference Proceedings - 6th International Information Technology Conference on From Research to Reality, Infotel Lanka Society Colombo, Sri Lanka, 29 Nov- 01 Dec 2004, pp. 49-55, ISBN 955-8974-01-3.