Vision

Modern technologies and practices are revolutionizing the way we gather and handle information. There is great potential for both science and society to build upon the fruits of this revolution, to magnify our ability to understand and manage ever-larger and more complex systems such as those in natural, social, material, financial, and manufacturing spheres. The vision of the Knowledge and Distributed Intelligence (KDI) initiative is to achieve, across teh scientific community, the next generation of human capability to:

• Generate or gather, and represent more complex and cross-disciplinary scientific data and information, from new sources and at enormously larger scales;

• Tranform this information into knowledge by combining, classifying, and analying it in new ways;

• Collaborate in groups and organizations, sharing this knowledge and working together interactively across space, time, disciplines, and scientific cultures to multiply results.

With this approach, KDI aims to foster distributed intelligence as a foundation for advancing all areas of science and engineering research and education, and to spin off important new results and technologies with positive benefit to society at large. KDI will support these advances by enabling more highly interactive, multidisciplinary, and collaborative paths to innovation. To achieve these aims will require major advances in collaboration processes and infrastructures, computational and representational capabilities, and our understanding of socio-technical interactions.

Challenges And Strategies

• Generates greater understanding of phenomena of distributed intelligence and collective behavior in human, automated, and natural systems;

• Creates the next generation of mathematical, computational, data-oriented, and organizational methods and infrastructure, which will exploit multidisciplinary distributed intelligence to advance science and engineering;

• Enhances human ability to create and use knowledge in groups, organizations, and communities through advances in human infrastructure, technology, and education.

As the primary governmental body that funds research in all of the relevant domains of knowledge, as well as in education, the NSF is the most appropriate agency to spearhead this initiative. NSF- funded research has already produced a body of knowledge in networking, computational approaches to complex problems, collaboration technology, socio-technical systems and impacts, learning, and education that provides a basis of community capability and infrastructure for addressing KDI aims.

This initiative is an intellectual focus for collaborative, multidisciplinary thinking on three complementary aspects of knowledge and distributed intelligence:

• increasing interaction, knowledge/tool integration, collaboration, and understanding within communities and across disciplines through Knowledge Networking (KN);

• extending the power of tools, models, and simulations to represent and manage complex systems through New Challenges in Computation (NCC); and

• extending our ability to learn and create through Learning and Intelligent Systems (LIS).

Learning and Intelligent Systems (LIS) seeks to stimulate interdisciplinary research that will unify experimentally and theoretically derived concepts related to learning and intelligent systems, and that will promote the use and development of information technologies in learning across a wide variety of fields. The LIS initiative focuses on fundamental scientific and technological research undertaken in the rigorous and disciplined manner characteristic of NSF-supported endeavors. The initiative will ultimately have a major impact on enhancing and supporting human intellectual and creative potential. Development of new scientific knowledge on learning and intelligent systems and its creative application to education and to learning technologies are integral parts of the initiative.

Knowledge Networking (KN) focuses on the integration of knowledge from different sources and domains across space and time. Modern computing and communications systems provide the infrastructure to send bits anywhere, anytime in mass quantitiesÑradical connectivity. But connectivity alone cannot assure (1) useful communication across disciplines, languages, cultures; (2) appropriate processing and integration of knowledge from different sources, domains, and non- text media; (3) efficacious activity and arrangements for teams, organizations, classrooms, or communities, working together over distance and time. In short, we have connectivity, but not interactivity and integration. KN research aims to move beyond connectivity to achieve new levels of interactivity and to deepen our understanding of the ethical, legal, and social implications of new types of interactivity, so as to increase the semantic bandwidth, knowledge bandwidth, activity bandwidth, and cultural bandwidth among people, organizations, and communities.

New Challenges in Computation (NCC) focuses on research and tools needed to model, simulate, analyze, display, or understand complicated phenomena, to control resources and deal with massive volumes of data in real time, and to predict the behavior of complex systems. Phenomena, data, and systems of interest so exceed in scope, multiplicity of scale, and dimensionality what can be handled by present techniques that incremental advances do not suffice. New computational schemas, such as quantum computing or biomimetic computing, are needed. Models, tools, and resources must be shared among many researchers in different places. Moreover, a key need is immediacyÑcontrol of networked resources in real time and tailored to people's needs and capabilities, Òon-the-flyÓ analysis of data to guide experiments or manage situations as they happen. These featuresÑscope, multiple scales, dimensionality, shared use, and immediacyÑdistinguish NCC from the earlier work on which it builds. NCC aims to enable collective understanding and effective management of complex systems. These aims will require major advances in hardware and software to handle complexity, representation, and scale, to enable distributed collaboration, and to facilitate real-time interactions and control.

Examples of Possible KDI Research Areas

• Systems as dissimilar as an economic market, the brain, behavioral norms in a society, large computer networks, and the scientific enterprise have this in common: information is widely distributed throughout the system; no identifiable entity within or outside the system coordinates information or makes decisions; no particular locus of coordination is evident (although many markets have a "place," many others do not); yet, information is coordinated and focused into sensible outcomes. Economists, neuroscientists, sociologists, and philosophers of science -- in frequent collaboration with mathematicians, computer scientists and others -- now study these systems, but separately. Collaboration may reveal important similarities in how these disparate systems produce knowledge from their distributed intelligence, and hence insights into improving system performance.

• Understanding the behavior of a system that emerges from collective interactions of its relatively simpler components is a central challenge. Typical problems include determining fluid flows in reservoirs at the level of the reservoir field from a knowledge of flows at the level of pores in the ground, predicting macroscopic properties of materials from a knowledge of atomistic properties, understanding the regulation of interacting metabolic pathways, and studying the interplay of natural and anthropogenic factors in issues of environment and biodiversity. The simulations needed to study these systems will use substantial computational resources that are just becoming available. The results of these simulations will provide an increased understanding of the development of emergent behaviors that will lead to better models of physical and social systems.

• Present abilities to manage experiments or numerical simulations in real time are inadequate. Advances in these capabilities, and the new paradigms of investigation that will follow, will open enormous opportunities to expand our understanding of complex systems. For instance, "on- the-fly" analysis of observations of a living cell via light microscopy can direct attention to areas of interest while the specimen remains viable. Similarly, in a numerical simulation, analysis of the simulation behavior as the simulation progresses, coupled with the appropriate invasive software, would permit changing parameters of the simulation during the calculation, thus providing greater insight in less time. Finally, enabling the dynamic interplay between observations of the behavior of a complex system and simulations of its behavior could dramatically improve the understanding and even the control of the system. This has wide and important consequences, for example in predicting weather, controlling communication and power distribution systems, and managing medical treatment regimens.

• Managing large datasets has become a critical task in every area of science and engineering. In astronomy and physics, sifting enormous volumes of data for indications of a few rare events is a key step in observations of both galaxies and subatomic particles. In geosciences and biology, effective ways to combine different datasets and manipulate elements of the sets are critically needed. The next three examples show different aspects of these issues. More generally, KDI requires advances in the interoperation, use, and construction of large distributed repositories by participants including people, artificial agents, groups, and organizations. Interactivity research studies how to build and maintain dynamic, content-rich, multi-media relationships among participants, instruments, tools, and data. Representation research studies how knowledge about processes and phenomena can be encoded, and how meanings for representations are reconstructed in their contexts of use. Cognition research investigates perception, reasoning, memory, learning, and action by participants, including groups and organizations. Studies of agents investigate the active and sometimes physically embodied algorithms, software, communications, and tools that can assist people in collaborating and networking knowledge. Finally, investigations of corpora study the entire lifecycle (creation, structuring, storage, maintenance, use and disposal) of general and community-specific knowledge collections, including ad hoc data collections, complex scientific databases, digital libraries and even such unconventional entities as digital forms of artifacts in museums. These five areas of research are critical enablers of KDI: people's ability to collaboratively access, retrieve and comprehend information from complex databases and distributed sources depends on how that information is created, structured, stored, presented, reasoned about, and managed.

• Geospatial data and information are vital components of science, constituting the link between scientific research results and their use by planners and decision-makers. New technologies (satellite and airborne sensors, automated and remotely-operated sampling stations) produce data that have generated an explosion of information. At the same time, our understanding of the biological, chemical, geological, physical, and social processes that operate in and on the environment has reached the stage where multidisciplinary integration offers enormous promise for progress. Full and functional interoperability of geographic information (GIS) systems is necessary for such multidisciplinary research. To achieve this will require basic research in the representation of the semantic content of GIS data, together with interagency and international standards activities. Once an interoperable distributed global GIS is available several scientific opportunites arise, such as the investigation through data mining of anomalous environmental phenomena (e.g., the Ozone Hole), and the possiblity of incorporating real-time data to understand and thus predict storms and similar events. Such capabilities could also be used in developing information systems for disaster relief.

• Taxonomists in biology are building a Web-linked network to share interactive specimen databases, programs for identification keys, image-analysis software, and links to phylogenetic trees, accessed through "expert workstations," to provide automated image-based means to identify specimens. Neural networks employ ÒlearningÓ rules to speed up identification after initial specimen presentations, while tomographic scanning and holographic images facilitate analysis of complex specimens such as bones and shells. The challenge is managing and coordinating the amount and complexity of information generated by such imaging technologies. However, the result will be rapid, accurate, reliable taxonomic identification, from microscopic to macroscopic scales, which will be invaluable to systematists who currently are limited to working with single taxonomic groups, at constrained scales of measurement, using information from limited databases. Other beneficiaries include agricultural extension agents encountering a new weed, customs officials interdicting imports of new biological materials, and geologists searching marine cores for fossils indicative of petroleum deposits, all of whom need quick and accurate taxonomic identification to do their jobs.

• Genomics is the analysis of DNA sequences resulting from the sequencing of all of the genetic material of selected bacteria, fungi, plants, and animals including Man. These enormous sets of sequence data contain the history of the evolution of life. The ability to extract meaningful and useful information from these datasets offers the keys to such wide applications as engineering micro-organisms to clean up contaminated sites, designing food crops to withstand drought and pests, and curing dreaded human diseases. Fundamental problems in this area include comparing sequences for similar genes and gene products from different organisms to study gene evolution, predicting the three-dimensional structure of proteins from the linear DNA sequences of the genes encoding those proteins, and ultimately predicting how the proteins might interact with each other to produce a differentiated cell. The size of datasets of DNA sequences and molecular structural information has grown explosively in the past few years; the need for new tools to understand the significance of those data has increased correspondingly. As the cost of sequencing DNA drops by an order of magnitude during the next five years, the amount of information to be analyzed will increase even faster than its current doubling time of a little over a year. To manage and make the best use of these resources will require unprecedented levels of collaboration among scientists in biology, computer science, mathematics, statistics, and computer and software engineering.

• Collaborations and sharing of resources between people in different places, especially in real time, are not yet effective enough. Advances in simulating or controlling complex systems depend on major advances in computing systems, both hardware and software. Conventional computing systems do not yet deliver the required performance; thus alternative schema such as systems constructed at the molecular level using quantum physics or biological components must be investigated. Similarly, the effective use of advanced computing systems, conventional or otherwise, will require the scaling of software and tools to levels of complexity and interactivity that are well beyond the usual evolutionary path of today's systems.

• Both the brain and its component neurons can learn and compute. Exactly how the massive networks of neurons in a brain perform computational tasks to produce behavior is not known. However, it is well established that the brain transmits information through distributed, parallel channels. At a basic level, elucidation of the computing processes used by neurons may provide insight into alternatives to or alternative designs for computer chips. At a higher level, understanding the way in which the distributed networks of neurons in the brain process information may reveal new paradigms, architectures, and approaches for solving complex computational problems by means other than those currently used by computers and supercomputers.

• How intelligent systems learn to develop strategies, to select among alternative plans, and to predict the outcome of complex performances is poorly understood. Engineers, mathematicians, computer scientists, psychologists, neuroscientists, and educators now have pieces of this puzzle. Putting the pieces together is the next challenge. Developing models and technologies that learn to maximize externally specified measures of performance in complex situations -- situations that require design, "foresight," or "planning," to optimize performance -- would increase understanding of this complex behavior and improve assessment of different learning methods in human and artificial intelligent systems.

• In both human learning and artificial intelligent systems, the management of information is a major practical concern. To provide a scientific basis for addressing this concern, a deeper and more multidisciplinary understanding of selective attention and memory management in learning systems is essential. The fundamental insights are at hand: Biological systems have extensive mechanisms for selecting and weighting information; statistical decision theory and filtering have produced methods to extract system parameters from uncertain and noisy measurements. By combining these and other insights, comprehensive selection designs and information management tools could be developed and tested for their engineering and educational value.