Social Laws for Artificial Agent Societies

WPCK 2 B P Z Courier 10cpi#|x K x 6 X@8；X@HP LaserJet IIISi HPLASIII.PRS x@，\，，7|X@#|x 2 Z B#X F`HP LaserJet IIISi HPLASIII.PRS x@，\，，7|X@Courier 10cpi Courier 10cpi(粗体)？X，宽x 6 X@8；X@？这是五年持续奖的第一年资助。这一搜索发展了一种关于计算e 2 i 9220645 Shoham``h h#(-p p 27？这是三年持续奖的第一年资助。这项研究发展了一种关于社会法则在计算环境中的效用的理论，即保证多个程序和程序员成功共存的法则。它的重点是协调计算机代理的活动，但调查将由人类也参与的社会中的协调指导和影响。在文献中已经讨论了许多活动协调的方法；本研究集中在一种方法上，尽管这种方法非常具有常识性，但在计算机科学界相对较少受到关注。粗略地说，方法如下。社会将采用一套社会法律；每个代理人将被要求遵守这些法律，并将能够假设所有其他人也将遵守这些法律。这些法律一方面会限制代理人可用的计划，但另一方面会保证其他代理人的某些行为。移动机器人的交通法规领域就说明了这一点：这种方法不是安装一个集中控制器(当今机器人学中最常见的方法)，也不是让机器人为避免碰撞而不断谈判(其精神代表了分布式人工智能的许多工作)，而是让机器人遵守交通法规，如“保持在道路的右侧”。同样，尽管这种方法很有常识性，但它并不是以系统的方式开发的；正如我们将看到的，它提出了一些非常重要的问题。拟议研究的一部分涉及按照上一段所述路线进行的社会法律的离线设计。另一部分与社会惯例的在线演变有关：即使我们作为设计师无法提前强加社会规律，我们也会希望以一种鼓励惯例随时间演变的方式来设计代理。为了实现这些目标，提出了如下建议：(1)定义多智能体系统的精确模型，其中明确地表示社会规律；(2)定义并研究有用社会规律的离线综合所涉及的基本计算问题；(3)在特定应用领域的框架内实现人工智能体社会的社会规律的思想，同时仍然证明这些规律在这些领域中的有用性的分析结果；(4)在计算框架中研究社会习俗的演变。更具体地说，了解不同的参数(如记忆和沟通)如何影响惯例演变的效率；以及(5)调查社会结构(如权力等级)和社会法之间的相互作用。H)0*0*0*

WPCK 2 B P Z Courier 10cpi #| x K x 6 X @ 8 ; X @ HP LaserJet IIISi HPLASIII.PRS x @ , \ , , 7|X @ #| x 2 Z B # X F ` HP LaserJet IIISi HPLASIII.PRS x @ , \ , , 7|X @Courier 10cpi Courier 10cpi (Bold) ? x x x , w x 6 X @ 8 ; X @ ? x x x , x ` B ; X This is the first year funding of a five year continuing award. Thi search develops a theory regarding the utility of social laws in a computational e 2 I 9220645 Shoham ` ` h h # ( - p p 2 7 ? X This is the first year funding of a three year continuing award. This research develops a theory regarding the utility of social laws in a computational environment, laws which guarantee successful coexistence of multiple programs and programmers. It focuses on the coordination of activities of computer agents, but the investigations will be guided by, and impact, coordination in societies which involve humans as well. A number of approaches to activity coordination have been discussed in the literature; this research concentrates on one approach which, although very commonsensical, has received relatively little attention in the computer science community. Roughly speaking, the approach is as follows. The society will adopt a set of social laws; each agent will be required to obey these laws, and will be able to assume that all others will as well. These laws will on the one hand constrain the plans available to the agent, but on the other hand will guarantee certain behaviors on the part of other agents. The domain of traffic laws for mobile robots illustrates the point: Rather than install a centralized controller (the most common approach in robotics today), or have the robots continually negotiate in order to avoid collisions (the spirit of which is representative of much work in distributed AI), this approach will have robots obey traffic laws such as "keep to the right of the road". Again, although very commonsensical, this approach has not been developed in a systematic fashion; as we shall see, it poses some highly nontrivial problems. One part of the proposed research is concerned with the off line design of social laws, along the lines mentioned in the previous paragraph. Another part is concerned with the on line evolution of social conventions: even when we as designers are not in a position to impose social laws ahead of time, we will want to design the agents in a way which will encourage the evolution of conventions over time. In service of these goals, the following is proposed: (1) define precise models for multi agent systems, in which social laws are explicitly represented; (2) define and investigate the basic computational problems involved in off line synthesis of useful social laws; (3) implement the idea of social laws for artificial agent societies in the framework of particular domains of application, while still proving analytical results about the usefulness of these laws in those domains; (4) investigate the evolution of social conventions in a computational framework. More specifically, understand how different parameters (such as memory and communication) affect the efficiency of convention evolution; and (5) investigate the interaction between social structure (such as authority hierarchies) and social laws. h) 0*0*0*