一个刻画Agent自主性的框架  

A Framework for Modeling Autonomy of Agent

作　　者：李斌[1,2] 陈韬略[1] 吕建[1] 

机构地区：[1]南京大学计算机软件新技术国家重点实验室,南京210093 [2]扬州大学计算机科学与技术系,扬州225009

出　　处：《南京大学学报（自然科学版）》2004年第2期137-145,共9页Journal of Nanjing University（Natural Science）

基　　金：973项目 (2 0 0 2CB3 12 0 0 2 ) ;国家自然科学基金 (60 2 73 0 3 4 ;60233010 ) ;863计划 (2 0 0 1AA113 110 ;2 0 0 2AA1160 10 ) ;江苏省自然科学基金 (BK2 0 0 2 2 0 3 ;BK2 0 0 2 40 9) ;江苏省教育厅自然科学基金重点项目 (0 1KJB52 0 0 10 )

摘　　要：自主性是agent最显著的特征 ,为了在理论上指导自主agent的设计和构造 ,促进基于agent的软件系统的开发 ,建立了一个刻画agent自主性的框架 .首先 ,在分析现有的agent自主性定义的基础上 ,总结了自主agent应具备的最小能力需求 ;然后 ,介绍用于描述agent的语言 ,agent的心智状态及其表示 ;在此基础上 ,提出一个自主agent的框架 ;最后 。One of the most notable attributes of agent is its autonomy.Although no definition of agent autonomy is universally accepted,exploring the concept of autonomy is important because autonomy influences and is influenced by the degree of trust that agent designers and users place in their agents,the effectiveness of agent_based problem solving,and the organizational structure of agent_based systems.This paper has two goals.The first goal is to motivate that autonomy is a complex and multifaceted concept that is best captured by using a multidimensional formalization.The second goal is to introduce our multidimensional conceptualization of autonomy that captures many aspects of autonomy,and then be devoted to presents a framework for modeling autonomy of agent.This paper reviews previous discussions of agent autonomy and presents a core definition synthesized from these discussions.This definition identifies three abilities:ability to decide future actions,ability to reason and ability to update mental state,which are basic capalities that an autonomous agent must have.Based on the definition of agent,a formal framework for describing autonomous agent and its mental state is presented.In this framework,four dimensions of an autonomous agent-representation of its mental state,ability to reason,ability to decide the future actions,ability to revise its mental state-are provided.An agent description language is introduced to represent the mental states of agent.By using reasoning complexity,a criterion for measuring the level of autonomy is put forward.Various architectures for different types of agents,such as simple agent,deliberate agent and planning agent,are classified and constructed by means of varying the ability of action decision and keeping the other components constant.These work supplies a necessary theoretical basis for implementing autonomous agent.Although our approach is based on the certain kind of representation of mental state,the resulting framework can serve as a flexible platform for hierarchizing

关 键 词：自主agent AGENT理论 心智状态 情境演算 人工智能 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论] TP18[自动化与计算机技术—计算机科学与技术]