Computing the Future: A Broader Agenda for Computer Science and Engineering (1992)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

2 Looking to the Future of CS&E BROADENING THE FIELD The time has come for the CS&E community to adopt a broader agenda that builds on the traditional strengths and interests of com- puter scientists and engineers. In particular, a broader agenda asks the community to: · Look outward as well as inward. A broader agenda would legiti- mize closer couplings to science, engineering, commerce, and indus- try. The committee believes that outward-looking interactions will enrich CS&E as a discipline by identifying new and challenging re- search problems, and will provide valuable assistance to those in science, engineering, commerce, and industry whose problems re- quire the best talent and expertise that CS&E has to offer. · Encourage greater interaction between research (especially theoreti- cal research) and computing practice. CS&E has a tradition of deriving inspiration and richness from practice, and, in turn, contributing clean concepts and fundamental theory that have been effective in further- ing computing practice. This tradition is well represented by the extensive interplay between theory and practice in programming lan- guages and compiler design, databases, machine architecture, operat- ing systems, distributed computing, and computer graphics. How- ever, as CS&E has matured, the theoretical side of many of these areas has become more inwardly focused. This is not altogether un 55

56 COMPUTING THE FUTURE desirable, but it is crucial that researchers working in these areas maintain an active effort to draw inspiration from practice and to continue to rise to the challenge of making a difference to the outside world. Box 2.1 illustrates possible connections between theoretical research and computing practice that arise in the context of the High Performance Computing and Communications Program. The committee's belief in the wisdom of a broader agenda for CS&E is based on several considerations. The first is that computing most often serves disciplines and areas other than CS&E; even the practice of such a characteristic CS&E topic as designing computer languages cannot be fully abstracted away from application domains, a point all too often overlooked in CS&E's search for the generally applicable. It would, for example, be folly to try to build even the framework of a computer language for music composition without a background in music. Beyond the inescapable engineering substrate of digital electronics and communications, computer scientists and engineers need to have some appreciation for the economics, finance, and administration intrinsic to business, the mathematics and phys- ics behind engineering, and the mathematics and other sciences that underlie computing applications in industry. Moreover, the number of problem domains to which CS&E is directly relevant will grow dramatically over time as a direct result of the increasing proliferation of computing into all sectors of soci- ety. Thus broadening presents major intellectual opportunities for 1 1 researchers in CS&E. A precedent to keep in mind in this regard is that of mathematics (Box 2.2~. Finally, nonroutine applications of computing technology to oth- er problem domains can be regarded as explorations undertaken to

[OO~G TO ~ TITLE OF CSSE ~7 understand empirically the actual utility of ~ given generation of computing technology. If computer scientists and engineers are in- volved in the design, implementation/ and analysis of these expert mental inadequacies in any given genershon of computing technolo- gy ~iH be better understood, laying the groundwork for the invention of the next generation.

58 COMPUTING THE FUTURE A second consideration is that regardless of whether computer scientists and engineers participate, computing will continue its march into the various sectors of science, engineering, commerce, and ir~- dustry. But as argued in Chapter 1, the future will belong to those who understand best how to apply new computing technologies to an ever wider range of problem domains; computer scientists and engineers are ideally situated both to create these technologies and to understand and articulate the appropriate application of these tech- nologies to other domains. Indeed, specialists in other areas are of- ten unable to articulate the computing aspects of the problem they want solved. If CS&E professionals remain uninvolved with other areas, the application of computing to those areas will most likely not reflect the most current or most relevant work that CS&E has to offer. The pace as well as direction of the information revolution will also be affected by the participation of computer scientists and engi- neers. Developments that may occur decades in the future without their participation may be only years away with it. The committee believes that dramatic improvements in computing efficiency and performance will be possible only with the full participation of com- puter scientists and engineers. The third consideration is one of recognizing social responsibili- ty. As Robert M. White, president of the National Academy of Engi- neering, has argued, ~_ ~ ~. . --r ~ Investments in research and development have to have an eco- nomic, social, or defense payback. Science and engineering research, like any other [federally funded] activity in this country, has a social purpose, and it must justify expenditures in ways that can be under- stood and lead to the social and economic betterment of the coun- try.i Given the growing ubiquity of computing in all sectors of society and the intimate connection between computing and CS&E, research in CS&E among all the C~iton~f~ anti niacin disciplines has a o ~ r particularly powerful justification with respect to social payback. The fourth consideration is that CS&E itself may contribute im- portant intellectual abstractions to other fields. Such contributions may be serendipitous, but when these applications do occur, their intellectual reach is often quite compelling. Consider the following: · The study of chaos, fractals, and dynamical systems. While work in this area goes back to the late 1800s (the days of Poincare), modern computation has rejuvenated this work and underscored its impor- tance. Many of today's insights into chaotic phenomena are the di

LOOKING TO THE FUTURE OF CS&E 59 rect result of extensive computational experimentation with dynami- cal systems and are often displayed in graphical form. A computer can be used essentially as a laboratory for experimental mathematics; as a result, computer-generated visualizations of chaotic phenomena at ever higher resolutions have led to conjectures about their proper- ties, which can then be addressed in a mathematically rigorous fash .2 · Cognitive psychology. The conceptualization of the human brain as a computational information processor, perhaps operating in par- allel, has emerged as an important paradigm for the investigation of human cognitive processes. A computational model allows indeed requires researchers in cognitive psychology to formulate explicit and testable models of cognition. · The study of algorithms in mathematics. The study of algorithms and computational complexity (i.e., the complexity of mathematical processes) has added completely new chapters to mathematical re- search. The classification by computer scientists of computational problems into large classes of problems of equivalent complexity (e.g., P. NP, PSPACE, EXPTIME) has led to new insights in game theory, logic, and recursive function theory. For example, the study of com- plexity has resulted in the systematic study of resource-bounded strate- gy selection as a part of game theory. Driven by the computer, the study of logic has also evolved from an emphasis on the foundations of mathematics to the design and study of effective, easy-to-use proof systems for use in the verification of programs and communication protocols. · City and building planning. Cities become more congested as they become larger, and they are most severely congested near the center. Theoretical analysis of the wiring of chips and circuit boards (analysis that computer scientists and engineers pioneered) helps to explain why congestion within cities occurs in this fashion and has influenced the planning of cities, factories, and office buildings. In each of these cases, intellectual insights have been gained not just by using a computer to perform some calculation more rapidly, but by understanding how the abstractions of CS&E might be rele- vant to some conceptual framework in another area of inquiry. Lastly, a broadening of CS&E speaks to economic realities faced by the field. As discussed in Chapter 1, the computer industry is undergoing a major shift, from selling thousands of million-dollar computer systems to millions of thousand-dollar systems. The mass- market nature of today's business calls for relatively fewer people who build computer technology (hardware or systems software) and

60 COMPUTING THE FUTURE relatively more people who know what to do with computers (e.g., write applications software or integrate complex systems for specific tasks).3 The importance of domain-specific knowledge relative to programming skills has increased, partly because new tools make programming much easier to learn and do (although this may change if new computing systems such as parallel processors require new programming paradigms), and partly because knowing a field (e.g., accounting) is often harder and more relevant than knowing a pro- gramming language. CS&E researchers also face economic concerns. Research budgets for all science and engineering will come under increasing pressure in the future, and despite the HPCC Program, CS&E is no exception. A broader research agenda for CS&E will enable CS&E researchers to make a better case for receiving support from nontraditional sourc- es.4 A relevant point of information is that over 42 percent of the entire federal science and engineering research budget (i.e., over $10 billion out of the total $24 billion) for FY 1991 was obligated by 12 federal agencies whose individual science and engineering research budgets each allocated less than 1 percent to computer science re- search.5 An action plan to develop a broader agenda for CS&E that recog- nizes the confidence, strength, maturity, and social obligation of the field calls for the CS&E community to broaden its research scope by expanding intellectual interaction with science, engineering, indus- try, and commerce, and to broaden undergraduate and graduate ed- ucation in CS&E accordingly. (Box 2.3 gives the view of the Associa- tion for Computing Machinery (ACM) on the need to broaden the CS&E agenda.) Concomitantly, other fields will need to develop some familiarity with modern CS&E if they are to maximize the benefits that computing can bring to them; this need for other fields to broad- en toward CS&E is discussed further in Chapter 4. A broader agenda for CS&E in research and education is elabo- rated in the sections "Research Opportunities in Broadening" and "Broadening Educational Horizons in CS&E." The section immedi- ately below provides some historical perspective and context for un- derstanding the relationship between CS&E and other fields. A HISTORICAL PERSPECTIVE Chapter 1 described the impact of computing in all aspects of society and explained the important role CS&E plays in computing practice. Increasingly, fields such as computational medicine and computational physics are emerging as subdisciplines of their parent

LOOKING TO THE FUTURE OF CS&E 61 fields-indeed, for every field X, it sometimes seems that someone creates a subfield, computational X. Cooperation and interconnec- tion of CS&E with these computational subdisciplines should be a major aspect of computing, as suggested in Figure 2.1. In the past, however, CS&E has been slow to participate directly in the research and development of these computational fields. This is understandable. Even though CS&E was initially populated main- ly by people from other disciplines,6 a natural tendency was to con- centrate on the development of the scientific base in core areas of CS&E. There were more than enough exciting problems in this core to keep the relatively small number of researchers busy without wor- rying about applications in other disciplines, and a lack of incentives to pursue interdisciplinary work kept most researchers working irk the core areas. There have been a few instances of interdisciplinary work. For example, computer science at the University of Michigan was closely allied with medicine and psychology, at the Georgia Institute of Tech- nology with library science. The University of North Carolina has had medical imaging and molecular graphics projects for many years. Stanford University was a pioneer in the application of artificial in

62 COMPUTING THE FUTURE Medicine //, Entertainment ~ _ ~ ~'' 1 / ~ <.._~ 1= CompL'fin9 Computer \ Science and Engineering / Humanities ~ ~/ Art :~ = Science \,, Business | // Engineering l FIGURE 2.1 Computer science and engineering, computing, and other problem domains. CS&E is central to computing, which in turn affects many problem domains. telligence to medicine. And from the beginning, numerical analysis was considered part of computer science in many departments many of these numerical analysts are now beginning to call themselves computational scientists and are playing a major role in computa- tional science. But by and large, the very nature of CS&E and its growing pains forced the field to look inward. A striking example of this inward-looking tendency today is the attitude of the academic CS&E community toward the general busi- ness community. Both the number of commercial users of computers and the dollar value of computers used for commercial purposes far exceed the analogous quantities for academic science, and yet, apart from a few in the database community, academic CS&E researchers have been extraordinarily reluctant to engage the problems faced by business and commerce (although they do contribute to and benefit from the activities of businesses that produce computer-related prod- ucts). A simple illustration can be found in the divergent attitudes to- ward the programming language Cobol. Among those involved in advancing the field, Cobol is derided as 30-year-old technology, an

LOOKING TO THE FUTURE OF CS&E 63 anachronism. But Cobol is the language in which the vast majority of business and commercial programs have been written and are sup- ported. A second point is that for the last 25 years' the need to solve computation-intensive scientific and engineering problems rather than business problems has motivated the design of ever faster proces- sors. Finally, during its deliberations the committee found relatively few academic computer scientists or engineers with research inter- ests that arise directly from the needs of the commercial domain. This important aspect of the field has generally been left to business schools, library schools, and departments of operations research and manufacturing. As a result, the mainstream academic CS&E commu- nity has not participated much in the development of the many com- puting innovations that have transformed the modern corporation and the practice of business today. The inward-looking attitude of CS&E manifests itself to a lesser (though still substantial) degree with respect to other applications as well. Although increasing numbers of computer scientists and engi- neers have research interests relevant to other scientific and engi- neering problems, the CS&E community still views with some appre- hension efforts to promote collaborations with other disciplines. For example, a recent CSTB workshop intended to bring together young computer scientists and engineers with molecular biologists in need of sophisticated computational systems elicited some concerns that pursuing such challenges would be inimical to progress in the aca- demic CS&E environment. The relevance and value of such work from a CS&E perspective are not widely recognized, and promotion opportunities for computer scientists and engineers who choose to work in this interdisciplinary area could thus be damaged.7 Conversely, various disciplines have likewise been mistrustful of CS&E and have not known whether to embrace CS&E as a real disci- pline. Wasn't computer science just programming? Was it really a science? Consider, for example, the following quotation, taken from a recent National Research Council report on physics:8 ... computer programming introduces problems.... [F]or the computational theorist the programming problems have led to spe- cial difficulties, including a great deal of misunderstanding and un- derestimation of the role and intellectual quality of computational physics. Computer programming and debugging is, in large part, a mind- dulling, menial task, in which hours and days and weeks are spent making trivial changes in response to trivial errors orfiguring out how to format the output. Yet one must be able at any moment to apply the deepest analytical skills in order to understand an unexpected result or to track down a subtle bug. "Emphasis added.]

64 COMPUTING THE FUTURE Although the statement does acknowledge the intellectual chal- lenges of debugging programs, it fails to do justice to the wealth of knowledge and talent needed to construct correct programs in the first place. Indeed, it suggests that knowledge of a programming language's syntax and the ability to perform low-level coding are all that a scientific programmer needs, whereas in fact knowledge of data structures and algorithms is the key to effective programming, and the structured decomposition of a problem and the stepwise re- finement of proposed solutions account for the largest portion of serious programming efforts. Even more problematically, it implies that the only function a program must serve is to solve a given prob- lem. Such a view is overly narrow, because it does not recognize that problems evolve, that therefore programs must evolve, and that CS&E is responsible for most of the tools and concepts needed to write evolvable programs. Put another way, it is understandable if physi- cists do not fully comprehend the intellectual challenges required to create the tools they use so freely. But rejection of those challenges as irrelevant to the business at hand may well discourage the intel- lectual work necessary to develop better tools. Beginning around 1986, CS&E as a field began to recognize the importance of interdisciplinary research and broadening. For exam- ple, interdisciplinary research became an issue at the biannual meet- ings of the chairs of Ph.D.-granting computer science departments as early as 1986. The HPCC Program, with its interdisciplinary orienta- tion, had its roots in various planning meetings held in 1986. Senior officials in NSF's Computer and Information Sciences and Engineer- ing Directorate in the late 1980s were important advocates for inter- disciplir~ary work. Concerns about the insularity of the field were raised at the ACM-CPA conference on Strategic Directions in 19899 and at the 1988 Snowbird meeting.~° In response to an inquiry from the committee, the ACM argued for a CS&E agenda that was broader and more closely linked to social needs. Today, one can find many more though still not substantial-instances of CS&E faculty mem- bers taking part in interdisciplinary work. At present, CS&E is in transition: many computer scientists and engineers are aware of its previous isolation and the need for a broader agenda, but the field as a whole has not yet taken sufficient action to remedy the problem or to change its culture. RESEARCH OPPORTUNITIES IN BROADENING One simple principle should guide the formulation of a broader research agenda:

LOOKING TO THE FUTURE OF CS&E Address substantive research problems in CS&E in the con- text of their application in and relevance to other problem domains, and derive inspiration for identifying and solving these research problems from these other domains. 65 By so doing, CS&E can be framed simultaneously as a discipline with its own deep intellectual traditions, as well as one that is appli- cable to other problem domains. CS&E can thus be an engine of progress and conceptual change in these other domains, even as they contribute to the identification of new areas of inquiry within CS&E.~2 In developing this notion further, it is useful to consider the tra- ditional distinctions between basic research (conducted to obtain a fundamental understanding of some phenomenon), applied research (done to investigate the nuances of this phenomenon with an appli- cation area in mind and perhaps to construct proof-of-principle pro- totypes), and development (which builds on research-based under- standing to construct engineering prototypes that demonstrate economic and manufacturing feasibility and results in items that are very close to marketable products).~3 This neat and orderly progression de- scribes the evolution of some products, but it often happens that in the course of bringing a product to market, it is not clear when a given activity fits into one of these categories. Indeed, some prod- ucts have bypassed the traditional development phase, going directly from research to use as the core of a new application. Although such products generally have not met the usual standards of quality ex- pected of more traditionally developed software products, they have established markets for the services provided by those products. In turn, these markets have then driven further improvement of those products. Examples include the Mach kernel for operating systems, the Scribe text formatter, the Emacs text editor, the Ingres relational database system, the Magic CAD system, the Query-By-Example da- tabase system, and the Unix operating system, all of which were first developed in a research environment and widely distributed initially at little or no cost. Such phenomena persuade the committee that the separation of basic research, applied research, and development is dubious, especially within CS&E. Given the way research in CS&E is actually done, distinctions between basic and applied research are especially artificial, since both call for the exercise of the same scien- tific and engineering judgment, creativity, skill, and talent.~4 A1- though the traditional areas of CS&E research (e.g., those discussed in Chapter 3) remain at the core of CS&E research and still present major and substantive intellectual challenges worthy of sustained ef- fort, they should not alone define the boundaries of the CS&E re- search agenda.

66 COMPUTING THE FUTURE Rather than a one-dimer~sional characterization of research lead- ing to development, a two-dimensional model may be more appro- priate (Figure 2.2~. The committee believes that research is any in- vestigative activity that results in the creation of renew knowledge (i.e., represented in the upper half of Figure 2.2), whether or not that activity is associated with a specific product item (i.e., irrespective of its horizontal coordinate). Thus research might well be an aspect of trying to improve the manufacturing or maintenance or upgrading of a specific product. Academics, who are generally free to choose their areas of research without constraint, should be encouraged to select problems that involve commercial products as long as significant new kr~owledge is created and demonstrable intellectual achievement is the result. As computer scientists and engineers engage research problems that arise in other problem domains, the center of gravity of tradi- tional CS&E research may shift. For example, a great deal of re- search irk CS&E is now devoted to increasing the speed of computa a) o o cn In a) - a' ~5 a) o By id o o a' .../ Traditional t~h~.~in r~.s~arch" \ / /~his shout\ also count as ~ research / ./ Traditional "development" '$2'~R"9 Product Orientation (low to high) FIGURE 2.2 A two-dimensional characterization for research and develop- ment. The vertical axis refers to the extent to which a given activity results in the creation of new knowledge, while the horizontal axis refers to the extent to which that activity is oriented toward a specific deliverable product intended for commercial sale (usually with associated deadlines).

LOOKING TO THE FUTURE OF CS&E 67 lion (i.e., developing faster and faster processors). Although researchers currently working on problems in this domain may well continue to proceed as they always have, it may not be surprising to see in addi- tion a larger effort in other areas of CS&E that, though always con- sidered "legitimate" CS&E, have not always been well represented in the discipline (e.g., design of faster and more capable input/output and storage technologies and of better user interfaces). Academic researchers often equate "basic" research with investi- gator-initiated research, and "applied" research with funder-initiated research. But even this distinction is not as clear today as it once might have been. Sponsors in the past may have been able to sup- port all proposals for good science regardless of specific area or top- ic, although this can be debated. But it is clear today that funding sponsors are more selective about the directions in which they wish to focus their efforts, and they find willing allies in the many re- searchers who submit grant proposals for "basic" research in spon- sor-preferred areas of interest. Framed as it is in the context of grand challenges in science and engineering, the HPCC Program is a good start toward a broader CS&E research agenda. But other sets of grand challenges can be imagined for different endeavors of social significance. For example, grand challenges relevant to business could include translating tele- phones that allow a Russian and an American to converse without difficultyi5 or copiers that reproduce a document and automatically generate a summary of key points in the document. Grand challeng- es relevant to medicine might include a "physician's assistant" (Box 2.4) with on-line access to patient data, physician's orders, and labo- ratory results that could monitor patients to provide status reports and alert the physician to important events, or an integrated medical information system that would give clinical practitioners convenient and flexible access to comprehensive, accurate, current medical infor- mation. There is no shortage of problem domains outside CS&E in which challenging and intellectually substantive CS&E research problems arise. Indeed, some research areas have developed directly in re- sponse to challenging problems, such as speech input or physical modeling and simulation. These areas require substantial interac- tions between CS&E researchers and those in other fields and often have a strong experimental component. Almost inevitably, it is re- search that addresses specific and concrete problems that captures the public's attention, since it is most easily understood by the lay public and also influences the public's perception of the benefits of CS&E research.

66 CO~G THE FUT~E . . ............... .

LOOKING TO THE FUTURE OF CS&E BROADER RESEARCH AGENDA SOME ILLUSTRATIONS 69 To suggest what a broader research agenda might entail for CS&E, four topics are discussed below to illustrate CS&E research problems that arise in embracing interdisciplinary and applications-oriented work. Note that these descriptions are intended to be illustrative rather than explicative of priorities for such work. Earth Sciences and the Environment Among the great challenges of computing is modeling the earth system, including the climate, hydrologic cycle, ocean circulation, growth of the biosphere, and gas exchange between the atmosphere and the terrestrial and oceanic biota. In this complex physical system are a multitude of phenomena that change on local, regional, and global scales. Detailed scientific models describe processes whose temporal and spatial scales differ widely. The data that drive and verify these models come from satellite- and ground-based sensors. By the year 2000, these sensors will have the spectral and spatial coverage and resolution needed to provide data to support accurate modeling and analysis by scientists and informed decisions by policy makers and legislators. The sensors and their associated scientific data products will gen- erate nearly a petabyte (10~5 bytes) of data each year, and these data will be integrated with local measurements of fluxes of water, ener- gy, and chemical species. (For scale, note that today, terabyte (10~2 bytes) databases are regarded as very large; an ordinary book is a few megabytes of textual information.) Improvements are needed in in- formation management systems for these data, along with techniques for their analysis, distribution, four-dimensional assimilation, and in- corporation into models. Construction and operation of valid scientific models that de- scribe and predict the dynamics and processes of the earth system will require interdisciplinary teams of experts from geophysical, bio- logical, and computer sciences and engineering. Needed are improve- ments in our understanding of how processes at different spatial and temporal scales interact, substantially more calculational power, bet- ter methods of accessing and storing large volumes of heterogeneous data of varying structures in distributed archives, and new ways of translating scientific ideas more rapidly and more reliably into work- ing computer code. The result of this collaborative theoretical, ex- perimental, and computing effort will be a much deeper understand- ing of the earth system as modified by human activities.

70 COMPUTING THE FUTURE The contributions needed from CS&E are fast, reliable networks that allow examination of large data sets on remote computers; algo- rithms that can be employed by fast computers, probably with paral- lel architectures; and improved tools for understanding and manag- ing staggering volumes of information. Moreover, the crucial dependence of research in the earth sciences on large-scale computer models of phenomena and processes suggests that CS&E expertise has a mean- ingful role to play in such work. Consider, for example, the historically important concept of re- peatability. In the past, a scientist could often read a paper describ- ing an experiment and redo the experiment, therefore verifying its correctness or identifying an earlier result as erroneous. But earth scientists whose work depends on large-scale computer models do not have this ability. It is generally not feasible for someone to read a few papers about predictions of a climate model, study the equa- tions, write his or her own model, and subject the model and the conclusions to the necessary scrutiny; as a result, careful study is rarely given to modeling software. Even when every line of source code is made available, a comprehensive understanding of someone else's model may not be achievable, due to possible interactions be- tween different parts of the model. The "same" model run on a different computer may give different results. As a result, the con- cept of repeatability is in danger of being lost. Thus collaboration between CS&E and the earth sciences will be necessary; Box 2.5 describes one example of such an interaction. An example of a large-scale problem of scientific and social inter- est that depends on advances in computing is NASA's Earth Observ- ing System (EOS). Global environmental change has become a top- priority issue in the public debate. Investigation of the causes and magnitudes of environmental change, especially at large regional and global scales, depends crucially on large data sets containing geo- physical and biological information that are reliable enough to en- able the detection of subtle changes. The computing challenge is the creation and integration of these data sets into a system for analysis. The U.S. Global Change Research Program was launched in 1989 in response to mounting national and international concerns about the global environment. The objectives of the program are to moni- tor, understand, and predict environmental change on a global scale The program calls for earth probes- satellite sensors dedicated to near-term observations of specific phenomena to be launched in the next few years. Beginning in 1998, EOS will be put into place and will collect data for 15 years. EOS will provide a much more capable space-based observing system, the EOS Data and Information System (EOSDIS), and a scientific research program.

LOOKING TO THE FUTURE OF CS&E 71 .,, ,,,, ,,., ,,., ,, , .,.,, i,,.,, ............. ... . . ,,,,,, ., . ,,,,,,, , ,, .. .... ... ~ .. ... . .. . . . . . T 'H ~. ~''A ' ~. ~' cEs AND cSsE ... . . ...... . ... . .................... .. ...... ...... ............. ...................... ... .. ... . '. ', ' '.'.' ~l, . ~',6e'l' Change '$ 'at the'Unt~s't~ of ca'tI'~a'have dIs ~ ~ ysmm.~..,,.,,3 . e.~..~..~.~.~....~.. them: nix their abj~j~ accents rams Eta and ~th~erekiy Ado research. . ~y,, , ~g,.U,,l.~., , ~ ~ | an ~ e|ect~n Tic ~ tepp5|t£ rid ~ They also ted u ~ He ~ ad ~ hoc co} {ec| . ~,. ~t ,,l ~.~.~. ~e qUl.~l .~d and tmnspo - d tO . ''''''"2w"''''0."~""2" '' ' ''' ' ' ' ' ' ''' ' "' ' ' '" ' ' '' ' "'' ' ' ' ' ' " ' '' '" ' '' '' ' ' ' ' ' ' ' ' ' ' ' ' '' ' ................................................. .. : ~ ~O ~ $ St0~ art: an~ ~ so3~n soluttons current a e a a a e $ pp e e s s c Illn~tyl911 1 11. 1 1.i 1 3 1 . i .................................................................................................................................................................................................................................................. Sterta N - ~a ~s ~ I- cen~d at the nta ~ @dms$ ~s$ues~ gI~} chan~' .................................................... ~.~. ~.~.~.~.~. ~, ~a,~. ~:.. : . ~ .............................. ~.O,l ,., , ~. ~s, ~. ~.~.e, ~.~.~.~.~.e , ~.~. ~. ~.,., ,8~.~., ~.~l,( ...... ~ ~ ~ ~. ~ ~ ... ... ~. . ~... . ~....... ~ ..... ~, ..... ~ ~ ~.. ~ ~ ~... ~ ~ ~ i. ~ . ~ ~. ... . , ~. ~. ~ .............. ,, ~, ~.~. ~.~.~.~ ~. ~. . ~ . . . . ~ . . . ~ ~ . . ~ . ~ . .......................... ~.~. mW. ~t,.~t ~.~.~. ~.~. ~A,~.~. 1~t ~ .~.~.~. gio15a~ change~ ~rese~a~cherls, pr~m~ar~y~fro~fn~tiNe Sa nta ~6ar:bara, ~Lios >\~ .................... ~.~. ~. ~, ~.~.n, . . . . . . . . . . . . . . . . ... .... ~. ~.. . . ~ ...... . ~ .. . . ~ ~. ...... ............ .. ........... > ~t ne pmtotype systems ano w't P~ '"'g''"~.'t""~."'a''''""'2 '' '''' '' ' ' ' '" '" '"''' ''' ' '' '" "'" 1'" ' ' ''' ' '' ' ' ' ' '"'' i"''' ~, :. ~ . : ~ ~ ~ .: ~ ~ ~ collect on of publ~r a~s that most ~t poltc~s a~0 ........................... - : sta ~ ~ ~ ~ ~ s o - a ( a se ~ ~ ~ h ~t element of the partne-~p ~s -~s f~r . ~ . ~ ~ ~ ~. .~.~. ~ ~. ~ ~. ~ ~ ~ . ~.. . ~ ~...... ~... ~ ~ ~ ........................ ~- ~d ~. ~. ~x, ~ ... ~ ' ' ' '' ' '' :' ...... : ~ . . . . , . . . . . . . . . . . . . . . . ~ . ~ . . . . . ~ ~ . . . ~ ~ . . . . . . ~ ~ ~ . . . . . . . ~ ~ , . . . ~ ~ ~ ~ . ~ ~ . . . . . . . . . ~ ~ ~ ~ . ~ . . . ~ ~ ~ ~ ~ . . . . . . . . . . . . . . . . . . ~ ~ ~ . ~ . ~ . ~ . . . . . . . . . . . . . . . ~ ~ ~ ~ . ~ ~ ~ . . ~ ~ . ~ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . a~j~ T~fRW, ~ the~ ~@@@e@ ^' j~@it corn} a@@r a@@ s>\~c @~@~ arno~@s ~ t@e < ~ ~, ,~ ~,~ ,~,~, { 2n d O s tt I'3, I pa ~ t,( I pa o2ts ~ ~ ~ ~'I'l''l'' ' ' ' se t~ $ .,.~.~.~.~. ~ ~ . . an., p.~.~.~.~ .~.n tec no ~y ,. ta? ~. ,. ... . . .... ~ ... ......... ....... .... .. , ~.~,.~ ~ . ~q ~. t ~th~.~ ,1 ~s~ .~a . na~.~o . ts ~ ~.~t {~t of ~.~.~..~ ..~ ~ ~.~..~ ~.~.............~. ~ .........~ ~. ~.~.... ~.~ ~ ....... ..............~..... ~ ..-... ~.....~. ~- .....~..........~......~....~......~. ...~.~.~..~...~.~...~.~.~.....~..~.~.~.. .~ x..... ~..~..~ ~.~..... ~. ~ ~ ~. ....~.~.. ~.~.....~....~.~.~.~....~.~..~.~...~..~......~.~-~.~.~....~.~........~..~.......~.~...~.~. : . .... . ..: .~....~.~.~...~.~.~....~.~......~.~.~.~.~...~..........~...~...~........~.~ ~.~........~.. ~..~.....~....~.~........~..~. ~...~...~....~.~......~..~........~.~.~...........~... ~...~ ~.......~..~.....~ , ~ Investigating the causes and magnitudes of environmental change on regional and global scales requires large volumes of biological and geophysical data, several terabytes per day over a lifetime of 10 to 20 years. EOSDIS is intended to manage this data, facilitating inexpensive, quick, and convenient access, integrating these data in a reliable manner, and promoting the interaction of scientists from a

72 COMPUTING THE FUTURE broad variety of disciplines and living all over the world. Several aspects of the U.S. Global Change Research Program present major challenges to computer science and engineering: · Large volumes of data require methods for transmitting an storing large data sets without loss of information, browsing these data sets quickly to identify interesting features and characteristics, displaying and organizing these large data sets in meaningful ways, and selecting representative data. To understand the volume of data involved, consider that a common technique for presenting data or- ganized by spatial position is to display an image. One hundred megabytes of spatial data can be reduced to a color image that can be taken in by the human eye in a single look; for comparison, a single typewritten page holds 2,000 bytes. When the amount of data pro- duced is 30,000 such images every day (many more image-equiva- lents will result from recombinations of the data derived by scientific analysis), even visual representations are likely to be inadequate. Automated vision and pattern recognition algorithms may well be necessary to allow comprehension of these large amounts of data. · Information must be distributed reliably among err international community of users from different disciplinary traditions. EOSDIS will have to validate data and prevent data corruption. Since re- searchers will have to rely on networks to provide access, EOSDIS will have to surmount barriers between different networks and ma . chines. · The character of the scientific problem is such that small differ- ences in the data or results can have large policy implications. Therefore, accuracy has a high premium. This requirement drives the need for large amounts of data and for sophisticated algorithms and models that can make meaningful predictions and identify long-term trends in a noisy environment. The development of such algorithms and models necessarily involves interdisciplinary expertise. The preceding discussion suggests how a tough scientific prob- lem requires solving generic and genuinely challenging CS&E prob- lems. Progress will depend on more scientists knowing something about CS&E as a research discipline, and computer scientists and engineers knowing something about the scientific and technical problems in other disciplines.~7 Computational Biology Computational methods have a long history of application to problems in the physical sciences and engineering. However, during the past

LOOKING TO THE FUTURE OF CS&E 73 decade biomedical research and technology have seen a comparable influx of computational methods. Three examples are widespread use of computer imaging techniques in medical diagnosis, computa- tional methods for drug design and structure refinement in molecu- lar biology and medicine, and computational neural science, which attempts to understand the principles of development and functional cooperation of the neurons in the brain, using computer simulations as a main tool. The primary driving force behind the proliferation of computa- tional techniques in biology and medicine has been the rapid devel- opment of computing technology, which has become increasingly less expensive and better adapted to the complex data-processing tasks required in these fields. Box 2.6 describes one example of a biological problem solved by applying a good algorithm developed by a computer scientist. How- ever, the available computational power and the algorithms for com- putational biology are in many cases still inadequate. But CS&E advances expected in the next decade will benefit mo- lecular and structural biology. Research in these areas currently xe- lies on computer simulations of the structure and dynamics of biopoly- mers, e.g., proteins and DNA. Today, simulations are possible only for small biopolymers of a few thousand atoms and over very short time periods (10-9 seconds); moreover, these simulations treat pro- teins mostly as systems of classical particles. A simulation of 10-6 seconds, much more useful but still relatively short on the time scale of many interesting biochemical reactions, would require about 100 years on a Cray-2 processor. In actual biological systems, proteins never function in isolation. Rather, a protein is typically surrounded by a membrane (itself a simpler biopolymer) around which is an aqueous environment. Such a configuration typically involves 105 atoms, or about 30 times the number of atoms in a single protein molecule. The time required to perform a numerical simulation of behavior at the molecular level increases with the square of the number of atoms involved (since all pair-wise atomic interactions must be computed), and so a simula- tion of a protein in its natural environment takes on the order of 1000 times as long as that for a protein in isolation. The computational power available to biologists is increasing by orders of magnitude because of faster hardware such as massively parallel processors and better algorithms (e.g., multiple time scale and cellular multipole methods). The result is that larger, longer, and more detailed simulations are becoming possible. For example, more computational power may permit the correct quantum-mechanical

CO~PUT~G ME FUTURE simulation of biopoly~er behavior. Long time simulations (i.e./ sim- ulations of ~ second or so of behavior) together with advanced algo- r~hms for predicting structure may finally enable the prediction of protein structures from their amino acid sequence. Simulations of large molecular assemblies ~i] advance the rational design of new

LOOKING TO THE FUTURE OF CS&E 75 drugs and allow the pharmaceutical industry to speed up develop- ment processes that today cost many millions of dollars. Another field that will gain from CS&E advances is neurobiolo- gy. Neurobiologists are concerned with describing and understand- ing brain activity at the neural level. An example is the problem of how neural activity encodes visual images in a brain area called the visual cortex. This area actually provides multiple encodings of any image that is seen. One such encoding of an image, as observed through so-called voltage-sensitive dyes in a monkey brain, involves about 106 neurons with 109 indirect synaptic connections to the reti- nas of the eyes that develop during the first few months of the mon- key's life. Massively parallel computers are very well suited to simu- lating this development process and may make possible simulations of large networks in which neurons are modeled as nonlinear, dy- namical, spiking units. These simulations may shed light on the hotly debated question of how temporal relationships between spikes contribute to information processing in the brain. Further opportunities for advances in structural biology through computational methods arise in connection with the use of two- and three-dimensional nuclear magnetic resonance (NMR) spectra for pro- tein structure analysis. Such spectra yield information on the inter- atomic distances between the large number of atoms in a biopolymer. Together with information regarding the native forces acting between these atoms (today not well known), knowledge about the interatom- ic distance constraints permits the determination of protein structure. It is expected that in the next decade the structure of many biopoly- mers will be obtained through this technique, which relies both on NMR measurements and on advanced computation. Diagnostic techniques such as magnetic resonance imaging are based on physical processes that need to be better understood if the diagnostic method is to achieve resolution on the scale of a single biological cell. A high level of understanding is reached if the mea- suring process for a sample can be simulated in its entirety, a task that requires monitoring the nuclear spin precession of millions of diffusing water molecules over many precession periods of their nu- clear spins. Again, such simulation currently requires many days on today's fastest computers. Even faster hardware and algorithms may allow briefer imaging periods and images with more detail. A final example of a rapidly developing role for CS&E in the life sciences involves biological databases. Reliable network access needs to be provided and use of computational resources promoted through tutorial documentation and workshops. CS&E researchers, through collaborations with biologists, should provide better opportunities

76 COMPUTING THE FUTURE for information "mining" (i.e., examining the data in search of unex- pected or unanticipated relationships). New opportunities would arise if the structural and sequence databases were maintained at one location, allowing the further development of existing tools (e.g., Gel- Reader (a creation of the National Center for Supercomputing Appli- cations), GenWorks, GCG, and Intelligenetics) to include cross-refer- encing and cross-linking of features between the databases. It is expected that geneticists and structural biologists working with the Human Genome Project will consult computer-based data- bases much more frequently than today for data analysis for example to identify genetic disorders. This data analysis will require distrib- uted computing in which the computing-intensive tasks are performed on a supercomputer and the interface is handled on a graphics work- station, where high-speed rendering and digital video will be indis- pensable. Supporting the necessary distributed computing environ- ment will be software, such as the Data Transfer Mechanism (DTM) software developed at the National Center for Supercomputing Ap- plications, that allows data exchange between a wide range of com- puters in a machine-independent manner. The preceding possibilities for computational biology will depend on the availability of advanced computer technology, including very large massively parallel computers deployed at the national super- computer centers and the national laboratories, smaller models of scalable parallel computers operating at many sites for program de- velopment and testing, concurrent computation exploited across net- works of workstations, and new visualization techniques (perhaps making use of digital video) for data postprocessing and interactive computation). Commercial Computing As mentioned above, academic CS&E has often kept the commer- cial and business world at arm's length in part because academic computer scientists and engineers tend to focus on the creation of the science, whereas business people tend to be interested in low-risk approaches that emphasize the best practice with currently available technology. But it is important to realize that in their quest to exploit business opportunities, industry and commerce (and often govern- ment agencies as well) are another rich source of intellectually chal- lenging problems. Manufacturing and service firms have driven the demand for computer-aided design, on-line transaction processing systems, and specialized portable information appliances; today, Amer- ican business is exploring the use of very sophisticated computing technology. By directly addressing the information demands of the

LOOKING TO THE FUTURE OF CS&E 77 business and commercial environment, academic CS&E research can be invigorated by new and demanding challenges, at the same time making contributions that improve the well-being of our society. As in the case of scientific computing, it is important to distinguish between the relatively routine uses of computers in organizations (e.g., spreadsheets and word processors on personal computers, large ac- counting programs or inventory control systems on mainframes) and the uses of computers that extend the state of the art. However, where the challenges of large-scale scientific computing center on the need to perform huge numbers of floating-point calculations and to display huge amounts of data in comprehensible form, the challenges of large- scale commercial computing arise from the need to: · process and store huge amounts of data, often with relatively little processing for each piece of data. In many ways, commercial computing is limited by the speed of input and output rather than by the speed of computation itself. · use computer systems with very high reliability and availabili- ty. Fault-tolerant systems, first used in life-critical applications (e.g., real-time flight control), have been spreading to applications such as banking, where the cost of down time is measured in megadollars rather than lives. · eliminate the deleterious effects of work dispersed across or- ganizational boundaries. In the course of ordinary business, workers must often interact with people who work in other locations. In addition, work styles and procedures may be different. Computing technology (e.g., e-mail) to promote and facilitate interaction is be- coming more common. The commercial environment today is characterized by globaliza- tion and worldwide competition, severe time and productivity pres- sures, and a rapidly changing and thus unpredictable business envi- ronment. Computers often play an important and even enabling role in managing this fast-paced environment; nevertheless, the need to respond rapidly will make even greater demands on computing tech- nology to be easily and quickly adaptable to new circumstances. Some important problem areas and promising research directions for CS&E with respect to business and commercial computing are highlighted below. Another major and relevant area, "better" soft- ware engineering, is discussed in Chapter 3. Model Management and Decision Support Modeling is an essential tool of modern business. Companies make decisions on the basis of likely or expected outcomes of possi

78 COMPUTING THE FUTURE ble courses of action, and computer models are more and more fre- quently essential to forecasting these outcomes. At present, models generate numbers without explicit articulation of the underlying prem- ises. Through great manual effort, the premises can be coupled to model output, but the farther one gets away from the original model (e.g., as one model feeds another one that there feeds a third one, and so on), the more likely it is that the premises will be lost. Users need the ability to inquire easily about the assumptions that underpin the analysis at any level, and to change these assumptions to test various scenarios. Thus tools that facilitate convenient model creation and management (e.g., by making assumptions obvious and explicit and easy to change at any level in a chain of models) would contribute a great deal to effective decision-support systems. Easily Usable Software Increasingly cheaper computer systems have led to the prolifera- tion of information technology into many offices. With such prolifer- ation has come greater access, by people (from chief executive offic- ers to beginning typists) who cannot be assumed to have the willingness or patience to develop serious computer expertise as a prerequisite for using the computer. Numerous computer users feel daily pain, anxiety, and frustration as they struggle with clumsy interfaces, in- comprehensible error messages, and technical details that are irrele- vant to what they want to do (Figure 2.3~. In the words of Mitchell Kapor (founder of the Lotus Development Corporation and principal HoW WAS VOICE MAIL ) ~= ~E W - AD C~15S~ AR£~00 ALL ~_ \ uprise' SEf no 05E OUR UEW ~ ) WAS VE~ ' - . it. S - ~M? 1 1 600D. SHE LMOSr ~1 HAD ~E ~ I COUVINC£D \\rMAr MICE ~ 1\ SAIL I n WILL ~ MAKE . 3 ~1fE EASIER. L~ ~ _ (dear HA~ED=~ SHE WADED Off - ~ ~-~ ThE GAL. FIGURE 2.3 The potential down side of "greater" functionality. Although the technology involved in this example is voice mail, the lesson of the exam- ple applies at least as well to computing technology. Copyright @) 1991 by North America Syndicate, Inc. Reprinted with special permission of North American Syndicate, Inc. '~1

LOOKING TO THE FUTURE OF CS&E 79 architect of the Lotus 1-2-3 spreadsheet), the lack of usability of soft- ware and the poor design of programs are the secret shame of the computer industry. Indeed, even as information comes to play a more and more im- portant role in society, the pervasiveness of difficult-to-use comput- ers in offices and homes is increasingly the factor limiting the wide- spread use of information technology. Attention to design resulting in computers that are almost as easy to use as telephones or fax machines would have an enormous impact on the number of people using computers and on the variety and effectiveness of different business applications to which computers can be applied. Software Development Metrics and Modeling Software development metrics and modeling present vast research opportunities. To a large extent these topics have "fallen through the cracks" because computer scientists often view these areas as the domain of management, while management scientists often view them as the domain of computer science. Rigorous approaches that ad- dress a broad array of software development issues holistically would be a valuable theme for research investigation. Technology for Interoperation Most companies must interact with other companies, and divi- sions within the same company must interact with each other. Al- though it is in principle easier to impose a single computing struc- ture on all divisions within a company, in practice it often turns out that intracompany computer-mediated communication is nearly as difficult as intercompany communication, which is a major problem in business today. Indeed, while the computers of one company can usually exchange strings of letters and numbers with the computers of another company, fax rather than computer networking is the standard means of interchange for documents involving images, notes, and for- matted text. (The writing of this report provides a good example of difficulties encountered in this environment; see Box 2.71. In fact, document interchange architectures exist that faithfully represent multimedia documents and that will eventually supplant fax, but the simple example shows the need for high-level machine-independent representation of information. With databases, processors, and net- works separately installed and maintained in different offices, auto- mated data conversion, interoperable network protocols, and trans- portable software systems are necessary to provide dynamic reconfig

80 ......... ... ......... .. . .......... . ~:. ~ ............... ~ ~ : : ~:. . ~. ~..... . ~. ~,~ · ; . ~ ~,.,~ ~ ., ,. I , . ~,i, COMPUTING THE FUTURE ...... .....~.. ....... . . . . oration. Research aimed at creating powerful but flexible technolo- gy and standards that facilitate interoperation among heterogeneous computer systems and convenient electronic data interchange will be a boon to all computer users, but will be especially valuable for busi- ness applications.l9 Although in many cases the lack of interoperability is a problem of choice for manufacturers that opt for proprietary architectures and data formats, the development of good standards nevertheless re- quires technical expertise. For example, interconnected devices or software conforming to a set of poorly designed or inconsistent stan- dards may exhibit unanticipated interactions or behaviors, possibly as the result of timing problems. As a result, even if standards are consistent, devices or software conforming to these standards may require excessive execution time. Design flaws in a given standard may become apparent only when it fails in some particular imple- mentation. For critical implementations, dynamic recovery techniques

LOOKING TO THE FUTURE OF CS&E 81 may be necessary to restore proper operation in such an event; these techniques themselves may become part of the standard. Collaborative Work Computers can help to facilitate cooperative work efforts among many individuals in geographically dispersed offices, reducing or eliminating the disruptive effects of distance on the work process. Commonly known as "groupware," such computing technology might, for example, provide ways for collaborative annotation of a single document, facilitate electronic interaction by keeping track of differ- ent threads of discussion in e-mail messages, or support decision- making processes in large groups. The development of groupware customized to the requirements of individual offices and different work styles will require careful attention to the social context in which such groupware will be used (Box 2.8~. The Electronic Library The dream of the electronic library dates from the very beginning of the computer age. Imagine accessing at modest cost, from home or work, the contents of a local library and, in case of need, escalat- ing to grander sources right up to the Library of Congress. (Perhaps the Library of Congress would itself be virtual" the networked ag- gregate of all the libraries in the land.) Enticing fragments of the dream have been realized, in the areas of cataloguing, storage, and search. Many general libraries have placed their catalogues on line. Regional library consortia with shared remote catalogues have come into existence. Documents from selected corporations are accessible electronically from commercial services that provide full-text search capability. Bibliographic databases exist in profusion. Chemical and biochemical information may be retrieved by performing pattern matches on formulae. Electronic libraries promise to accommodate information formats that fit uncomfortably in traditional libraries, e.g., images, sound, and small documents such as letters. All are more manageable in electronic media than in print. And the electronic library is the natu- ral home of multimedia "documents" and electronically published journals, which are just beginning to appear in the marketplace.20 To achieve the fully electronic library, many imposing technical problems will have to be solved: acquisition, indexing, storage, re- trieval, transport, presentation, and performance. There are chal- lenges enough for almost every branch of CS&E (Box 2.9~.

82 COMPUTING THE FUTURE ~ . , .. , . ~ , . ~ ~ ~ .. ~ ~ ~ . , . ~ . . . ~ ~ ~ ., , ~ ., ~ . ~ ...... i ..... i ~ . i ~ ~ ~ i . . ~ . ~ ~ . ~ ~ ... ~ ~ ~ ~ . ~ . i. ~ . . i. ~ .......... ~. ~. ~.~ ~. ~ . ~ ~ ... ~ ~. ~.~. ~it ~ i. ~. ~. ~.~ ~ .. ~ ~ ~ ~ . ~ i ........ . A. , . ~ ~ ., , . ~ . ~ . ~.,~ ., ~ .... ~ ~ i i. :. ~. ~ ~ i ........... i. i i. ~ ~ ~.~ ~ . ~ ~ ~ ~ . i. i . . . . . . . ...... ......... . . .......... . .... . , i. ~ . . ~ . . ~ i. i, ,, ~ :, .~ i, , ,...,., i, .... .... ........ . ...... . ........... ..... . .. . ~. ~. ~i i. ~ . ~ ~. ~ .~ i ~ i. ~ ~ i .~ i. ,.: ~ : - .~ ~.: ... . . ... ............. ....

LOOKING TO THE FUTURE OF CS&E 83 Input Even attending only to printed matter, the electronic librarian is faced with source materials of two radically different kinds: printed documents and an electronic source for digital typesetting. The only feasible way to enter printed material is optical scanning, which cre- ates digital page images. But for retrieval other than simple regurgi- tation of pages by number, one needs digital text. Making optical character recognition practical on a library scale requires advances in natural language processing, to say nothing of new special-purpose architectures for image and pattern recognition. Documents originating in electronic form are not an unmixed blessing, either. There are hundreds of distinct encodings input on diverse media for diverse typesetting systems. Extraction of "mean

84 COMPUTING THE FUTURE in"," even at the level of a simple stream of text characters, is a daunting task. Furthermore, typesetting is not typically fully auto- matic. Title pages, page numbers, figures, and proofreading correc- tions are likely to come from separate places, and so there may be Rio complete electronic version of a document. The printed version may have to serve as a guide for the reconstruction of a full electronic document from partial electronic sources. We thus have the problem of correlation of multiple texts. Retrieval A document in a library is useful only insofar as information can be extracted from it, either by direct retrieval or by processing. In- formation retrieval systems usually depend on indexing (manual or automatic) to home in on documents, and then perhaps on full-text scanning to find exact information. The suitability of various index- ing and scanning techniques depends strongly on scale; there is much room for innovation and experiment. At a higher level, the quality of retrieval should be enhanced by "text understanding." Still not com- monly used today, statistical methods for analyzing documents are likely to be the first scalable techniques. (For statistical analysis, the details of language are unimportant arid sample size is a boon, not a bane.) Understanding at the level of identifying certain formal parts of a text, such as titles and table of contents, will be important for . . nc being purposes. Searching, even among indexed documents, on a library scale is a challenge for both architecture and algorithms. And searching for Contextual matter visual or audio is almost virgin ground. The possibility of novel and massive search techniques, however, is a prime motivation for developing the electronic library. In a print library, images can be found only by leafing through the holdings. Presentation Electronic libraries promise simultaneous availability to all read- ers, access at a distance, and easy capture of relevant passages. Off- setting these advantages is the fact that electronic presentation of substantial amounts of static information is rarely as satisfying as print, either for browsing or serious reading. That judgment may be altered by the advent of new modalities, such as hypertext,2i for navigating documents. One thing is certain: the availability of large bodies of text for experimentation will stimulate creative new ways to present and interact with the documents and with search proce

LOOKING TO THE FUTURE OF CS&E 85 cures, and bring new models to the attention of CS&E. How, for example, can the enormous numbers of "hits" that automated search- es often return be summarized for effective further selection? Or again, how with reasonable speed can a reader "see" a whole book as effectively as one does today by leafing through it? Performance It is easy to conceive of automatically "reading" whole books of text over a high-speed fiber-optic network; a book is one or a few megabytes of textual data, and a megabyte takes ten milliseconds to transmit at gigabit rates. It is less easy to imagine, say, an art book coming as page images at a megabyte apiece. Issues of data com- pression akin to those present in high-definition television come to the fore, in storage as well as in transmission. Memory hierarchies, probably distributed, will be needed for economical storage of infor- mation, the demand for which differs by many orders of magnitude. Simultaneous searches on behalf of multiple readers pose a challenge to information retrieval technology, likely involving massive paral- lelism, distributed computing, and scheduling. The matter of survival poses problems, too: how can a library that archives material for the ages exploit technology that goes utter- ly obsolete in a decade? And how can indexing and retrieval strate- gies, which will surely evolve rapidly in the light of experience, be introduced gracefully? BROADENING EDUCATIONAL HORIZONS IN CS&E A broader research agenda for the field requires people willing to engage in a wider scope of activity than they have been accustomed to pursuing. Thus changes in the educational milieu of both gradu- ate and undergraduate CS&E education will be necessary if a broad- er agenda is to win wide acceptance. Computer scientists and engi- neers may not need to fully master other disciplines, but they will need to know enough about other domains to understand the prob- lems in those domains and thus how to apply their own unique ana- lytical tools to their solution. Employment opportunities may well be wider for broadly educated computer scientists and engineers than for those who know only about computing per se. In addition, CS&E education will need to reexamine some of the values with which it socializes its graduates. At present, CS&E stu- dents are led to believe that doing "pure" CS&E research is the high- est pinnacle to which all good students should aspire. Values consis

86 COMPUTING THE FUTURE tent with a broader agenda would teach budding computer scientists and engineers that in the information age, they should learn to make contributions to a wide range of fields and problem domains. And finally, CS&E has a responsibility to help those in other areas to understand the implications of the new information age. Thus it must take a broader view of its responsibilities for service education to practitioners in other disciplines and problem domains. Chapter 4, "Education in CS&E," discusses these issues in greater detail. A SPECIAL ROLE FOR UNIVERSITY-INDUSTRY-COMMERCE INTERACTION Ties between universities and the industrial and commercial world have a special role to play in promoting a broader agenda for both research and education. One overarching reason is that industry and commerce, concerned with developing products and services for cus- tomers who want their problems solved, assemble multidisciplinary project teams and research efforts with much greater ease than do universities with their discipline-centered departments.22 Computer hardware and software vendors have a vested interest in being responsive to the needs of the user community. Over the long run, software packages and hardware systems improve, or their vendors go bankrupt. Because of its need to gauge accurately what its customers are willing to buy, the computer industry can play a special role in specifying for computer scientists and engineers re- search areas that have relevance to the user community as a whole- general-purpose advances that make computers easier to use or more practically powerful from the perspective of individual users. A good example of such a role is found in the industry-driven spread of graphical user interfaces. (Of course, such contributions will be pos- sible only with the involvement of people whose vision can tran- scend narrow company perspectives.) Commercial users of computers can also help to define a broader research agenda that is relevant to particular segments of the user community. Problems that arise in specific applications are often an instance of a more general and incompletely understood issue with substantive intellectual challenge. Research undertaken to solve the specific problem may well shed light on the more general issue. Fur- ther, by working with the ultimate end users, academic computer scientists and engineers can help those users to better understand their future needs in their particular settings and to develop technol- ogy that better meets those needs.

LOOKING TO THE FUTURE OF CS&E 87 On the educational front, both the computer industry and com- mercial computer users have an important role to play in broaden- ing. As the need emerges for businesses of every possible descrip- tion to manage information of all types, individuals who understand the possibilities of computer-mediated management of such informa- tion will be in demand by both industry and users. This imperative has fueled the development of a host of computer-related programs in information sciences, information systems, management sciences, and so on, in addition to programs in CS&E. However, a broadly educated CS&E graduate is most likely the person who will under- stand how or whether existing technology can be adapted to meet existing needs and how to specify and design new technology that may be required. Thus a move by industry and commercial users to widen the employment opportunities they offer to CS&E graduates beyond the narrow computer-related jobs that CS&E graduates now fill may well benefit these firms as they move into the 21st century. These issues are discussed at greater length in Chapter 4. PREREQUISITES FOR BROADENING Although the committee found a reasonable consensus that aca- demic CS&E would benefit from a broader agenda, the inward-look- ing and applications-avoiding traditions of the field are likely to make implementation of a broader agenda difficult. The present structure of CS&E as an academic discipline often impedes the participation of faculty members in applications-oriented or interdisciplinary work. Reorienting academic CS&E to embrace interdisciplinary or applica- tions-oriented work will require serious attention to several factors, including the following:23 . Adequate departmental or university support. The research hori- zons of many faculty (especially junior faculty) could be expanded if they believed that good applications-oriented or interdisciplinary re- search would lead to tenure or promotions. Senior faculty, even though protected by tenure, are not immune to the pressures of their colleagues, and if other departmental faculty believe that such work is not intellectually worthy of attention, they too may be inhibited from pursuing such activity. Many CS&E departments believe that the evaluation of interdis- ciplinary research is daunting when assessment of work related to other fields is required. Even the definition of a peer in interdiscipli- nary research is unclear. In the words of H.E. Morgan, "Is a peer a person knowledgeable primarily in the technical aspects of the ap

88 COMPUTING THE FUTURE proach that is to be applied, or is both technical expertise and a broad knowledge of the field encompassed by the hypothesis and questions to be addressed also a requirement for designation as a peer?"24 When even the general characteristics of those who should be making assessments are unclear, departments may well shy away from encouraging work that requires such assessments. · Provision of appropriate funding. Funding to pursue interdisci- plinary or applications-oriented research is certain to encourage such work, especially in times of tight research budgets. Partly because of its novelty, interdisciplinary or applications-oriented research is of- ten seen by the typical funders of research as high-risk or irrelevant. In the absence of funding specifically targeted to such work, more traditional, discipline-oriented work often appears the safe route to follow for seekers of research funding. · Strong communication between CS&E and other problem domains. The sine qua non of most academic work is the published paper or book. But interdisciplinary or applications-oriented work often lacks suitable forums that will provide appropriate attention. The solution of a given problem may require collaboration between researchers in CS&E and another field, but journals in the other field may be inter- ested only in the results relevant to that field, while CS&E journals may be unwilling to give space to describing details of the other field relevant to the solution of the problem. Thus special outlets for such work may be necessary. · Common educational experiences and mutual respect. Collabora- tions between researchers in CS&E and other disciplines and applica- tions areas are most successful when computer scientists and engi- neers have a modicum of knowledge about those other areas and disciplines, arid when people from those other areas have some fa- miliarity with current concepts in CS&E. Moreover, each side of the collaboration must respect the basic intellectual interests of the oth- er the interest of computer scientists and engineers in the challeng- ing CS&E aspects, and the interest of other party or parties in the problem at hand. Without such respect, it is all too easy for the computer scientist or engineer to be regarded merely as a hired hand responsible for the intellectual equivalent of washing test tubes. · A broader definition of research. Even when interdisciplinary research is considered, prevailing notions in the academic CS&E com- munity limit the definition of research to fundamental intellectual work that underpins a product or may have no connection to any product now or in the future. Thus academic CS&E research may

LOOKING TO THE FUTURE OF CS&E 89 involve theoretical work and proof-of-principle and laboratory pro- totypes, but nothing closer to product application. In fact, a great deal of intellectually substantive work and inquiry can be associated with "productizing" a concept. As an example, chemical engineering and, to a lesser extent, chemistry both include within their defini- tions of Ph.D. research work that improves chemical manufacturing processes. Certain challenging computing problems (e.g., the con- struction of large-scale software systems) have solutions that in prac- tice often do not require a single key insight but rather many small ideas solving subproblems across many areas. Such problems are best solved by people with breadth, but breadth often comes at the expense of the depth that characterizes most traditional research. In addition, traditional notions of academic research call for work in which students and faculty are expected to make their mark as individual scholars and researchers, rather than as members of teams or groups (as would better characterize an industrial environment). Since many interesting and substantive problems in CS&E involve as a primary or secondary activity the construction of large systems that require extended efforts by large groups, those with interests in such areas may be left at a disadvantage. · Leadership. By definition, the leaders in any given field play a major role in setting the tone and character of that field. The judg- ments and opinions of these leaders determine the standards to which other participants in the field are held. Thus, expanding the bound- aries of CS&E research will require the intellectual leaders in the field to proselytize vigorously in favor of such expansion. They must lobby for departmental or university support of a broader agenda. And, most importantly, they must engage the public policy process on behalf of change with an intensity and persistence that they have not often demonstrated in the past.25 As a general rule, individuals can participate in or contribute to the public policy process through either the executive branch or the legislative branch. Interaction with the executive branch is especially meaningful when it involves sustained effort (e.g., serving as a pro- gram officer), simply because such service generally involves deci- sion-making authority. Interaction with the legislative branch is po- tentially more profitable for the field, since the legislative branch determines actual funding levels. However, it is often much more frustrating, because the Congress is often unable to consider the full implications of various proposals from the scientific community. Box 2.10 describes some of the opportunities available to computer scien- tists and engineers to engage the public policy process.

So COMPUTING THE FUTURE it, ........................................................................................................................... ............ ... ............................................................................................................. SUMMARY AND CONCLUSIONS Broadening academic CS&E offers benefits from several perspec- tives. From the perspective of the field itself, extending its bound- aries will identify new challenges and offer new opportunities for students and research support. Those in other areas and fields will also benefit from the application of state-of-the-art hardware and software technologies customized to their specific problems. And finally, the interaction of CS&E with other disciplines is likely to lead to intellec- tual insights and developments in both CS&E and those other disci- plines that would not otherwise be possible. The broadening of CS&E will lead to a flowering of new ideas, advancing the knowledge of humankind as well as promoting the growth of industry and the economy. Intellectually substantive CS&E issues and themes can be found in many problem domains, from biology and the earth scienc- es to commercial computing and electronic libraries. But broadening the CS&E field will require concerted university and funding agency support, educational programs to support a broader conception of the field, and a rethinking of what constitutes research for an aca- demic computer scientist or engineer. NOTES 1. See Robert M. White, "The Crisis in Science Funding," Technology Review, Vol- ume 94(4), May/June 1991, p. 47. Lest the reader believe that the need to justify science on the basis of its social and economic return is a new sentiment brought about today by increasingly tight budgets and short-sighted political leaders, it is interesting to recall that Vannevar Bush, in the July 1945 document widely regarded as the semi- nal statement of philosophy underlying creation of the National Science Foundation, argued for the support of science on the basis of its ability to contribute to society. Advances in science when put to practical use mean more jobs, higher wages, short- er hours, more abundant crops, more leisure for recreation, for study, for learning how to live without the deadening drudgery which has been the burden of the common man for ages past. Advances in science will also bring higher standards of living, will lead to the prevention or cure of diseases, will promote conservation of our limited

LOOKING TO THE FUTURE OF CS&E 91 national resources, and will assure means of defense against aggression.... [S]ince health, well-being, and security are proper concerns of government, scientific progress is of Nrital interest to government. Without scientific progress the national health would deteriorate; without scientific progress we could not hope for improvement in our standard of living or for an increase in the number of jobs for our citizens; and without scientific progress we could not have maintained our liberties against tyranny. (Van- nevar Bush, Science-the Endless Frontier, NSF 90-8, National Science Foundation, Washington, D.C., 1945/1990, pp. 10-11.) 2. For example, mathematically rigorous investigations of the Mandelbrot set were begun only after Benoit Mandelbrot had examined many computer-generated visual- izations of the set. Mandelbrot observed that the islands present in low-resolution pictures were apparently not present at higher resolutions. As a result of these exam- inations, Mandelbrot conjectured that the set was connected. A rigorous proof of this conjecture has subsequently been developed. 3. This point was reinforced at the recent CSTB Workshop on Human Resources in CS&E, a report on which is forthcoming. 4. An Association for Computing Machinery (ACM) position paper notes that "analyzing how computer science and engineering R&D can assist with solving national and in- ternational needs can result in new opportunities and directions, such as increasing funding and more diverse funding sources." See Association for Computing Machin- ery, "The Scope and Directions of Computer Science: Building a Research Agenda," Communications of the Associationfor Computing Machinery, Volume 34(10), October 1991, p. 123. 5. The basic data for this claim are given in Table 1.1. The agencies in question include the Departments of Education, Justice, Agriculture, Health and Human Servic- es (including the National Institutes of Health), Labor, State, and Veterans Affairs; the Smithsonian Institution; the Nuclear Regulatory Commission; the Tennessee Valley Authority; the Arms Control and Disarmament Agency; and the International Trade Commission. Even if the National Institutes of Health is omitted from this list, the research budgets for the remaining agencies still account for $3.4 billion. 6. At 38 key institutions, academic computer science was seeded by a number of different disciplines, including mathematics, electrical engineering, business, physics, psychology, physiology, linguistics, philosophy, cognitive science, and management information systems. (See Lois Peters (Rensselaer Polytechnic Institute) and Henry Etzkowitz (State University of New York at Purchase), "The Institutionalization of Academic Computer Science," p. 5. Paper presented at the Study of Science and Technology in the 1990s, a joint conference of the Society for Social Studies of Science and the European Association for the Study of Science and Technology, Amsterdam, November 16-19, 1988.) Even today, the majority of CS&rE faculty who have Ph.D s received them in other fields (as noted in Table 8.11 in Chapter 8), although projecting forward from the approximately 300 new Ph.D.s in CS&E who took faculty positions in the 1990-1991 academic year, this may change soon. 7. Some of the intellectual issues in this area are reported in Eric S. Lander, Robert Langridge, and Damian M. Saccocio, "Computing in Molecular Biology: Mapping and Interpreting Biological Information," Communications of the ACM, Volume 34(11), No- vember 1991, pp. 33-39. This article describes some of the key computational challeng- es in molecular biology as discussed by participants in a CSTB workshop. 8. National Research Council, Physics Through the 1990s: Scientific Interfaces and Technological Applications, National Academy Press, Washington, D.C., 1986, p. 121. 9. Association for Computing Machinery and the Computing Research Associa- tion, Strategic Directions in Computing Research, ACM Press, 1990, pp. 1-2.

92 COMPUTING THE FUTURE 10. David Cries, Terry Walker, and Paul Young, "The 1988 Snowbird Report: A Discipline Matures," Communications of the ACM, Volume 32(3), March 1989, pp. 294- 297. 11. Association for Computing Machinery, "The Scope and Directions of Computer Science," Communications of the ACM, Volume 34(10), October 1991, pp. 121-131. 12. This approach to building a research agenda has much in common with one discussed in an ACM position paper that argues for a strategy that "proposets] a set of goals and needs, and recommend[s] computing research that can help attain those goals." See Association for Computing Machinery, "The Scope and Directions of Com- puter Science: Building a Research Agenda," Communications of the ACM, Volume 34(10), October 1991, p. 122. The use of "computing research" in this reference is equivalent to the use in this report of "CS&E research." See also John Rice, "Is Com- puting Research Isolated from Science?", Computing Research News, Volume 2(2), April l990,p. 1. 13. The definitions used by the National Science Foundation are the following (Na- tional Science Foundation, Federal Funds for Research and Development: FY 1988, 1989, 1990, NSF 90-306, NSF, Washington, D.C., 1990, pp. 2-3): "Research is systematic study directed toward fuller scientific knowledge or under standing of the subject studied. Research is classified as either basic or applied accord- ing to the objectives of the sponsoring agency. In basic research the objective of the sponsoring agency is to gain fuller knowl- edge or understanding of the fundamental aspects of phenomena and of observable facts without specific applications toward process or products in mind. In applied research the objective of the sponsoring agency is to gain knowledge or understanding necessary for determining the means by which a recognized and spe- cific need may be met. Development is systematic use of the knowledge or understanding gained from re- search, directed toward the production of useful material, devices, systems, or meth- ods." The U.S. definition of "basic research" as research without application in mind stands in marked contrast to the Japanese notion of "basic research" as research that is basic to the future of industry. See David Cheney and William Grimes, Japanese Tech- nology: What's the Secret?, Council on Competitiveness, Washington D.C., February 1991, p. 4. 14. Indeed, a powerful argument can be made that the linear model of basic re- search leading to applied research, applied research leading to development, develop- ment leading to product manufacture, and manufacture leading to sales is highly oversimplified and in many ways downright misleading. Product innovation rarely resembles the popular view of one revolution followed by tedious development (e.g., invent the transistor, and the rest is reduction to practice). Rather, the process more resembles something like this: invent the transistor, then invent technology to place 10 transistors on a chip, then invent technology to place 100 transistors on a chip, . . . then invent technology to place 100,000,000 transistors on a chip, and so on. This model, often called the cyclic development model, is discussed in R.E. Gomory and R.W. Schmitt, "Science as Product," Science, Volume 240, May 27, 1988, pp. 1131- 1132, 1203-1204. 15. See, for example, Computer Science and Technology Board, National Research Council, The National Challenge in Computer Science and Technology, National Academy Press, Washington, D.C., 1988, pp. 34-35.

LOOKING TO THE FUTURE OF CS&E 16. As used in this report, the terms "interdisciplinary research" and oriented research" are not synonymous. ~ ~~ ~ ~~ ~ ~ 93 "applications lnterdisciplinary research is research that requires and draws on intellectual contributions from CS&E and some other discipline together. Applications-oriented research is CS&E research pursued in the context of some specific problem that may well be fully understood from an intellectual stand- point but whose scale or nature may overmatch the capabilities of current computing technology. 17. The National Research Council's interim report on EOSDIS noted the synergy possible in a collaboration between the earth sciences and CS&E, arguing that "EOSDIS, as it evolves, must maintain the flexibility to build rapidly on relevant advances in computer science and technology, including those in databases, scalable mass storage, software engineering, and networks. Doing so means that EOSDIS should not only take advantage of new developments, but also should become a force for change in the underlying science and technology where its own needs will promote state-of-the-art developments." See National Research Council, Panel to Review EOS- DIS Plans: Interim Report, Washington, D.C., April 9,1992, p. 3. 18. For example, the American Express Company and Schlumberger, both stal- warts of the American business community, will be among the first organizations to purchase a massively parallel computer recently offered for sale by the Thinking Ma- chines Corporation. Such purchases indicate that problems faced by these firms can- not be solved economically with routine computing technology. See John Markoff, "American Express to Buy Two Top Supercomputers," New York Times, October 30, 1991, p. C-7. 19. However, it should also be noted that technology changes rapidly enough and the lag time in making purchases is long enough that it is often difficult for any standard to be widely used and accepted. Still, electronic data interchange of various types is growing rapidly. 20. For example, as this report goes to press, the American Association for the Advancement of Science and the On-Line Computer Library Center are about to launch an on-line, peer-reviewed journal titled "The Online Journal of Current Clinical Tri- als." Manuscripts will be submitted, reviewed, and published in electronic form to as great a degree as possible. See Joseph Palca, "New Journal Will Publish Without Paper," Science, Volume 253, September 27, 1991, p. 1480. 21. Hypertext is a way of presenting text that is not structured linearly. A hyper- text document has cross-references and other links that allow the reader to peruse the document in an order that makes sense for his or her needs at the time. 22. As the value of interdisciplinary work is recognized, it may become easier to perform interdisciplinary research in universities. The NSF-sponsored engineering research centers and the science and technology centers appear to represent a positive step this direction. 23. The first four factors listed are inspired by a presentation in National Research Council and Institute of Medicine, Interdisciplinary Research: Promoting Collaboration Between the Life Sciences and Medicine and the Physical Sciences and Engineering, National Academy Press, Washington, D.C., 1990, pp. 12-15. 24. H.E. Morgan, "Open Letter to NIH- Review of Cross-Disciplinary Research," in The Physiologist, Volume 31(April), 1988, pp. 17-20. Cited in National Research Council and Institute of Medicine, Interdisciplinary Research: Promoting Collaboration Between the Life Sciences and Medicine and the Physical Sciences and Engineering, National Academy Press, Washington, D.C., 1990, p. 12. Although the letter concerns interdisci- plinary research in the life and health sciences, the moral is the same. 25. An example of past indifference to participation in the public policy process is evident in the experience of NSF's Computer and Information Sciences and Engineer

94 COMPUTING THE FUTURE ing Directorate, which provides a considerable percentage of research funding for academic CS&E and thus exerts a substantial influence over the field. Naturally, NSF looks to the field to provide knowledgeable individuals who can help to shape a research program and make reasonable decisions about funding directions. But, ac- cording to NSF officials, finding appropriate individuals willing to fill staff and high- level management positions within the CISE Directorate has been extraordinarily diffi- cult. Why is it difficult? Some people argue that a period of inactivity in research of even a few years can place an individual at considerable disadvantage. Without special provisions such as "exit grants," faculty may be hesitant to enter public service even temporarily. (An "exit grant" is a grant provided to program officials returning to academia that enables them to restart their own personal research programs and thus facilitates their reentry into academic life. Such grants may be provided formally through a designated program, or informally through a mutual understanding of the participants involved.) Others argue that the salaries paid for government service tend to be lower than those that could be earned by qualified computer scientists and engineers working outside of government. Still others contend that most CS&E de- partments are so "thin" that the departure of an individual for a few years could cripple such a department's ability to cover an important subarea of CS&E. Finally, the relative youth of academic CS&E tends to increase the number of individuals who, in earlier stages of their career, quite naturally and reasonably focus on their own personal research agendas.