Outfitting the Mesh means a shift from few complex items to many simple items knit together by software. For systems acquisition this means building from dual-use parts, fostering open systems, and defining a new role for software.
It is no big secret that today's acquisition system is under stress both from its own dysfunctional internal dynamics and the difficulty of accommodating the post Cold War drawdown. Problems range from very extended cycle times, continuously escalating costs, and excessive overhead, to a technology base that lags commercial developments and has not made an easy transition to commercial conversion. Important questions are being raised on the relative priority of prototyping versus equipping the force, the control of proliferation, and how the industrial base drawdown should be managed in case the United States might need it again.
These discussions tend to overlook how changes in the weapons of warfare may, themselves, affect how the acquisition system ought to be run. Not surprisingly, the optimal method of developing and acquiring elements of the Mesh will, for that reason alone, differ radically from optimal methods of developing and acquiring industrial-age weaponry.
A typical defense system starts life as an operational requirement. This requirement is converted into a basic system design which drives development programs. The design, in turn, is broken down into sub-systems and components. As presently constituted, defense acquisition is predominantly demand-driven. The alternative model, which looks at what is out there and develops innovative ways of using it in defense is far less appreciated. More generally, system designs are only modestly affected by the cost tradeoffs that routinely go into, say, automobile design decisions.
True, this logic is coming under increasing attack on its own merits. Yet, a shift from complex platforms to networks of sensors, emitters, and microprojectiles will (or at least ought to) accelerate the trend to greater cost sensitivity and growing reliance on commercial capabilities in defense acquisition.
One reason that cost competition plays such a small role in major systems issues is that contractors rarely see the sales gains from lowering their own costs. A thirty percent cut, for instance, in the cost of an F-14 carrier fighter is unlikely to result in commensurate increases in the number purchased. Other factors -- prior force planning, the logistics infrastructure, the number of pilots, or carrier deck space -- put an upper limit on the number of F-14s acquired. By contrast, the cheaper are the elements of the Mesh, the more densely they can be dispersed, and thus the more capable the overall system. As the elements of the Mesh become less expensive, networked sensors, for instance, can increasingly substitute for large platforms in the same function. The logic of the Mesh, overall, is heavily driven by economics. Indeed critical architectural and operational issues (e.g., what is the proper density of flooding, jamming, spoofing, decoying, and coverage) have to be decided, in large part, on the basis of which side can afford to throw what kind of resources into a thing-on-thing attrition campaign.
As the number of elements in the Mesh runs into the millions, economics will force systems to be designed around technologies already extant in commercial markets, because the latter alone are large enough to offer economies of scale. Even if military items are not, themselves, commercial items, this linkage requires closer attention to using commercial production facilities and practices. What about the counterargument: if acquisition rates are really so high, might not defense procurement alone generate the economies needed? This fails for two reasons. First, the elements of the Mesh are consumables and, in all likelihood, of relatively quick manufacture. DOD would be better off stocking a few months of them and rely on post-crisis production to make up the rest. A commercial base permits a faster ramp-up in emergencies. Second, the culture of commercial production makes more sense for quantity items than the culture of military production. One reason that defense goods are so expensive is that each is engineered, tested to ensure that every single item has the highest chance of working. The more fussing, the higher the cost; the higher the cost, the greater the urgency of assuring that each works. Elements of the Mesh, however, can do with statistical quality control. Because of how they are likely to be deployed, a certain percentage failure can be assumed. The system, rather than the individual element, is what needs to be configured for reliability -- which it does through planned redundancy.
As it is, information technologies, because they perform similar functions whether in military and commercial employment, are already best suited for non-MILSPEC treatment. The few information technologies hitherto thought unique to military needs (e.g., encryption, spread-spectrum, frequency-hopping) are being adapted to commercial users anyway.
For these reasons, the build-or-buy decision for elements of the Mesh can proceed in a progressive winnow. Some capabilities will be available off straight off commercial shelves, and some will need to be modified for military needs. Others could be helped by defense-led efforts to accelerate the development of dual-use items so that they can appear in defense systems at the right time (and under control of U.S. producers). Still others could be developed in conjunction with new civilian infrastructure projects. The remainder, primarily defense- oriented programs, would represent a reorientation of existing work plus altogether new starts.
Key Technologies: This winnowing can be illustrated by examining the key technologies for the Mesh: electronics, micromechanics, sensors, space, and energy all undergirded with improved manufacturing processes.
Commercial users will drive most electronic technologies (notably digital computation, neural-net hardware architectures, parallel processing, and digital signal processors). DOD could help advance non-silicon optical and electronic materials in its usual ARPA-like ways, and support generic advances in manufacturing such as Sematech. DOD is likely to be more independently active in analog areas such as microwave and extra-high frequency communications, emitters, compact antennas, and counter-EMP hardening.
Major advances in micromechanics are likely to lag similar advances in electronics by one or two decades. Nevertheless, fields with promise include ultra-light exoskeletons and very small legs, some of which could locomote penny-sized sensors and others which could manipulate a windsail (to support airborne sensors aloft for long periods of time). Some micromechanical devices may find use in chemical and pressure sensors.
Among sensors, visual and near-visual (IR/UV) passive collectors are most important. DoD will probably have to be the primary funding agent for improved sensors. Commercial versions could be spun off to uses such as medical instrumentation, optics, and robotic systems. Similar patterns would prevail for acoustic/pressure sensors, seismic sensors, and various chemical sensors. The latter can find use in medical, agricultural, and environmental fields. Fiber optics is showing promise as a the basis of very fine movement detectors.
The successful use of space-based eyes and brains in the mesh is more likely with every drop in the cost of lifting a pound of material into low-earth orbit, and with every method to shrink components found in large spacecraft (power, stabilization, maneuverability, common busses). Ultra- stabilization -- to permit satellites to communication down to specific earth collectors -- would also improve the ability of space assets to do continuous tactical monitoring as would improve hand-off methods as satellite coverage keeps changing. DoD-sponsored improvements could be shared with NASA (and visa versa), but commercial space activities -- unlikely to grow for another decade or two -- should not be counted on for much help.
Three energy technologies which need further development are batteries, photovoltaics, and remote deliveries of energy infusions. Better batteries would extend the life of sensors, particularly those used underwater. Photovoltaic collectors and energy beam delivery would allow continuous energy feeds in remote locations. Battery and photovoltaic technology (which the Japanese lead us in) have strong commercial applications, and both are consistent with a revitalized interest in renewable energy sources -- thus piggyback opportunities. Remote delivery of energy infusions has applications in space, as well.
Manufacturing technology is also critical for its contribution to the affordability that a system composed of many small items needs. Although a specific research agenda must be tailored to specific product lines, two thrusts, miniaturization and more effective cost/quality control, are likely to recur. Any DoD effort to improve manufacturing technologies is best pursued within a consolidated federal thrust and need not be separately programmed.
Civilian Megaprojects: In developing technologies that are needed for the Mesh, DoD may want to look for opportunities to piggyback on top of civilian megaprojects planned for this decade and the next.
One candidate, born Mission to Planet Earth, monitors the earth's environment with low-earth orbiters. The advance of Earth surveillance in general should support better remote multi- spectral sensing, high-bandwidth data dumps from space, sophisticated software especially for distributed access, and orbiters useful for tactical surveillance in general. A successful National Aerospace Plane could slash per-pound costs to orbit.
The High Performance Computing and Communications program seeks thousand-fold improvements in supercomputer speeds, and very high capacity communications lines -- both with Mesh applications. If the program connects schoolrooms to global libraries, it may promote information standards (as a key aspect of systems integration) that could help integrate all those sensors, emitters and nodal processors in the Mesh.
Another program, Intelligent Vehicle/Highway Systems (IVHS), is, like the Mesh, is also concerned with the problem of coordinating millions of objects. IVHS enables highways to talk to cars (to warn them about traffic conditions), cars to talk to highways (to predict traffic flows), and cars to talk to each other (letting them travel more closely packed together without fear of collisions). Sensors and software promoted by IVHS may have defense applications, as would associated developments in non fossil-fuel energy.
Health care is an area whose synergies with defense technologies are underexploited. Medical instrumentation, for instance, is similar to defense systems in their cost, complexity, and the fact that their functionality is a matter of life and death. Cost control requirements may require that more people be monitored outside expensive hospital settings. Doing so would impel the development of remote health sensors that engage in periodic and emergency communication with health networks -- a technology with many resemblances to the Mesh.
Reorienting Defense Research: After absorbing what is available from the commercial world, and gathering what might become available from commercial piggybacks, DOD needs to fill the gap with innovations that it develops itself.
At present, the DOD research establishment -- with their bewildering mix of technologies, wide variety of paths, and diverse clientele -- is largely devoted to countering current capabilities and threats. Too little effort is designed to ward off anticipated threats from emergent technologies. Current parameters emphasize performance maximization (faster, more sensitively, over a wider range of environments), and robustness (versus building redundancy into the system vice components). The latter method produces satellites that cost a billion dollars and carrier battle groups costing ten billion.
To develop the Mesh requires a different direction for DOD's research and development program. The first requirement is some top-down dicta in favor of information technologies that can be deployed as millions of items in a networked information environment. Such dicta would have to be translated into parameters for systems that can be composed of smaller and cheaper components that can be adapted or derived from commercial products. Second, a developmental bias needs to be inserted in favor of methods that divide system functions into decomposable parts, and develop open interfaces so that it can fit into both today's information mesh and tomorrow's. Third, a bias for bench-scaling, building, and testing should be part of the development process. Fourth, systems planning should anticipate that telematics technology will continue to advance roughly fifty times from one end of the ten- year development cycle to the end. Researchers should look for ways to solve problems in software or silicon-embedded microcode rather than with hardware. Thus one should avoid, for instance, developing precision machinery to align multi- spectral photographs if a computer of sufficient power could correlate the various spectral images and determine what is the most probable correlation between various photographs of the same scene.
Rapid technological change virtually dictates open systems design. To understand why requires appreciating how much of today's acquisition cycle time is spent integrating component to subsystem and subsystem into a final system. At every level, each of N subsystems have to be fit with each other requiring the simultaneously solution of the n-square problem. Typically, defense systems are very tightly constructed for maximum efficiency. Altering one component changes a subsystem's performance which in turn changes a system's performance and so on. Thus, minimizing unnecessary changes between specification and integration is important. So how are parts to be specified? If parts requirements assume current technology (which most do), parts will be ten years behind the state-of-the- art when fielded. Calling for components with then-current capabilities may be necessary in some cases for mission accomplishment but overdoing it risks the possibility that such performance is not possible or affordable. If so, the program is delayed; conservatively specified components fall further behind the state of the art. Systems that result from the process tend to contain far too much old technology, but at least with stable technologies the benefits of tight integration cover the costs.
When technologies advance rapidly and unpredictably, however, this model breaks down. The alternative is building systems not from subsystems fit to each other, but to subsystems each fit to a standard interface which is careful specified. This is precisely the approach now being developed for the new generation of object-oriented software modules. As with software, this approach loses efficiency because modules cannot take advantage of known aspects of other modules. However, any lost efficiency is more than made up by greater flexibility. Other modules can get major updates without forcing the whole system to be reintegrated. A new capability, suddenly possible, say, two years prior to fielding, can be inserted with less damage to the original schedule.
Systems integration takes on new meaning for meshes -- at that level, it is almost all software. The combination of common components, open systems, and external systems integration would redefine defense industry. Today's typical prime contractor, ostensibly a frame manufacturer has, over time, become a systems integrator and software writer. The prime imposes hardware-originated contract specifications upon what are, even now, defense-oriented subcontractors. Tomorrow's prime will be almost entirely a software house. Most subcontractors will have to find markets in the commercial world to achieve the low price, and compatible tools and parts that future systems need.
The rise of the Mesh also informs the current debate over what to do with today's shrinking defense giants. The United States, as well as its allies, possesses an excellent defense industrial base whose existence and capacity are imperiled by expected defense cuts. Many defense analysts are looking for ways to keep them alive: more R&D, extra maintenance work, or weapons purchases, foreign military sales, or direct preservation.
The usual argument is that such subsidies are wasteful; the more pertinent argument may be that they are counterproductive. Why? The current force was designed to counter opposing and comparably capable Soviet forces almost weapon-for-weapon to engage in like-on-like combat. For the next decade or two, the odds of a new peer competitor are low. The United States has enough good systems in its inventory to avoid needing many major systems starts. Although new systems would be more survivable and perhaps more efficient, neither fact justifies multibillion dollar development programs. Beyond two decades, a peer competitor and thus feature-for-feature competition may re-emerge. But by then, the value-price ratio for information technology may be a thousand times higher than today's and like-on-like platform combat may be obsolete. Many skills (software aside) needed to build ships, tanks, and planes will not be relevant to building meshes. Worse, the persistence of a large platform-oriented industrial base may be inimical to promoting the revolution in operational concepts needed to change defense paradigms. The current defense structure may retard rather than promote defense reconstitution.
Major improvements in software will be necessary to realize the Mesh: remote systems integration (how to get two different systems to recognize and talk to each other), pattern recognition, adaptive algorithms, data-flow architectures, image compression, and simulation. The algorithms required in the Mesh will need to mix deductive digital components (with their formal logic) and inductive analog components (with their dynamic minimization techniques). Software tools per se are inherently dual-use, and many of the algorithms will find use in the commercial world. Some techniques for remote systems integration may be developed for the infrastructure projects mentioned above. At the level of specific software for particular applications, though, the code will almost always be exclusive to defense applications.
Training and testing will become a greater component of software development. Few complex systems work well the first time out; they will miss some targets and identify other objects as false threats. Neural net components, in particular, need to be tuned by repeated example until they are reliable. The Mesh will have to be tested against wily foes; B-teams could generate decoys and false images as well as real targets with unexpected parameters. Meshes will have to learn, as humans do, how much evidence to collect, and which anomalous readings to pitch out.
Wars hitherto fought in real media will increasingly be fought in abstracted media. Although the same banging and shooting will take place, the cue-search-locate-categorize-target- shoot-assess cycle will require not direct analysis of sensory data, but its abstraction in the realm of oughts and noughts. The offense will be as good as the algorithms that power the cycle. The defense will be judged on how well offensive estimates can be frustrated. Many theaters of conventional warfare -- space, strategic warfare, and naval warfare -- have already been abstracted in that warriors already sit in a simulated environment, one that they no longer directly perceive. This tendency will only become deeper and broader (e.g., pilots will increasingly look at their screens rather than out their windows).
Abstraction implies that what military personnel do -- regardless of service -- will converge. Successful performance will mix an increasing percentage of generic software skills with a decreasing percentage of media-specific ones. True, the algorithms that train sonobuoys to find submarines differ from those that pick out small satellites from those operating in cluttered jungle or urban environments. Experience with a physical medium yields better algorithms. Yet the underlying skills remain the same: writing maintainable code, using computer-aided software engineering, evolving well considered objects, taking advantage of network resources, conducting fuzzy and discrete logical analysis, tuning neural networks, recognizing images faster and more accurately, countering deception, improving the efficiency of learning algorithms, integrating systems, wringing more sensitivity from statistical processes--and so on.
The evolution of the aforementioned Information Corps may yield two distinct types of software skills. One group would develop the system, train the Mesh, and maintain the code to new circumstances. The elite force would specialize in restoring a systems in real-time against unexpected situations or enemy action. Both forces would work together -- original writers often do the best repair -- but the grab-bag of tricks necessary to rewrite and retest code quickly may need to be developed especially for military field uses. Such an elite will get considerable use in wartime; the benefits of fooling a Mesh - - which is only as affective as its treatment of new threats or new spins on old threats -- even for just a day, can be considerable. The elite of the Information Corps would travel globally to install, oversee, reprogram, or trouble-shoot the massive automated systems that tomorrow's armed forces will have become.
During the waning days of the Cold War, both manufacturers and controllers of American export worked themselves into a mutual frenzy trying to differentiate weapons from dual-use items that might have a military application. The advent of the Mesh will make this distinction even less meaningful. Because the Mesh requires dense coverage to work, economics requires adopting and adapting commercial items -- already made by the millions and billions. The same batteries that power consumer cameras would be candidates to power militarized optical sensors. As world markets continue to broaden, what prevents an enemy from building systems from the same materials the United States does?
Nothing, really, but therein lies a dilemma. Our current military, composed of large expensive systems, is based on hardware that has no commercial substitute, and is largely unmatched by anything in world markets. Even after spending a full day at the world's armaments mall, an adversary starts from a weaker position than ours. To scrap our advantage in favor of a system whose technology base is common to all might seem to aid national security as much as the switch from mainframes to microcomputer clones helped IBM. True, America can still afford more equipment than its adversaries. But even during the Cold War, though, America's military philosophy always emphasized qualitative over quantitative measures of superiority.
The use of common parts need not translate into common capabilities. The hardware may be the same, but the secret is in the software -- as the entire computer industry is learning. True the computer-aided systems engineering tools and many of the fundamental computer algorithms will be the same for both civilian and military applications. But many tasks that the Mesh has to perform -- pattern recognition, learning, auto- configuration, counter-deception tactics, and data fusion -- need to be reified in specific code. Such code would generally differ sharply from what similar tasks look like in commercial applications. To the extent that the United States would invest tens of billions of dollars a year in building and refining such code, an adversary would have to spend comparable sums to develop a similar system. Capturing a code-intensive device would not be as revealing as for instance, capturing a panel from a B-2 bomber. Microcode embedded in silicon is extremely difficult to reverse engineer; some chips in use in the intelligence community already self-destruct upon opening. Capturing the original source code would compromise security (if it is well-documented code). However source code could only be stolen from the factory, not -- as with hardware -- in the field.
America's wide lead in software is another advantage to concentrating our military functionality there. This lead is evident everywhere from our dominance of packaged applications to our lead in systems integration for telecommunications and aerospace. Simply put, the American dollar goes further in software than the German mark or the Japanese yen. Thus is multiplied the advantage that our GNP affords our defense. By contrast, American manufacturing skills, dollar-for-dollar are nothing special and, if anything, may be falling farther behind those of our rivals. That said, the ability to manufacture lots of little items -- without having to depend on trade partners reluctant to serve U.S. military interests -- remains important.
Can the United States still lead in software, or are we facing (pace Yourdon) the "decline of the American programmer"? Without discounting the Japanese threat in software, their inroads into American markets have yet to come materialize. Cities as diverse as Budapest and Bangalore have cadres of over-educated but under-employed programmers who write good code for peanuts. Nevertheless, the best foreigners are more often drawn into our corporate orbits than our orbits are rendered asunder by their companies. The lead is there if we choose to maintain it.