Designing Fault Tolerant Systems with Multiagent AI

A model is presented for design and implementation of software systems with multiagent AI and concurrent software engineering techniques. The stages of conceptualization, design and implementation are defined by AI methods. Software systems are proposed to be designed through knowledge acquisition, specification, and multiagent implementations. Multiagent implementations are proposed to facilitate a fault tolerant software design methodology, applying our recent research that has lead to fault tolerant AI systems. A particular approach to and an AI formulation of designing fault free and fault tolerant software is presented, which is based on the agent models of computation. An approach using object orinted design coupled with a novel multi-kernel approach is presented. A system is defined by many communicating pairs of kernels, each defining a part of the system, as specified by object level knowledge acquisition. An overview to agent morphisms and algebras are presented and appli ed to the design. Keywords Fault Tolerant AI, AI Agents, Multi Agent AI Techniques, Abstract Multi Agent AI Design AII, Agent Morphisms Project_METAAI@CompuServe.com Copyright © Photo reproduction for noncommercial use and conference publications is permitted without payment of royalty provided that the Journal reference and copyright notice are included on the first page.

Fault Tolerant Specification are triples consisting of objects, actions and relations. Actions are operations or processes. However, the problems of abstraction , object-level programming, and agent view of AI computation are the important components of inter-play in the present paper. A structural design technique with objects is defined to address knowledge abstraction problems.FTKA has a requirement that each object to be defined must have a dual definition in terms of the actions that are taken for exception and recovery. The term "agent" has been recently applied to refer to AI constructs that enable computation on behalf of an AI activity [see for example Geneseth-Nillson]. It also refers to computations that take place in an autonomous and continuous fashion, while these are considered high-level activities, in the sense that their definition is software and hardware independent [Nourani]. Such functionality is typical of what is required to implement complex autonomo us planning systems[2]. The specifications , once expanded further, are in fact of the form , where A is actions, F is the faults and exceptional functions, and (RNA,RFA) the respective relations. NA is for normal action and FA for fault action. In the example on the figure many objects and their fault functions are presented. Let us define a pair of systems, each consisting of a set of objects, actions and relations. In the figure below objects are represented as , where the coobject is a copy of the object on which faults and recovery functions are defined. Figure 1 The Thrust Module Implemented By AgentsCircles are objects, the squares are agents, and object-coobject pairs are enclosed by a rectangle. The dotted lines are agent message passing paths. The objects and the operations of one set of kernels once defined specifies the FTN, while those of the FTF are defined by the dual kernel. The set of kernels defining FTN and FTF are synchronized by cross operations and interact by some operations that are implemented by message communications between FTN and FTF. FTF's major task is that of fault handling and recovery. If fault recovery takes place, in each kernel, the active kernel (a collection of agents) for a particular function, will be the FTN component, while the FTF component does concurrent checks for further exceptions should they be encountered. In each of the kernels there are objects, processes defining the operations, and objects to which there is a corresponding function in the other kernel. Thus FTN and FTF are a collection of objects and processes. Figure 2- Agents Implementing Functions

References

Nourani, C.F.,"Modelling,Validation,and Hybrid Design of Intelligent Systems," February 1997, KEML98, January 1998, Karlsruhe, Germany. Genesereth, M. R. An Agent-Based Approach to Software Interoperability, In Proceedings of the DARPA Software Technology Conference , 1992. Nourani, C.F.'Double Vision Computing," December 1993, IAS-4, Intelligent Autonomous Systems, Karlsruhe, Germany. Nourani, C.F. "A Multiagent Approach To Fault Free and Fault Tolerant AI," Proc. FLAIRS-93, Florida AI Symposium April 1993. Genserth, M, and N.J. Nilsson , Logical Foundations of Artificial Intelligence< Agent Morphisms Project_METAAI@CompuServe.com Copyright © Photo reproduction for noncommercial use and conference publications is permitted without payment of royalty provided that the Journal reference and copyright notice are included on the first page.

Fault Tolerant Specification are triples consisting of objects, actions and relations. Actions are operations or processes. However, the problems of abst.

1. Author contact:METAAI, Project_METAAI@CompuServe.com.