"Architecture in general is frozen music."
-Friedrich von Schelling

A library serves a community of users by making available information content and services that are valued by that community. A traditional, physical library is thus not simply a building that houses information, but rather a complex configuration of information goods and services that have been carefully selected and organized around the needs of a user community.

In a digital library, the content and services are electronically available, and user communities are no longer geographically defined. Realizing a digital library therefore includes difficulties in digitizing contents, computerizing services, and networking together users. Even if these difficulties are overcome, however, the result can well be an overwhelming tangle of possible information sources without the structure and selectivity that renders a traditional library navigable. In other words, if the administration of a traditional library is challenging, the administration of a digital library can border on impossible due to the magnitude of content and services available, the rate of change in what is available, and the size and evolution of a user population that is not bounded by physical proximity.

One answer to this challenge is to rely on traditional methods that put administrators at the center of the enterprise, to attract, register, and track a user community, to seek and include the content that will benefit the community, and to provide the most valuable services for tasks, such as organizing, searching, abstracting, and deseminating the content. An alternative approach, however, is to move as much of the administration into the digital infrastructure as possible. The goal of this approach is to provide mechanisms by which a digital library can continually reconfigure itself as users, contents, and services come and go. These mechanisms should encourage:

• Flexibility: They should be able to embody a wide variety of policies to realize different flavors of libraries (public, corporate, university, personal,...)
• Extensibility: Providers and consumers of information goods and services should have incentives to join the library and abilities to find their counterparts.
• Scalability: As the plethora of users, goods, and services grows, the underlying, computerized administration of the library should not bog down.

Toward this end, the University of Michigan Digital Library (UMDL) is structured as a collection of agents that can buy and sell services from each other (or the user) using our commerce and communications infrastructure. While one of the emphases of the UMDL is to provide a working testbed to improve secondary education, a second emphasis is on the definition and design of the infrastructure, and the kinds of agents that exist in it, that allow decentralized (scalable) ongoing configuration of an extensible set of users and services. We refer to the services/protocols offered by this infrastructure as the Service Market Society (SMS).

The SMS requires the integration of numerous agent technologies for knowledge exchange, commerce, learning, and modeling. In our research on SMS, we are investigating how we to bring these technologies together to create a prototype system in which a changing population of agents can find each other, enlist each other's aid (for a price), decide on the terms of an interaction, and learn to differentiate among providers. We use our prototype to demonstrate how these technologies contribute to providing a flexible, extensible, and scalable digital library.

Recently, we have concentrated on developing technologies that, for example, manipulate ontological descriptions of the elements of a digital library to help agents find and categories services and auctions for exchanging goods and services under various conditions. These technologies allow flexibility in the UMDL configuration policies, extensibility in what can be bought/sold in the SMS, and scalability by using demand (as represented by price) as incentive for replicating services. Robust performance in such a society, however, also requires that participating agents make informed buy/sell offers for the goods and services, and that they are able to recognize and learn from whether another party keeps its end of the bargain. Therefore, we have also developed specially suited learning and bargining methods.

If you would like to get a first-hand glimpse of our work, please take a look at our Ontology-Based Metadata demo.