Cooperative Intelligent Software Agents

The availability of network-based information sources and services, along with the wide use of of World Wide Web, presents a great opportunity for enhanced problem solving support that incorporates information gathering into the problem solving framework. The concept of an intelligent information agent has been explored for information accessing and ltering tasks (e.g., ltering of newsgroup information). To provide support for more complex tasks that involve information gathering along with decision making capabilities, an agent must communicate and cooperate with other agents. One of the main issues is how to structure a multi-agent architecture that will allow access, ltering and fusing of information from many sources and services, integrated with exible decision support. In this paper we present such an architecture. It has two layers: The top layer consists of taskspeci c software agents which help users perform tasks by formulating problem solving plans and carrying out these plans through querying and exchanging information with other software agents. The bottom layer consists of information-speci c agents which provide intelligent access to a heterogeneous collection of information sources. The layered architecture has been implemented in the domain of everyday organizational tasks (e.g., hosting visitors, nding information about persons on the internet, managing personal calendars) at Carnegie Mellon. We illustrate this architecture and the agent coordination protocols in the hosting visitor domain. ii