Coordinating Activity in Knowledge-Intensive Dynamic Systems

In this paper we consider the problem of coordinating activity in knowledge-intensive dynamic systems (HDS) -large-scale, multi-agent systems in which agents with significant individual capabilities work together 10 accomplish complex, knowledge-intensive tasks. Given the time pressure and resource limitations under which a KIDS typically operates, the establishment of plans and schedules can significantly improve organizational peiformance. However, there are several complicating factors: (5, there is diversity and novelty in the structure of processes that must be executed over time, requiring tight coupling of action selection with resource allocation, (2) processes are unpredictable in their outcomes and require continual dynamic adjustment and revision. (3) the collective capabilities of the agents of a KIDS are its primary asset and, to minimize future resource limitations. task allocation should consider the side effects of acquired expertise, and (4j KIDS are large-scale enterprises, requiring an ability to eflectively distribute decisionmaking. Our previous research has developed constraintbased search techniques for continuous, dynamic scheduling that have been successfully applied to complex, large-scale transporration and manufacturing domains. We outline current work aimed at extending these models to address the above issues and provide an eflective basis for managing KIDS. 1. INTRVDUCTIVN The establishment of a schedule is crucial to effective management and control of KIDS; it is the means by which global coordination of executing agents (actors) is achieved and maintained. Thus, as evolving requirements, changing priorities, and new resource availability constraints continually force changes to previously planned tasks and resource assignments, attention must be paid to the cost of continual solution change and to maintaining continuity in currently executing processes. Like many other practical domains, KIDS are further complicated by the presence of multiple performance objectives, the need to optimize under complex (and often idiosyncratic) constraints and the need for flexible human decisionmaking involvement. KIDS are also systems whose components build expertise as processes are performed, and hence constituent resource capabilities evolve over time. Finally, KIDS are large-scale enterprises. The scheduling requirements inherent in the coordination of KIDS are at odds with classical scheduling models and most current scheduling tools, which are conceived and designed as batch-oriented, black box solution generators. Most basically, this type of approach advocates a mode of operation wherein the scheduling problem is (re)solved from scratch each time there is a change to any problem constraint or requirement. This presents significant problems for use in dynamic and continuous scheduling environments, where such changes are the order of the day. First, there is generally no persistence of decisions over time. Small changes in inputs can lead to large changes in outputs, and it is quite difficult to control and manage solution change. Second. a resolve-from-scratch Daradiem Dresents obvious . Knowledge-intensive dynamic systems (KIDS) present computational barriers to integrating with execution; in complex scheduling and coordination problems. By KIDS, large-scale domains it becomes problematic for the scheduler we refer generally to large-scale, milti-actor systems in which agents with significant individual capabilities work together to accomplish complex, knowledge-intensive tasks. Examples of KIDS include large news organizations such as CNh' and AP, the research and development division of a large company, and the World Bank. KIDS plan and execute production processes which: (1) principally involve the collection, manipulation and management of knowledge products, (2) exhibit diversity and perpetual novelty in process structure, and (3) are unpredictable in their execution and require continual adjustment and revision. KIMAS 2003. Octaoer 1-3.2003, Boston. MA, USA. Copyright 0-7803-7958.61031S17 00