Western scholars use the concept of the “dual economy” to explain how the Soviet Union fails to compete in the international industrial economy, while at the same time achieving parity with the United States in the development of strategic weapons. The USSR concentrates its scientific resources on the military side of its economy, and little that is developed is allowed to filter over to the civilian sector.
In the United States, a similar yet more complex dual economy has emerged. Two sectors, military and commercial, function side by side, often within the same corporate organizations. They compete to exploit the latest technologies. In some areas, the military sector stands in the lead. But in technologies with proven commercial value, the commercial sector has dramatically outdistanced the military.
Once ideas and information flowed freely back and forth between the commercial economy and the military sector. In fact, during World War II, America’s depressed commercial 260industries were able to rapidly produce planes, tanks, ships, and guns for the war effort.
However, both the needs and the styles of the two sides of the economy have gradually diverged. During the 1960s and 1970s, for example, the Pentagon sponsored the development of specialized technologies in the machine tool and metals industries. The resulting innovations may have served the military well, but the commercial fortunes of those industries sagged as foreign producers grabbed the technological lead in mass production.
Today there are two fundamental differences between military-sponsored and market-driven production in the United States. First, the military is interested in small production runs, such as 10 satellites, 100 strategic missiles, or maybe 10,000 tanks. Companies serving the mass market, however, must produce hundreds of thousands of cars, computers, chips, or phones each year.
Second, the military and its allied agencies deliberately control the flow of information.9, 10 In the past, security classification has been used to conceal the designs of specific weapons systems. Under the Reagan administration, the federal government has attempted to restrict the dissemination of basic research results, and it has concealed even the existence of billions of dollars of so-called black weapons programs.11 Though commercially oriented companies also restrict the flow of information—especially before products are released—they always depend for success upon public knowledge of their products’ capabilities. Often they deliberately make available design information as well.
Nowhere are the diverging paths of commercial and military industry more clear, and nowhere is the tension between the two sectors more important than in high-technology electronics. Both the data processing and components industries—to the extent that they can be defined distinctly—are the offspring of military research and purchasing during 261and soon after World War II. Through the mid-1960s, at least, government agencies shaped the growth of those industries and essentially had first rights to their innovations.
Following the invention of the microprocessor, however, in the early 1970s, the civilian chip and computer industries experienced rapid technological advances. They not only grew as important sectors of the U.S. economy, but they also began to transform the rest of the economy, through a combination of technological innovation and low cost manufacturing techniques.
At the same time, the military’s dependence on electronics increased, being used in items such as smart missiles, battlefield sensors, and strategic warning systems. Though the marketplace-paced civilian sector still was creating technologies of value to the Pentagon, commercial products such as war-simulating video games were employing more sophisticated hardware and software than the real thing. Worried that the Russians might absorb the new technologies first, the Pentagon began a series of initiatives designed to regain leadership in advanced electronics.
As in other industries, these initiatives emphasize secrecy and the development of specialized technologies. By diverting resources and controlling the flow of information, the American national security establishment is threatening the health of the commercial high-tech electronics economy. In fact, the tension between military and civilian priorities in chip and computer technology is more significant than competition with Japan in determining the long-term vitality of high-technology industry in the United States.
Rise of the Semiconductor Industry
Without the early involvement of the U.S. military, it is unlikely that semiconductor devices would have been developed and mass marketed so early. Yet this obscures the 262fact that if the military had guarded the secrets of the semiconductor technology more closely, the solid-state electronics industry would have floundered.
During World War II, the U.S. government sponsored a program to investigate the properties of silicon and germanium. At least thirty academic and industrial laboratories studied solid-state detectors for radar. Following the war, Bell Labs, the research arm of the telephone monopoly, applied this research to the creation of the first transistor, in December 1947.
Bell, interested in the application of its new invention to telecommunications rather than to weaponry, publicized the transistor and licensed the technology widely. Reportedly, officials at AT&T feared that the military would classify the invention, so Bell chose not to disclose it to the military until one week before its public announcement.
The military continued to sponsor semiconductor research, but its major influence was as a customer. In the first five years in which the transistor was available commercially, the armed services, particularly the Army Signal Corps, ordered millions of dollars of the new devices, promoting the development of production techniques.
Significantly, the military did not demand that its transistors meet specifications beyond those which commercial customers required. Thus, when the low-cost mass production of transistors became a reality, computer and consumer electronics manufacturers could easily substitute them for the bulkier, less reliable vacuum tubes.
During the 1950s the military supported a number of semiconductor research and development projects, but commercial firms set the pace of innovation. At the end of the decade, teams at both Fairchild Semiconductor and Texas Instruments independently invented the integrated circuit, which combined more than one transistor on a flake of silicon.
263The first integrated circuit chips were expensive, but the Air Force and NASA were willing to pay the price for the miniaturization of the guidance system on the Minuteman missile and the navigation system of the Apollo moon probe. Again government agencies did not require any specialized design, so once producers learned how to make large numbers of chips cheaply, commercial customers were able to purchase and use them.
The integrated circuit industry quickly outgrew its government roots. Though in 1962 all integrated circuit shipments were made to the federal government, six years later it only consumed 37 percent of all ICs sold.
The 1960s also marked the maturation of the San Francisco Bay Area’s semiconductor industry. The high-technology culture created by Stanford University’s provost and dean of engineering Frederick Terman proved a fertile ground for creating new high-technology firms. These ventures thrived upon the informal transfer of technical information. So many dissatisfied researchers and managers from Fairchild had established new semiconductor firms that by the early 1970s insiders started calling the area Silicon Valley. Today, it is the world’s leading center of semiconductor design technology.
Mil-Specs
At the same time, the military began to apply a bureaucratic mentality to semiconductor components. Companies like Intel, the Fairchild spin-off that introduced the first microprocessor and first commercialized memory chips, avoided military contracts. Ted Hoff, the Intel engineer generally credited with the invention of the microprocessor, later said:
Military electronics people would come to us and say, ‘We’ve got to have this special program because we want a certain high performance electronics part and the other commercial sons of 264bitches won’t build it for us.’ But what usually happened was that the military’s plans for a custom exotic device had about an extra two or three years’ delay added in because of all the extra negotiations, the paperwork, and the qualification process. By the time the military gets what it asked for, the part they wanted is usually at least two years behind the times because the commercial guys just keep right on moving.6
More than any other event it is the invention of the microprocessor that best marks the eclipse of the Pentagon by Silicon Valley’s commercial sector. In 1969, a time that the Pentagon was demanding that chip manufacturers build specialized circuits to rigid specifications, a Japanese electronics manufacturer asked Intel to build circuitry for a small electronic calculator. Rather than follow the calculator companies’ approach, Intel proposed the first single-chip processing unit, and it began delivery in 1971. Vastly more flexible than previous generations of integrated circuits, microprocessors quickly found a place in the commercial electronics industry.
The Pentagon, however, took years to make use of the breakthrough. Regis McKenna, Silicon Valley’s public relations guru, recently recalled the spread of Intel’s popular “8080” microprocessor, developed in 1974: “It was second-sourced by Japan in 1975, and by 1977 we had seen a Soviet reverse-engineered copy of it. But the joint Army-Navy task force didn’t approve it until 1979.”
The military had not overlooked the microprocessor. Rather, they were trapped by a cumbersome set of qualifications and testing procedures known as military specifications, or mil-specs. Mil-specs are designed to standardize components and to make sure that they’ll perform reliably in extreme conditions, such as very hot and very cold temperatures. The Navy set the first specifications for military semiconductors back in 1959. Most semiconductor companies accepted the inconvenience of mil-spec work in 265exchange for the higher profit margins associated with military contracts. But the slow moving mil-spec bureaucracy could not match the dynamism of semiconductor technology.
First, simply qualifying a commercial chip required as long as two years of re-engineering, testing, and paperwork. By the time chips met military specifications, they were obsolete. The commercial sector had moved on to a new generation of more capable, and even more reliable circuits.
Second, specified testing procedures did not keep up with the increasing complexity of the circuits. For example, as late as 1976 one military standard required the visual inspection of interconnects, the fine lines of metal, imbedded into chips, that connect the various circuit elements. This worked fine on small-scale integrated circuits, where the interconnects were about 2 mils wide. However, on large-scale integrated (LSI) circuits, where the interconnects were about 0.2 mils wide, inspections that had formerly taken several minutes then took several hundred minutes. One contractor said, “This problem is like sending one person to find a contact lens somewhere in a football field with knee-high grass,” and he warned that the most advanced semiconductors were going into industrial and commercial, not military, uses.4
In the early 1980s, Pentagon investigators found that the mil-spec branches of most U.S. chipmakers were not fully testing their chips. While in some cases firms were actually guilty of supplying fraudulent documents, the Pentagon never claimed that the chips were defective. Industry responded by arguing persuasively that the old testing procedures had become pointless. By then the semiconductor industry had adopted stringent quality control measures during production. One executive contended, “It should not go unnoticed that a vendor’s process-quality guarantee of 150 parts per million is substantially better than any known military 266IC’s when they arrive at the prime contractor’s [weapons-maker] factory.” In fact, the military testing process could actually damage perfectly good chips.1
The military’s insistence upon the mil-spec approach was widening the gulf between the commercial and military sectors. But the Pentagon may have had no other choice. It could not match the rewards of the commercial sector to hire or retain engineers and managers qualified to make flexible judgments about the quality of chips. It needed standards that any technically trained bureaucrat could implement.
By the end of the 1970s, the military’s demand for semiconductors had slipped to less than 10 percent of the total U.S. market. Chips designed to mil-spec standards typically lagged two years behind comparable commercial designs. And the brightest young engineers—including those who supported U.S. military policy—were choosing the commercial sector over military employment or contracting.
VHSIC
In late 1980, the Pentagon launched the very high speed integrated circuits (VHSIC) program, which has since spent several hundred million dollars. VHSIC has a clear purpose: the restoration of the military to the forefront of semiconductor technology.
Some proponents portray the VHSIC program as a response to government-industry collaboration in Japan. The business press has featured many articles promising commercial spin-offs. However, the VHSIC program has never emphasized the development of new commercial technologies. The commercial impact has been minuscule.
The VHSIC program followed a two-pronged strategy. First, it sponsored the development of advanced design and production techniques to produce dense, high-speed chips 267for specific military applications. Second, it formed contractor teams, linking commercial chip firms with weapons-makers, to speed the insertion of new chip technologies into critical weapons systems. This approach appears to have been designed to serve Pentagon goals. Gordon Bell, assistant director of the National Science Foundation, says, “The VHSIC program simply sucked the technology out of the commercial industry and brought the military into the mid-80’s.”
Three characteristics of the VHSIC program kept it from strengthening the commercial sector. First, from the start the Pentagon clamped controls on the flow of goods and the dissemination of technology. Second, the chips sought by the Pentagon were not well suited for mass production. Third, the Pentagon chose to work with large semiconductor and aerospace companies.
Soon after the VHSIC program was announced, officials made clear that its new technologies would be subject to controls. The VHSIC director proposed restrictions on the publication of unclassified university research sponsored by the DoD, prompting an open letter of opposition from the presidents of five of America’s top research universities. While academia is still fighting the Pentagon over publication rights, the VHSIC program’s industrial contractors never questioned this approach to security. Any chip developed under the VHSIC program may only be sold to military contractors eligible to receive arms under the International Traffic in Arms Regulations. Many VHSIC technologies are actually classified secret by the Department of Defense.
Thus, the Pentagon has prevented direct technology spinoffs, and the threat of restrictions has diminished the exchange of ideas between commercial and VHSIC labs, even at the same companies. At least one chip maker is said to maintain detailed files to show that its commercial state-of-the-art chip development is distinct from its VHSIC work, 268in case the Pentagon moves to classify more VHSIC information.
Even where VHSIC design technology can be made available outside the weapons industry, it is of limited value. VHSIC has supported the development of advanced chips designed for particular military missions, even integrating chip design with the development of military systems. Most VHSIC contractor teams have chosen to create custom chips, such as radiation-hardened signal processors, that have little relevance for commercial markets. When designing these custom chips VHSIC contractors have chosen design tools that do not maximize the utilization of the chip’s surface. This makes sense in the military’s world of short production runs. However, commercial firms seek high-density designs that will pay off at higher volumes.
By working only with large firms, the VHSIC program missed a chance to work with many companies operating at the leading edge of semiconductor technology. Ever since AT&T began to circulate transistor know-how, a major share of significant innovations in semiconductor technology has come from small, new companies. Of course, given the emphasis of VHSIC on noncommercial technical problems, the program’s hands-off approach might have actually made it easier for small companies to come up with commercially viable technologies.
Faster Than Silicon: Gallium Arsenide
While it is unlikely that the VHSIC program will transfer new silicon technologies to the commercial sector, it is possible that the military will fund a new semiconductor technology until it becomes commercially viable. That is, the Pentagon helped launch the silicon transistor and integrated circuit industries; why can’t it repeat its success?
A prime candidate for repeating earlier success is the use 269of gallium arsenide (GaAs) as a substitute substrate for silicon. Electrons move four to five times as fast in GaAs as they do in silicon, and the processors operate at a wider range of temperatures. Market research firm Frost & Sullivan suggests that “the speed that electrons can move within gallium arsenide . . . will be yielding fewer busy signals on the telephone, less error in workplace data transmission, much faster computers, even better TV and superior electronic weapons for the military.” Frost & Sullivan says that the military held nearly half the $464 million U.S. gallium arsenide market in 1985.
Silicon is still the medium of choice for most applications, since gallium arsenide is made of much rarer elements and is more difficult to work with. GaAs wafers cost five to ten times as much as silicon and are only available in three-inch diameters. In digital logic chips, GaAs circuit densities are at least two generations behind silicon. Supported by the VHSIC program, Honeywell Inc. and Rockwell International have recently announced breakthroughs such as a 4-kilobit static RAM chip. While this is an accomplishment for GaAs, it is orders of magnitude behind silicon RAM circuits.
Conceivably, military research and procurement dollars could help researchers and producers overcome the technological obstacles encountered when designing and building GaAs chips. Numerous Pentagon programs, including the VHSIC-like microwave and millimeter-wave monolithic integrated circuit (MIMIC) program, support GaAs work. Since equipment and techniques can be adapted from silicon production, catching up to silicon-based technology is not an impossible goal.
There are, however, differences between gallium arsenide development and the early years of silicon. First, in the 1950s and 1960s, the military wanted the same things out of solid-state components that the commercial sector needed: high-speed circuitry, miniaturization, low cost, and the efficient 270use of power. The military is supporting gallium arsenide work because GaAs is much more resistant to interference and damage from radiation. GaAs circuits are more likely than silicon to work in a radioactive environment. The Pentagon is seeking the development of high-performance components for space warfare, and cost reduction is a minor goal. It should not be surprising, therefore, that Frost & Sullivan sees the military’s share of the GaAs market increasing over time, not declining, as it did with silicon components.
Even if the Pentagon were to propel GaAs technology into the commercial sector, the benefits to the economy could be nullified by environmental risks. One of the elements in GaAs is arsenic. In high concentrations, arsenic and many of its compounds are lethal. At lower levels, it is still carcinogenic. Arsine, a highly lethal gaseous compound of arsenic and hydrogen, is already used in the manufacture of silicon semiconductors. But it is used in small amounts. In GaAs circuits, arsenic is present in much greater volumes. The formation of gallium arsenide generates toxic vapors that must be contained. When ingots are sliced into wafers, dust is created. Waste disposal is a problem, since waste drums of the compound can generate electrical currents (like a battery) causing corrosion and eventual leakage. Even the GaAs circuits themselves can be the source of hazardous materials when they are discarded. Two GaAs workers have reportedly been killed by such hazards thus far, and cumulative occupational exposure is likely to create a long list of injured workers. The potential for an environmental disaster is serious, with even a Bhopal-type incident from an arsine gas leak possible.
The commercialization of nuclear power may be the appropriate precedent for the GaAs case. In the case of nuclear power, military funding (through the Navy and the Atomic Energy Commission) created an industry that was not fully 271capable of competing commercially, especially when safety rules were belatedly imposed upon its operations. Still, the industry developed the clout to win subsidies, lax environmental regulation, and sales. Only now are power companies realizing that nuclear energy was the wrong choice, and they still remain unwilling to give up their investment in an inappropriate technology. Similarly, in the unlikely case that military funding helps spawn a commercially significant gallium arsenide industry, we may all be the worse for it.
Sematech
In March 1986, a task force set up by the Defense Science Board recommended massive new Pentagon involvement in the semiconductor industry. It proposed that the Defense Department supply $200 million a year in contracts to a Semiconductor Manufacturing Technology Institute (SMTI). The institute would be a consortium of U.S.-owned semiconductor houses, which together would match the military funding. The SMTI proposal dovetails with, but differs in detail from, the Semiconductor Industry Association’s proposed Sematech consortium.
On its face, the proposed SMTI differs from VHSIC, MIMIC, and other earlier military chips programs because it is specifically designed to support high-volume production by the commercial sector. It would focus on manufacturing technology, not design. Task force participants, led by Martin-Marietta’s Norman Augustine, reasoned that a healthy American semiconductor industry was vital to the generation of future military technologies. SMTI would produce some military chips, but it would also be required to sell products on the open market to demonstrate its ability to produce at low cost.
While cooperative chip-building technology programs may make some sense, in the rush to “beat the Japanese” policymakers 272may overlook their shortcomings. First, there is no guarantee that jobs generated by such massive subsidies would go to Americans. The U.S.-owned semiconductor industry employs about as many people outside the United States as it does domestically, and unless other conditions are built into SMTI funding, this practice will continue.
Second, the consortium may freeze out smaller semiconductor firms. Historically, small firms have contributed a disproportionate share of new chip design technology. However, the proposed consortium is likely to be a combine of the country’s largest computer and chip firms. Since Congress must waive antitrust rules to permit the formation of such a consortium, provisions could be inserted into the legislation to require that small companies be allowed access to the technology or even to the consortium’s production facilities.
Third, the manufacturers of semiconductor equipment, which have always been important sources of new production technology, may be excluded. They appear supportive of Sematech and SMTI, but their interests are not the same as the chip houses. They too would like to see the subsidized development of new manufacturing technologies, but their business depends upon their ability to sell to the Japanese, Koreans, and Europeans. It is not clear how semiconductor equipment firms will be able to participate.
Fourth, and probably most important, Defense Department involvement is likely to encumber the work of any semiconductor manufacturing consortium. Despite suggestions in the Augustine report, it is likely that Pentagon subsidies will be linked once more to the specific needs of the armed services. The military is unlikely to sponsor any research that will leave it further behind the commercial sector. In addition, it is hard to imagine a Pentagon investment in critical technologies without stringent controls over the sharing of information.
273The recent historical record makes it clear that if the goal of government policy is to strengthen the domestic semiconductor industry, then aid should be channeled through an agency that is not mission-oriented, such as the National Science Foundation. Many of the supporters of Sematech would prefer this approach, but they recognize that the Reagan administration is much more likely to increase Pentagon funding than support for the NSF. As for the Pentagon, it is likely to support a massive undertaking such as SMTI or Sematech only if there are direct benefits. That is, the DoD will fund a manufacturing research consortium only if the program helps the military regain the control over the semiconductor industry that was lost nearly two decades ago.
The Many Faces of Computer Science
The programmable electronic digital computer was an outgrowth of World War II, as detailed in Chapter 2. Teams in Britain, Germany, and the United States worked on specialized machines for calculation and code breaking, but the machine that launched the American data processing industry was ENIAC (the electronic numerical integrator and computer). ENIAC, a vacuum-tube based monster built at the University of Pennsylvania, was sponsored by the Army Ordnance Corps to calculate artillery tables. ENIAC was dedicated in February, 1946, too late to contribute to the war effort, but timed perfectly to stimulate the development of more practical computing machines.
As it had been with semiconductors, the federal government was a key source of research dollars and a primary market for the earliest generations of computers. As late as the mid-1960s, about half the sales of IBM (already the world’s dominant supplier) were to the Pentagon.
The most significant chapter in the Defense Department’s 274commitment to computer science began in 1962, with the formation of the Information Processing Techniques Office (IPTO) at the Advanced Research Projects Agency (now known as DARPA). While described in detail elsewhere in this book, certain developments had an impact on commerce and should be mentioned here.
DARPA, which had been formed in 1958 in response to the Soviet launching of Sputnik, had a broad mandate to develop new technologies not adequately supported by the more mission-oriented labs of the armed services. IPTO, under the direction of J.C.R. Licklider, funded computer science research primarily at universities and think tanks.
Since then, DARPA has been the preeminent sponsor of American computer science research and graduate training. Other branches of DARPA worked on specific military problems, such as submarine detection, social science studies for counterinsurgency in Southeast Asia, and the monitoring of nuclear tests. Meanwhile, IPTO, under Licklider, supported more basic research, with a mandate to advance the state of the art in computer science.
During the 1960s and early 1970s, DARPA-sponsored researchers came up with technologies that literally changed the world. These included the first computer time-sharing system, a method of telecommunications networking known as packet switching, the bit-mapping method of generating computer graphics, and tools for interactive computing such as the mouse.
Various military agencies had funded efforts to develop time-sharing computers—machines capable of running several tasks simultaneously. Licklider focused that effort in Project MAC (multiaccess computing) at MIT. MAC not only was the breeding ground for some of computer science’s most well-known researchers—including John McCarthy, Marvin Minsky, Seymour Papert, and Joseph Weizenbaum—it also was a first home to a new breed of computer 275hackers, who among other things created the first videogame, called SpaceWar.
Licklider also focused on interactive computer graphics, both at MIT’s Lincoln Laboratory and at the University of Utah. Ivan Sutherland’s doctoral thesis, called Sketchpad, developed on an experimental computer at Lincoln Lab, formed the basis of bit-mapped graphics technology, successfully commercialized much later by Apple Computer’s Macintosh computer. At about the same time, at Stanford Research Institute in California, computer scientist Douglas Engelbart attracted DARPA support with a now legendary article entitled “A Conceptual Framework for the Augmentation of Man’s Intellect.” Engelbart developed his ideas into a system, which he called Augment, that is today widely credited as having fundamentally shaped modern thinking about personal computers. Engelbart was the inventor of the mouse as well as a pioneer in advanced computer display interfaces. Other early DARPA funding led to the development of packet switching, a fundamental technology that is the basis of today’s commercial data communications networks. Potentially even more important in the future, DARPA funded the first research into distributed computing.
Despite DARPA’s leadership in computer science, the military proved ill-equipped to take advantage of progress in computer technology, particularly on major projects. For example, in 1979 the General Accounting Office found that the World Wide Military Command and Control System (WWMCCS), a computer network with responsibilities as imposing as its title, worked unreliably, not only during international crises, but even during day-to-day operations. The GAO found that WWMCCS hardware and software were out of date.
While DARPA was leading the world in computer network structure and interactivity, WWMCCS was relying heavily on batch processing, an approach that created traffic 276jams in computer memories. In fact, the architects of WWMCCS did not even take advantage of technologies that were commercially available at the time the network was designed, ten years earlier. Officials at GAO wrote:
The WWMCCS computer system’s circuitry was designed for batch or sequence processing, and although well suited for this type of processing it does not contain a capability to function economically, efficiently, or effectively in an on-line interactive mode. By the late 1960s several vendors, including Honeywell Information Systems, Inc. (supplier of the WWMCCS processors), had marketed computer systems that were designed to operate in this fashion.3
In 1973, DARPA installed the Illiac IV, at the time the world’s most powerful computer and a landmark in parallel processing, at NASA’s Ames Research Center in Silicon Valley. Just down the street, however, the Pentagon’s sole Satellite Control Facility was using obsolete Control Data 3800’s. When the Air Force brought in newer machines, it handicapped them by design. To avoid developing and testing new software, it actually programmed new Varian 77’s to emulate the old 3800’s. With the growing importance of military satellites, the old computer systems just couldn’t do their job. At least as late as 1986, engineers were scheduling contacts between spacecraft and Air Force ground stations—about 100,000 per year—by making marks on graph paper three feet wide by eighty-four feet long!
In 1980 the Air Force hired IBM to install newer machines, but the work is still going on. Jack Robertson, Washington correspondent for Electronics News, wrote, “The IBM 3081-class computers involved will be outdated by the time they are installed—the fault not of IBM but of the strait-jacketed military procurement bureaucracy.”8
The military acquisition of large computer systems is handicapped in two fundamental ways. First, several years usually elapse between the submission of contractors’ bids 277and the completion of any given system. This may be an adequate way to procure aircraft carriers, but it means that the military must always rely on obsolete data processing equipment.
Second, the military, with the exception of DARPA, tends to quantify technological progress. When Pentagon officials consider the need to develop new computer software, they don’t discuss conceptual breakthroughs. Instead they count lines of code. Air Force reports, “The military is currently developing as much new software as the 100,000,000 lines of code that it now has in use.” The demand for software is growing rapidly: “For example, the core avionics of today’s fighter aircraft embody about 200,000 machine instructions. The next generation of such aircraft—defined generically—will need 1,000,000 such instructions in their core avionics. Those future aircraft will also require about 250,000 machine instructions just for the signal processors that will be mated with their sensors.”1 Larger systems require even more. WWMCCS, for instance, is expected to require 20 million lines of code.
Today the Pentagon is attempting to close the gap between the state of the art in military-sponsored computer science research and the reality of military computer systems. Three major initiatives discussed in more detail in other chapters mark that effort: the Strategic Computing Program, the Strategic Defense Initiative’s battle management effort, and the Software Engineering Institute.
Strategic Computing
DARPA and the armed services launched the Strategic Computing Program in 1983. Funded at over $100 million per year, SCP is DARPA’s largest project. Like the VHSIC program, the project was designed not only to advance the state of the art, but to speed the insertion of new technologies 278into actual weapons systems. Initially, the Pentagon targeted three applications: a pilot’s associate for the Air Force; a battle management system, or admiral’s adviser, for the Navy; and an autonomous ground vehicle for the Army.
DARPA’s move from basic to applied research raises three basic issues. First, by rushing to apply new technologies before they are ready, it risks leading the Pentagon up innumerable blind alleys. Second, it raises serious ethical questions for computer scientists who, while opposing the thrust of U.S. foreign and military policy, have felt comfortable in the past accepting DARPA’s funding for basic research. Third, and most important for this discussion, it could slow or cut the flow of commercially significant ideas from the DARPA fountain.
Depending upon the audience, DARPA highlights or downplays the potential for commercial spin-offs from Strategic Computing. Dr. Robert Cooper, head of DARPA, told Congress in 1984, “we expect that there will be considerable commercial spin-off, particularly to the computer industry in this program, and that it will strengthen the commercial base in computing in the 1990’s.”7 On the other hand, DARPA officials told Defense Electronics, also in 1984, that the “Focus will be on building high-speed super-intelligent machines for defense, so it is not an issue of U.S. vs. Japan in this particular program, but rather how to meet the defense needs in the future.”2
While it is likely that SCP will advance some technologies useful to the commercial sector, the money would better serve the needs of U.S. industry if it were channeled through IPTO programs as they were administered in the past, or through a nondefense agency such as the National Science Foundation.
SCP is diverting scarce talent to solve problems particular to the armed forces. Instead of working on ways to put 279together machines that will be built by the millions, computer scientists are developing hardware, techniques, and software for battle systems that will be made in lots of tens or hundreds. Furthermore, resources that could be supporting creative thought are being used in expensive demonstration programs such as obstacle courses, so that military contractors can show how stupid “smart” ground vehicles can be.
The NSF’s Gordon Bell is concerned that SCP is moving research into the arms of corporate contractors. He says, “Contractor research only has about one tenth the efficiency of university work. There are two major advantages to university research: They have sacrilegious undergraduates and they have a high turnover rate.” Researchers working for weapons-makers learn that the practices of the military-industrial complex, such as a lack of competition and cost-plus contracting, mean there is little incentive to hold down costs.
Finally, the Pentagon’s practice of classifying and otherwise controlling the publication of military-sponsored research results is on the rise. There is no guarantee that SCP-generated know-how will be allowed into the hands of commercially oriented companies or divisions. Even proprietary restrictions are likely to increase with SCP.
Strategic Defense
As explained elsewhere in this book, major advances in computer technology are necessary if the Strategic Defense Initiative is to remain politically viable. Most of the SDI program’s battle management budget, worth hundreds of millions of dollars each year, goes for computer-related work. Any program that large will undoubtedly generate some ideas of commercial value—that is, if the ideas are allowed to circulate outside the military community.
280However, SDI computer research is even less likely than Strategic Computing to feed innovative concepts to the commercial computer industry, for many of the same reasons. The problems of SDI are even more specialized. These include: devising ways to test software that can never be tested in the real world; constructing software, expected to contain over ten million lines of code, to govern a huge, global real-time network; and designing systems to operate during a nuclear exchange.
Software Engineering Institute
Recognizing that the military and its contractors do not have enough qualified personnel to create and maintain millions of lines of computer code in military applications, the Defense Department created the Software Engineering Institute (SEI) at Carnegie-Mellon University in 1984. The institute expects to spend about $100 million during its first five years.
SEI is not trying to develop new ways for computers to function. It is working on tools to generate software more efficiently for existing military applications. If the results of its efforts are made public, SEI can make productivity tools available to the commercial software industry. However, in the commercial computer industry and particularly in microcomputer software development, individual creativity and group synergy have been the essential elements of success. Though America’s strong educational system and government-sponsored research have played key roles in the development of the computer industry, the freedom of hackers to experiment with new technologies is the secret weapon that has kept the U.S. computer industry far ahead of its rivals in Japan, Europe, and the USSR. By promoting automated tools and overlooking the potential of individualism in student programmers, the military could deny the commercial industry its lifeblood.
281A Battle for Hearts and Minds
The Pentagon wants to get first crack at new developments in semiconductor and computer technology. No matter who is president, the DoD is likely to have a massive war chest for the establishment of research, development, and manufacturing programs. If the military attracts the best and brightest engineers, programmers, and computer scientists and puts them to work on focused military problems behind closed doors, then those segments of American high-technology industry that compete in the marketplace will be severely hurt.
Yet military domination is by no means a certainty. Many creative individuals refuse to work on weapons technology, for a variety of reasons. Some have moral objections, opposing the nuclear arms race or America’s interventionist foreign policy. Others, as practically oriented engineers, want to be able to see their innovations put to use. Some don’t like the waste characteristic of most military programs. Some find working on massive projects too impersonal. Others prefer the stock options, informality, and independence that are more prevalent outside the military sector.
The inability of the Pentagon to win the hearts and minds of the country’s scientific elite means there is hope that America’s commercial high-technology industry will survive even the clumsiest efforts of the Pentagon to control advanced technology. And potentially, by supporting more basic research, buying innovative products, and reducing secrecy, the Pentagon could play a more positive economic role. However, one point remains clear. There is no economic reason to support the current stream of military high-technology projects.