transmission. The use of multiuser dialogue is likely to extremely useful if there is usable software available, such as in the case of scientific research and engineering design.7

2.

Machine sensors for HCI and general I/O to facilitate telepresence and teleoperation. Current computers are ''sensory deprived," representing a barrier to both human-computer interaction and machine learning. Such sensors as microphones in single and array configurations, infrared and other means of scanning and computing distances, optical sensors at lightwave range including charge-coupled device (CCD) cameras of small size, haptic interfaces, and alternatives to click-and-point devices should be studied. Fusing sensor inputs to the computer with intelligent or learned action-response behavior would create a more realistic approach to machine learning and complex inferencing techniques, involving symbolic, fuzzy, and probabilistic approaches. This area has been researched with different objectives, but seldom with that of trying to improve the human-computer interface. Standardizing environments (e.g., via a human-computer interaction workbench; HCI-WB) can improve measurements. Such an experimental environment is also useful in the study of human behavior in real and virtual modalities related to the NII, and provides comparisons in human subject variabilities between real and virtual environment behavior, navigation, and orientation. The potential for research in the fusion of the modalities is enormous.8 The challenge of this research area is to fuse multiple sensor inputs to the computer in a cohesive and well-coordinated manner. One such example would be the integration of a CCD camera input with a haptic experiment using force feedback and synthesized video output. Another helpful experiment could involve mechanisms for the localization of sound in virtual environments9 using the HCI-WB.

3.

Large storage (archival and nonarchival), database, and indexing technologies, including multiresolution and compression for different modalities. Video and audio technologies will require large compression factors and mechanisms for rapid encoding and decoding and are difficult to index and access for retrieval, and even then, mass storage database techniques will be required. This area is also indirectly related to the speech and video synthesis technologies, since high-resolution synthesis approaches imply efficient encoding, possibly at different resolution levels. Similarly, virtual environment research requires efficient storage and compression technologies for input and output. There are good reasons to believe, for example, that high-quality audio can be encoded at rates of 2,000 bps using dynamic adaptation of perceptual criteria in coding and articulatory modeling of the speech signal. Therefore, encoding research should include both generation and perceptual factors.10 Additionally, multimedia databases require techniques for providing temporal modeling and delivery capabilities. A novel interface, called "query scripts," between the client and the database system adds temporal presentation modeling capabilities to queries. Query scripts capture multimedia objects and their temporal relationships for presentation. Thus, query scripts extend the database system's ability to select and define a of set objects for retrieval and delivery by providing a priori knowledge about client requests. This information allows the database system to schedule optimal access plans for delivering multimedia content objects to clients. One more example of an area of concern related to the overall throughput capability of the NII is the Earth Observing System (EOS) of NASA. This system is coupled with a data information system (DIS) in a composite EOSDIS, which is expected, when operational in 1998, to require transport of one terabyte per day of unprocessed data and possibly an order of magnitude more when processed, roughly equivalent to the total daily transport capacity of the current Internet. The question is, Will the NII provide the capacity for even a fraction of such volumes of data?

4.

Virtual environments and their use in networking and wireless communication (tethered and untethered) networked environments11 will have an impact on the NII. Virtual environments relate to telepresence and telecommuting, as well as to personal communication services for digital voice. The technologies for telepresence and telecommuting involve a mixture of multimedia and networking. Wireless communication technology also includes techniques such as geopositioning measures, local indoor infrared sensors for location, communications technologies at low, medium, and high bandwidth, and so on. The technical challenges of wireless messaging are well known.12 In particular, the proposed use of ATM LANs will integrate virtual environment research at different sites with communication research.13 The concept of virtual environments is taken here in a broad sense, including both head-mounted and enclosed CAVE-like environments,14 telepresence, and their human factor considerations for the real-time and residual long-term psychological effects of immersion. Strong encourangement for a research emphasis on the human-computer interface is provided by the National Research Council's Committee on Virtual Reality Research and Development, whose final report15 makes specific