Click for next page ( 32


The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 31
5 Recommendations The most unport ant element of the challenge before us is the potential of computer science and technology to Reprove the na- tion's economic productivity and competitiveness. We recognize that improvements in these areas will not be achieved by technological progress alone; changer must also address a long list of economic, educational, managerial, labor, financial, trade, and policy issues that have been partially identified and are still the topic of current investigations and discussions. If progress in these areas is sufficient to maintain the United States at a level of rough parity with other nations, then national primacy in computer science and technology could supply a much-needed competitive edge. Another significant element of the challenge of continued lead- ership in computer science and technology involves the national d~ tense. The U.S. military depends heavily on computers in the oper- ation of strategic, tactical, intelligence, logistic, and command and control systems and is supporting significant research on a number of new-generation computing technologies with potential military ap- plications. Continued and enhanced leadership in computer science and technology ensures a healthier and stronger U.S. defense capa- bility. Indeed, the vital dual role played by computers in defense and in the civilian economy plus their contribution to scientific research underscores that leadership in information technology is essential to U.S. national security. 31

OCR for page 31
32 The United States has no coordinating agency corresponding to Japan's Ministry of International Trade and Industry, and our na- tion has neither the appetite for nor the inclination toward centrally directed Year plans. If the U.S. government does not continue its established role in guiding and supporting basic research in computer science, no one else mill. Government funding, particularly through DARPA, NSF, DOE, and NASA, is the de facto lever of U.S. na- tional policy for setting the future direction of computer science and technology. Government also plays a role in shaping the application of computer science and technology as, for example, through tax incentives or military procurement. We present below two broad recommendations that are aimed at helping the nation meet the national challenge in computer science and technology. The recommendations, which are mutually rein- forcing, should be regarded as guides to strategic directions. In its ongoing and future work, the board will develop more specific rec- ommendations. At this time, it is the board's purpose to underscore the importance of the issues and the need to assign high priority to addressing them. IMPROVE AND EXPAND INFO1IMATION NETWOR1lING Options should be evaluated for improving the computer com- munications infrastructure of the United States providing for such features as fast, high-quality data transmission, support for a range of computer systems, ewe of use, economical nationwide intercon- nectivity, and access to a range of information services and resources. At the same tune potential costs and financing should be considered, along with risks, design alternatives and technology requirements, social and economic unpacts, legal and regulatory aspects, and roles for industry and government. As the ongoing debate surrounding the more limited but component msue of improved computer network- ing for scientific researchers has shown, the msues associated with providing a nationwide service are numerous and complex (CSTB 1988~. The board itself has begun and will continue to evaluate is- sues surrounding the information infrastructure. It recommends that the Congress and the Executive Branch also pursue such study as a first step toward developing appropriate policy. As emphasized above, a substantially improved information in- frastructure could help the United States to achieve many of the

OCR for page 31
33 potential benefits promised but not yet necessarily achieved by com- puters, in particular those for unproving productivity. The infra- structure could also accelerate implementation of many of the ad- vances expected in computer science and technology, especially those conducive to net~vork-based access or those that involve intercon- nection themselves. This promise, combined with the challenge to preserve U.S. leadership in the computer field, makes it unperative to explore information infrastructure options now. SUPPORT FUNDAMENTAL ADVANCES IN COMPUTER SCIENCE AND TECHNOLOGY Continued national strength in any industrial sector calls for sustained, long-term research and development. At a tune when other nations are targeting and strengthening their own national re- search and development efforts in computer-related basic research- examples include Japan's SIGMA project and the European Eco- nom~c Community's Esprit project (U.S. Department of Commerce 1988~- it is altogether appropriate, if not imperative, that the United States step up its own efforts. We recommend that the strengthening of U.S. basic research in computer science and technology be achieved by identifying and funding grand challenges in the field, investing in human resources, strengthening the research environment, and increasing funding for basic research. Some of the relevant issues are discussed below. Identify and Fund Grand Challenges As in other fields of scientific endeavor, in computer science and technology there are a number of grand challenges worthy of long-term research support. Such challenges, if successfully met, would generate major advances in the field and create significant spin-offis in industry and government. We present here a partial list of such grand challenges, without suggesting that every item should be pursued. Instead, a few of these challenges combined with others of comparable magnitude from other lists should be selected and funded by government on a long-term basis. 1. Technology for large, correct software systems. The major component of the cost of computation ~ now the development and maintenance of software. Not only is the cost high, but so is the uncertainty: it is exceedingly difficult to know whether a piece of

OCR for page 31
34 software is correct, and development schedules and budgets often far underestimate actual costs. This challenge calls for tools, methods, and Redefined components that will allow development at reason- able, predictable cost of large software systems (i.e., systems that require a minion lines of code) that are knowably correct when they are released and that admit of modification at a cost proportional to the magnitude of change. ! 2. An ultra-reliable computer system. This challenge would en- tai! development of hardware and software technologies for computer systems that could run 20 years or more on average without fail- ing. Hardware-based fault tolerance requires research in redundant architectures at the component level. Software-based fault tolerance requires research in software-based dynamic reconfiguration of sys- tems, involving concepts of virtual memories, virtual processes, and virtual data paths. Fault tolerance based on artificial intelligence requires research in intelligent monitoring, automated diagnosis, and repair. Ultra-reliable computers would benefit all applications, par- ticularly in space and in hazardous, critical environments. 3. A trillion operations per second ultracomputer. Development of the so-called tera-ops computer system would require advances in a number of component technologies, such as a microprocessor with 1 to 10 nanoseconds cycle time, a billion bits of memory on a square inch chip, and a multiprocessor-memory-~vitch system with a trillion bits per second bandwidth. Achievement of such an ultracom- puter would make possible novel scientific explorations and advances (see Chapter 6~. To achieve this goal, research ~ needed in multi- processing, parallel algorithms, switching, graphics, and memory management. 4. A translating telephone. This challenge cads for the devel- opment of a telephone by means of which people, speaking different languages, can converse directly. The Japanese have already un- dertaken this challenge by recently initiating a 7-year $120 million (16 billion yen) project toward developing a phone system in which a Japanese speaker can converse with an English speaker in real time. This challenge requires a speech system capable of recogniz- ing large-vocabulary, spontaneous, unrehearsed, continuous speech; a natural-sounding speech synthesis approach that preserves speaker characteristics; and a natural language translation system capable of dealing with ambiguity, nongrammaticality, and incomplete phrases. Since achievement of a translating telephone would, in effect, be a si- multaneous achievement of speech recognition, it would have a broad

OCR for page 31
35 and dramatic unpact on human-machine communication throughout essentially all computer applications. 5. Specific systems that learn from practice. There has been a long and continuing interest in systems that learn and discover from examples, observations, and books. Currently, there is some research on system that can learn from signals and symbols. Two longer-term grand challenges in this area are to develop computers that (a) can read a chapter in a college freshman text (say physics or accounting) and answer the questions at the end of the chapter, and (b) learn to assemble an appliance (like a food processor) from observing a person doing the same task. Both are extremely hard problems requiring advances in sensory computing, language, problem-solving techniques, and, most important, learning theory. Achievement of these challenges would open the door to systems that can learn by observation and practice and would therefore result in systems that improve continuously and adapt to change, thereby liberating us from the tedium and inflexibility of prograrrlming at the outset against all foreseeable eventualities. 6. Self-replicating systems. There have been a few theoretical studies in this area since the 1950s. The problem is of some practical interest in areas such as space manufacturing. Rather than launching a whole factory into space, it may be possible to send a small set of machine tools that can produce perhaps 95 percent of the parts needed for such a factory, using locally available raw materials and assembling the factory in situ. The solution to this problem involves many different disciplines. Research problems in this area include knowledge capture for reverse engineering and replication, design for manufacturability, and robotics. Each of the above grand challenges would require and would gen- erate significant breakthroughs and fundamental advances in com- puter science and technology. In each case, success or failure could be clearly established and appreciated by nonexperts. And each of these tasks would require long-term stable funding at significant levels. As we noted above, the Japanese are already budgeting some $120 mil- lion over the next 7 years on the translating telephone alone (and it is estunated that such other major Japanese computer research efforts as the Fifth Generation project and the Superspeed project will each cost more than $100 million annually (OSTP 1987~. Each of the other grand challenges on the list would probably call for funding at comparable or greater levels. Success would by no means be guaranteed, but the payoffs from

OCR for page 31
36 these efforts, even if they achieved far lem than total success, would be substantial. Besides advancing the state of the art, the pursuit of such challenges would create a new generation of leading computer researchers, who in turn would contribute to the creative and effective use of computers throughout the nation. These challenges are grand, not only because of the immediate accomplishments sought, but because achieving them would give rise to immense technological spin-offs benefiting industry, defense, and society. Barest m Unman Resources The single most important factor crucial to the success of U.S. computer science and technology is a reservoir of experienced, knowI- edgeable, and creative people. Developing and benefiting from this invaluable human resource requires action across severer fronts to increase the number and enhance the quality of the people involved. For example, increasing the number of people well-educated in the productive use of computers depends on increasing the number of qualified teachers at all levels, beginning tenth elementary school. At the college level, computer science has the highest student-to-faculty ratio of any of the physical science and engineering disciplines (Gries et al. 1986~. Because of the shortage of properly trained teachers, in many schools those teaching the subject are inadequately qualified. It is especially important for graduate teaching and research that we increase the national supply of Ph.D. holders in this field, for which demand from educational institutions as well as industry has exceeded supply by an estimated 4 to ~ ratio (Hamlin, cited in Cries et al. 1986~. In 1987, U.S. universities awarded only 466 computer science Ph.D.s, 2 percent of all Ph.D.s in science and en- gineering. Almost half of those degrees went to foreign students. By contrast, 845 Ph.D.s were awarded in mathematical sciences. The average computer science department has 18.5 faculty, while the average mathematics department has 30.2 (NSF 1987, Gries and Marsh 1988~. Since the top-ranked universities are already produc- ing Ph.D.s to the limits of their capacities, the pool of research talent can be expanded only by substantially increasing the funding for additional computer science departments and by funding efforts to unprove the teaching of computer science and technology at ad educational levels. While the number of students entering graduate school in com- puter science is fortunately still increasing, there may soon be a

OCR for page 31
37 downward trend. Undergraduate enrollments, after 5 years of dra- matic annual increases and severe straining of human and physical resources, have declined over the past few years almost as precipi- tously as they had been increasing. Moreover, the quality of students pursuing college courses in computer science (and other sciences and engineering) appears to be declining. Students are not weD-prepared and seem unwilling to undertake the hard work of mastering subjects with a mathematical basis. It is more import ant than ever that we catch the imagination of youngsters in school and create an aware- ness of the intellectual excitement and rewards of the field. We feel that this can best be done by teaching computation as a mode of thought as important as the disciplines of mathematics and the nat- ural sciences. Student and teacher interest should also be stimulated early with hands-on experience In schoob, business, and industry via summer jobs, internships, sabbaticals, and the like. To continue to cultivate a growing pool of researchers at the college level, significant increases in the number of graduate fellowships and other forms of financial support for both students and faculty are required. Investing in people to become knowledgeable in computer science and technology goes beyond the satisfaction of basic research needs; it increases the talent available for more widespread use of computers throughout the economy. This is especially unportant in the areas of software clevelopment, with its severe mismatch between programmer demand and supply. The board wall be examining human resource issues in computer science and technology In more detail to focus attention on problems and options for their resolution. Strengthen the Research Environment Computer science has a strong experunental component whose past progress has been intimately linked to the availability of ad- vanced experimental resources. Accordingly, we recommend making possible the acquisition by research centers of advanced computer workstations, high-performance machines, multiprocessors, and other advanced computer technology research tools on an ongoing basis to guarantee that computer science researchers have at Al times the best available tools to continue advancing the state of the art. We also recommend funding the use of advanced emulators and tools for prototyping to encourage and facilitate the design of innovative new architectures and software tools. Moreover, we recommend increased support for efforts in VI.ST design and processes by maintaining and

OCR for page 31
38 improving foundry services such as MOSIS. Finally, access to high- speed, high-functionality computer networks is critical for progress in this field, as in others. Acreage Ennding for Basic Research The board recommends that increased funding be directed to the promising technological areas outlined in Part ~ and discussed in more detail in Part Il. Those areas fall within the categories of machines, systems, and software; artificial intelligence; and theoret- ical computer science. The board further recommends that special initiatives be undertaken to strengthen promising research that has been underfunded in the past, in particular in the areas of theory, software productivity, and commercial applications of computer tech- nology and infrastructure. The latter two areas are directly relevant to U.S. productivity and competitiveness. In the course of its ongo- ing and future work, the board will address specific research needs in various segments of the computer science and technology field.