From the walls of caves to the pages of books, human knowledge has found its way onto discrete surfaces as mostly printed matter. For the most part, the Internet has merely extended this display, albeit with added dimensions such as multimedia, virtual worlds, and hyperlinks—a "Version 0.8 (beta)" of Vannevar Bush's memex. 1
As the network transforms our work and facilitates our interactions with others, human knowledge is increasingly authored in the network and augmented by it. Researchers in business-process engineering, workflow management, and even architecture and interior design have long realized that organizational structures significantly influence workplace dynamics. The network is now the most flexible of structures, capable of adapting instantaneously to different projects and needs—allowing human knowledge to be connected and augmented in astoundingly complex ways.
Knowledge networking has been the subject of numerous research programs (Grid Computing, Knowledge and Distributed Intelligence, and others). The fundamental idea behind this research is the need for substantially new ways of thinking and working to leverage the Internet revolution in the advancement of human knowledge. We need new tools for capturing human knowledge as it evolves over the network, as well as tools for indexing and retrieving it for later reuse.
Consider the problem of capturing and reusing design rationale in the design of a large software system, where component development is outsourced, system integrators create custom solutions to enable code interoperability, and so on. Where does the reasoning behind design decisions and implementation solutions go? It exists in the ether of the interaction webs created by e-mail and collaborative work tools, in the software code and documentation, and in the minds of the developers and managers involved in the project. Our technologies and tools for capturing and reusing this knowledge are in their infancy, and draw on ideas from disciplines as diverse as distributed AI, computer-supported collaborative work (CSCW), psychology, and information science.
Knowledge networks are networks of people and computer systems that work on collaborative, knowledge-intensive tasks in a distributed—often virtual—enterprise. These networks will ensure that knowledge becomes an integral part of business practices, readily accessible for the task at hand—either proactively or passively. Knowledge needs to be gathered, filtered, and kept up-to-date. It must also be protected from unauthorized access.
From a business perspective, knowledge is pretty useless unless it affects action. If your decision is independent of a piece of information, there is no point in collecting the information. Hence, several approaches tie access to knowledge to the tasks of a business process (for example, see Maurer and Holz 2
and Kühn and Abecker 3
). The major problem here is how to make sure that these knowledge-enriched process descriptions evolve according to the changing business environment: Maintenance is the big issue.
Knowledge networks utilize the Web as their basic means of communication. Distributed or even virtual enterprises can integrate their business processes and exchange data over the Internet. The Web makes knowledge accessible to humans. They can use a browser to read information that is made available elsewhere by a knowledge provider. As a knowledge seeker, the core problem is how to find the few diamonds for a task in the vast rough of information that is accessible on the Web. For finding information, we currently rely on tips, portals, and search engines:
• The fastest way to find something on the Web is by knowing where to look. The best tips often come from colleagues and friends who work in the same area, have the same interests, or otherwise simply understand the context of your investigation. Unfortunately, the tipsters may not be around when you need them the most.
• Portals are the Web facade of an old idea: they define a context and gather information that is relevant to it. If you are interested in a specific topic, you go to a portal that covers it, just as you buy a journal that addresses the topic.
• A search engine indexes the Web and uses the index to answer a knowledge seeker's keyword queries. In principle, there are two ways to create an index: automatically by using information-retrieval techniques to scan text or manually by having an editor read it. Both approaches have their limitations. The state of the art in automated text understanding limits the accuracy of automated approaches; they also have problems in determining relevancy. The manual approach, on the other hand, is costly and difficult to pace with the evolution of the Web. So far, no search engine is able to index the whole Web.
How can we improve the precision (all information found for a query is useful) and recall (all useful information is found) of knowledge on the Web? One way is to use a knowledge representation language to annotate the information available and use an inference engine to overcome the limitations of keyword annotations. This would allow search engines to formally reason about the meaning of a specific page (that is, determine its semantics).
The article by Decker et al. ("The Semantic Web: The Roles of XML and RDF") in this issue describes one possible approach toward a semantic Web. They use RDF to represent ontological knowledge. (For other approaches that are based, for example, on XML representations, see the special issue of IEEE Intelligent Systems
on Knowledge Management and the Internet. 4
Using declarative AI-like representations and general-purpose reasoning engines will not answer all the questions regarding support for knowledge networks.
First, there is the question of effort. Who annotates all the Web pages with formal knowledge representations, who maintains it, and who makes sure that all the knowledge represented stays compatible so that you can post a query to a "semantic" search engine and actually get back results from several different Web sites?
Second, will these representations really improve precision and recall on an Internet scale? So far, there is no empirical data to answer this question.
Third, is the future of knowledge networks to provide services instead of information? Objects may exist on the Web and offer their standardized services to interested parties. Instead of defining semantics declaratively, a service defines semantics operationally. There are several trends in this direction, such as application servers based on component models like Enterprise JavaBeans and Microsoft COM, or the CORBA approach.
The article by Lu and Cai ("STARS: A Socio-Technical Framework for Integrating Design Knowledge over the Internet") shows what can be built on top of a service model. It examines knowledge networking in the context of distributed engineering design and manufacturing. Traditional design environments have focused on tools, integrating data so different tools can work together. Cai and Lu propose creating tools that address how people work together to resolve conflicts and move toward a final product. They call this a socio-technical framework for design and introduce us to some ideas that cross engineering, computing, and social science.
The theme articles in this issue of IEEE Internet Computing touch on some recent work in knowledge networking. We hope readers find them useful in understanding some of the fundamental research issues and enabling technologies by which we can all extend our own personal and professional knowledge networks.
is an associate professor at the University of Calgary, Canada, and co-director of the Alberta Software Engineering Research Consortium (ASERC). His research interests are in e-business engineering, experience management, and software process support. Maurer received a PhD from the University of Kaiserslautern, Germany. He is a member of the IEEE, the ACM, and the editorial board of IEEE Internet Computing
. More information on his work is available online at http://sern.ucalgary.ca/~maurer/.
is an assistant professor of computer science at Drexel University, Philadelphia, and director of Drexel's Geometric and Intelligent Computing Laboratory. His research interests are in artificial intelligence and engineering design. Regli received a PhD from the University of Maryland at College Park. He is the recipient of a 1998 NSF Career Award and a member of the editorial board of IEEE Internet Computing
. More information on his work is available online at http://www.mcs.drexel.edu/~regli.