"A study of history shows that civilizations that abandon the quest for knowledge are doomed to disintegration."
-Bernard Lovell: The Observer, 'Sayings of the Week' ,14 May 1972

Combining traditional notions of libraries with contemporary technological capabilities (such as the WWW) is a meeting of dissimilars. Libraries have traditionally stressed service, organization, and centralization. The WWW has embodied flexibility, rapid evolution, and decentralization. Digital libraries somehow need to bring these together.

Much digital library work has begun from the centralized, structured view of a library and sought to provide access to the library through digital means. In the University of Michigan Digital Library Project (UMDL) we believe that this approach loses the advantages of decentralization (geographic, administrative), rapid evolution, and flexibility that are hallmarks of the web. In UMDL, we are instead embracing the open, evolving, decentralized advantages of the web and introducing computational mechanisms to temper its inherent chaos. However, we are also embracing the traditional values of service, organization, and access that have made libraries powerful intellectual institutions.

The challenges we face are providing an infrastructure that lets patrons (and publishers) feel like they are working within a library, with the traditional emphasis on providing service and organized content, when in fact the underlying space of goods and services is volatile, administratively decentralized, and constantly evolving. Moreover, the decentralized and flexible infrastructure can be exploited to allow information goods and services to evolve in a much more rapid, diverse, and opportunistic way than was ever possible in traditional libraries, for the good of consumers and providers.

In the UMDL we are meeting these challenges by defining and incrementally developing interfaces and infrastructures for users and providers such that intellectual work (finding, creating, and disseminating knowledge) is embedded in a persistent, structured context even though the underlying networked system is evolving. The infrastructure supports extensible ontologies (meta descriptions of collections and services) for allowing components in the digital library to self-organize, dynamically teaming to form structures and services that users need. Principles from economics are also being used to efficiently allocate resources and provide incentives for continual improvement to networked goods and services. This approach enables third parties to join or use UMDL technologies to define and manipulate agents, facilities, and ontologies so that the web of resources grows in an orderly but decentralized way.

The core of the UMDL has been the agent architecture that supports the teaming of agents to provide complex services by combining limited individual capabilities. In the early stages of the project, the architecture was defined and has been in use for some time now. Our ongoing efforts have been to deploy the UMDL in real-world settings, which has stressed the need for advanced user interfaces and on deliberately populating the agent architecture with a diverse set of services. Despite its open nature, the UMDL can already support inquiry by user communities (high school classes) that are very reliant on service and structure. The UMDL testbed is being used to support authentic "inquiry-based" approaches to science education in middle and high schools.

Our continuing objective is to show also that the inherently decentralized information economy that the UMDL architecture encourages will lead to capabilities beyond the reach of centralized approaches. Our work has already made measurable progress on the frontiers of theories and algorithms for pieces of this overall vision: economic mechanisms for distributed resource allocation; algorithms for determining appropriate offers for goods and services; learning methods for allowing computational agents to evolve with the evolving agent population; ontological search methods for discovering new relationships among separately-developed components of the library, and so on. We are working on bringing all of these pieces together, so that their collective advantages can be exploited and measured, to demonstrate the promise of the UMDL architecture and to satisfy the needs of information consumers and providers.

Subject matter/content:

The content will emphasize a diverse collection, focused on earth and space sciences, which can satisfy the needs of many different types of users. The content will be supplied by publishers, although the project will eventually allow all users to publish their work. A related project, the Journal Storage Project (JSTOR), will digitize and make available all issues from the first publication through 1990 of ten economics journals to the NSF-UMDL.

Testbed description:

The UMDL will consider a complex array of technical and socioeconomic issues and will focus the research through the design, construction, and evaluation by real users of a testbed system. The testbed will consist of a cooperating set of three types of software agents: user interface agents, mediation agents, and collection agents.

User interface agents will conduct interviews with users to establish their needs such as what they need to know, and the breadth and depth of the information they require. The interface agent will also enable the user to specify areas of interest so that the system can notify the user of items of potential relevance.

Mediation agents will coordinate searches of many distinct but networked collections by taking orders from the interface agents. This will allow the user to search many libraries simultaneously in ways that meet time, relevancy, and economic constraints. The mediation agents will depend upon a conspectus that describes the contents of the various collections on the network.

Collection interface agents are associated with each specific collection and can handle searching within specific collections of text, images, graphics, audio and video. Information held in the collections may be owned by various entities, some of which may demand some control over dissemination of contents or compensation for access to their copyrighted material. The system design will provide mechanisms to protect information access and support remuneration operations.

The users of the digital library testbed will include expert researchers, students, and the general public. The library will include media types ranging from page images to interactive, compound documents and eventually real-time interaction with data. Critical to the exploitation of these resources will be ongoing programs of evaluation, training, user assistance, and outreach.

Prior Results:

Investigators associated with the NSF-UMDL have expertise in many areas related to the project including distributed architecture agents, economic models, information retrieval, user interface design, and deployment and evaluation of learner centered software tools. The collaborative effort among researchers draws upon the experience and expertise of each individual. Initial prototypes of the testbed will be based on the TULIP system, an electronic searching and browsing interface to 40 materials science journals published by Elsevier Science.