Computation has always been an intrinsic part of military science. Among the early requirements were the need for ballistic tables, navigational information and astronomical techniques. Today individuals have immense computational power readily available, and computers have an impact on all aspects of life. In examining the military prospects for future developments in computer science we will chart the exponential rise in capabilities since the application of the transistor, and identify the projects, both military and civil, which have powered the developments.

While the tasks demanded of computational devices range from displaying the time on a clock to predicting world wide weather patterns, there are a number of common features to all systems. A computer takes in information and instructions, stores this data, processes the information according to a set of instructions and then feeds out the processed data. The system may be as apparently simple as the pocket calculator. Here the information is fed in through the keyboard as numerical data and mathematical operations. The calculator stores the numbers and processes them according to the requirements of the mathematical function. It then feeds out the numerical result to the display. Or it may be a jet engine fuel control system. Here the information will come from many temperature, pressure and control sensors. It will be processed according to preplanned instructions and the output will be in the form of electro-mechanical control of the fuel system.

The route to the small, powerful, cheap computing device starts from the military pressures of the Second World War. In 1944, J. Presper Eckert Jr and John W. Mauchly were developing a machine to compute artillery tables for the US Army. This machine, the Electronic Numerical Integrator and Computer (ENIAC) was completed in 1945. The machine could complete 5000 additions per second, and could do in a day calculations which would take a man 6 years. The major drawback was that the sequence of calculations had to be wired in, and altering the sequence was therefore a complicated undertaking. The computational power would have been of great help to another military project. The scientists working on the atomic bomb project at Los Alamos were having to complete long and vital calculations with mechanical calculators and paper and pencil. One of them, John von Neumann saw the possibilities of developing the ENIAC type machine so that the sequence of operations could be changed through the use of stored instructions instead of rewiring. His ideas were not ready in time to help the scientists on the Manhattan Project, but it was the catalyst of two military needs which brought the ideas together which led to the programmable electronic computer. In a paper in 1945 (1), von Neumann described the building blocks for the modern computer today.

Following the war, research continued in the universities and the commercial possibilities added impetus to the development work. The first large commercially available computer systems appeared in 1951 in the form of the Ferranti and the Univac models. Since that time, power and reliability have increased by many orders of magnitude, while cost and size have decreased radically. This has led to a proliferation in applications of computing devices greater even than the spread of machine tools in the last century. Today computers are found everywhere. The average European home will have them in clocks, televisions, heating controls, cookers, washing machines, toys, cars, typewriters and video recorders, as well as the more obvious calculators and personal computers. How has a machine like ENIAC(2), with some 19000 valves, 1500 relays, hundreds of thousands of resistors, capacitors and inductors, housed in a large air-conditioned room, and consuming 200 kilowatts of electricity been reduced to the size of a pack of cards, able to run off a torch battery, and all for the price of a book?

Looking first at the most obvious reason, there have been two developments in electronics which have led to significant miniaturisation. The early electronic computers depended on thermionic valves, which were inherently bulky, power intensive and unreliable. The invention in 1948 of the transistor, a solid state switching and amplifying device, offered a much smaller component which needed no power to run heated filaments. Difficulty in developing reliable mass production techniques, and perhaps also a lack of foresight on either military or civil users, meant that it was 1963 before the first transistor desk calculator was in production. By this time the pressures of space technology and the missile developments were generating much more interest in miniaturisation of electronic components in general and computers in particular. Individual component size reduced, and the use of printed circuit boards reduced circuit sizes somewhat. From there the idea of etching the transistor, or more than one, with its associated components, on to single pieces of suitably treated semiconductor was developed. The integrated circuit became the successor to the transistor. The driving force of the missile and satellite programmes increased the packing density of components. Since 1959 the number of elements contained in each integrated circuit has doubled annually. Once the power, reliability and widespread application of these devices became apparent, commercial development led to mass production techniques and reduced costs. The production techniques have been further improved by computer aided design: the power of the modern computer is harnessed to make even more powerful computers.

Can these trends of greater component density at less cost continue for ever? There are a number of physical limitations. Individual component size is now down to fractions of a micron. That is a size of the order of the wavelength of light. It is not therefore possible to use light to 'draw' the components. X-rays or electron beams provide the shorter wavelengths necessary to focus and draw such fine detail. The packing density will therefore be limited by the technology available to manufacture the integrated circuits. As the system moves into the high energy physics regime costs may begin to rise. Another concern is that as size of components decrease to atomic like dimensions, quantum effects become significant, and the predictability of operations may decrease. In one analysis of the fundamental physical limits of computation (3) the authors assert that the uncertainty principle need not theoretically constrain computer capability. Other constraints arrive as numbers of components increase. The time that information takes to travel is limited by the speed of light, and processing time is also a limiting factor. We shall examine potentially useful techniques for increasing power per unit volume later in this chapter.

The advances in integrated circuits have allowed the production of the computer on a chip. All the components necessary to accept input of data and instructions, a store for the information, a processing unit and an output of the processed information are etched into a single piece of silicon. The storage of data has seen a similarly spectacular decrease in size. Early computers depended on punched cards or paper tapes for the input of data, and bulky delay lines or magnetic drums for temporary storage of data within the computer. These gave way in the 60s to magnetic tape input and output of data, with the computers using magnetic core memory for its internal operations. The drawbacks of the sequential nature of data stored on tape have since been overcome by the use of magnetic disc systems which can be accessed at any point. Internal memory has been taken over by integrated circuits giving large arrays of switches. We have moved from the punched card which could hold 80 bytes of information to the hard disc which can hold several 1000 million bytes. For internal memory the delay line of the early computers was some 5 foot long and could hold around 1000 bits with an access time measured in milliseconds. Today the solid state Random Access Memory can hold 64 million bits on a chip measuring less than an inch square, with an access time of a fraction of a microsecond.

The most powerful computers are designed to minimise the time taken for signals to travel within the machine, and speed of operation is measured in megaflops. A megaflop is one million floating point operations per second. The speed at which the signal pulses travel through the wires connecting the various components of the computer is of the order of 15 cm per nanosecond. Despite the advances in miniaturisation through the development of Ultra Large Scale Integration (USLI: the manufacture of over 1 million components on a single chip), a powerful computer needs large amounts of memory. While individual integrated circuits can be designed with cycle times of one nanosecond, the physical size of single processor super-computers increases the minimum cycle time for the machine to more than 10 nanoseconds. (4). To make further progress in improving speed of operation, the inter-component distances need to be reduced to a configuration where operations are not constrained by the time it takes for the signals to travel. Not only does this present severe manufacturing difficulties, the problem of heat dissipation becomes critical.

There is however a different approach to the problem. One technique has been the use of many processors connected in such a way that they work simultaneously on a problem with their own dedicated memory. One such working parallel computer using 65,536 simple parallel processors averages about 2,500 megaflops (5). Theoretically such a design system could be used to produce a computer with a thousand million parallel processors. Using current technology it would be as large as a building, cost 20 times as much as the largest commercial computer, but be able to carry out 100 million million instructions per second. Such a system requires quite different approaches to problem solving, so that the operations can be processed in parallel rather than sequentially (6). This may be the limiting area, and the research on software development may become more critical than novelty in the computer hardware. The commercial supercomputer of 1996 uses a limited number of complex parallel processors. With this technique, the Cray J932 using 32 processors achieves 6400 Megaflops. It weighs 1400 lbs, needs 12 square feet of floor space and uses 8 kW of power. (7).

It may be that the semiconductor junction which has held sway since the transistor has run its course, as the thermionic valve did before it. There are those who argue that it will remain the preferred method for some time to come given the advantages of its high gain ratio (7a). However, different switch devices are already available. The Josephson junction depends on quantum-mechanical effects displayed by thin insulating layers at temperatures near absolute zero. A junction can be arranged which is superconducting until a magnetic field is applied when it switches to being resistive. The switching time between the two states can be as low as 6 picoseconds (6x10 secs), and the power consumption is very small. A superconducting computer with an overall cycle time of two nanoseconds could be constructed in a two inch cube.(8). Another switch offering low power consumption, high speed and high component density is the photonic switch. The Self Electro-optic Effect Devices or SEEDs are activated by the arrival of a photon of light, which generates a voltage, which in turn changes the light transmission characteristics of the SEED. Current technology can manufacture a SEED of 2500 alternating layers of gallium arsenide and gallium aluminium arsenide with supporting circuitry all within a thickness of six thousandth of a millimetre.(9). Advances using this type of device are particularly promising for the parallel processing of data, which can enter from the top of the chip and emerge at the bottom, rather than the more conventional sequential processing systems.

While there is still room for development through the utilisation of electronic switching processes, the capacity of the human brain suggests that the most productive long term research will be in the field of biochemical computers. The brain is still poorly understood but appears to be able to cram at least 10 junctions into a box the size of the skull. The processing system is parallel, redundant and self manufacturing. It would be surprising if the chemical and electrical connections of the brain had no lessons to offer to the computer scientist. Research is already well underway into electronic simulations of the brain's neural network (10). Such systems are already ale to carry out collective decision computing.

Computers of whatever power remain useless boxes until they are instructed how to process information. The earliest machines were wired up so that the operations required were carried out in the necessary order. The development of variable instructions as part of the input data allowed the development of modern computers. Originally these instructions had to be laboriously written to tell the computer in detail each step required. Thus if numbers were to be added they would have to be converted into the binary system, stored at specified locations in the computer memory, brought into the central processing unit to be added one by one, and the result sent to another memory location. To ease this task, operations which were used repeatedly were stored for quick access, and the computer could translate a single mnemonic into the necessary sequence of instructions. Different applications called on different routines and so a range of computer languages has developed. The problem of the development has been to make it as easy as possible to give the computer instructions, while at the same time making them both correct and unambiguous. Machine-dependent languages are different for every processor and are totally specific. However they are difficult to trace errors in, and take a long time to write. General purpose higher level languages like BASIC, Pascal, C, FORTRAN, and COBOL were easier to use but could lack rigorous checking systems to ensure that the machine is correctly instructed. What is still required is a programming system with universal application, which is standard between the whole range of computing devices, does not use up the machine's processing capability in translating the instructions, has the ability to detect errors, and allows programs to be easily understood and modified.

The advantages of such work on standardisation led to the adoption of ADA by the US armed forces as their preferred language. However languages currently require specialist training for the giving of instructions to computers. The tendency has been to make the job of the operator as easy as possible. As computing power has increased it has been possible to make available capacity for the computer to guide the operator through the instruction process. For the future, the realms of artificial intelligence seem to offer the greatest potential for new programming arrangements exploiting the capabilities of the greater computational power.

The definition of artificial intelligence remains a somewhat emotive issue. As the capabilities of computers have expanded the definition of intelligence has been pushed back further and further, until it appears that it becomes defined as that process of reasoning which is beyond the capability of current machines. Current research focuses on the nature of consciousness. The first computers were configured for a particular process. The development of programming languages allowed machines to be used for many applications through different programs. More recently the operator has been able to instruct the machine more generally about problems, and it has generated the program to cover the task. Programs which allow a machine to learn by experience and modify the program accordingly are already in current usage. In the same way computers can operate on incomplete data and make a best guess. The increases in computer power and speed allow these abilities to be improved, and hence the machine becomes more intelligent. The design of the program is also a vital part of this process. Brute force analysis of problems limits the size of the problem which can be solved. If chess playing computers attempted to examine every possible move and counter move they would have to consider of the order of 10 possibilities. Successful chess players, both human and computer, reduce the number of possibilities they examine by looking at 'promising' options. The rules which guide the computer to select a manageable number of options determine the intelligence of its game. If those rules can be modified by experience by the computer then it can improve its performance further. Such systems already exist, and the developments of techniques to improve intelligence is a major area for further research. (11).

One relatively recent computer development, the Internet, is an interesting example of military research, which was subsequently extended and improved by civil users, and then taken back into military use for different purposes. In the 1960's, the United States Department of Defence funded a range of research activities through the Advanced Research Projects Agency (ARPA). To connect the computers at a number of research sites, and reduce costs, ARPAnet was developed. In 1969, the importance of data connectivity in war was becoming apparent, and the ARPAnet was used as the research vehicle for developing resilience in computer networks. The aim was to provide systems which could route around those sites lost in a nuclear exchange. The system which was developed was to send data in packets which were automatically routed around the shortest available route between sending and receiving site.

University research sites were also connected to this network where they had military related research projects. The interconnections grew as the academic and military users realised the usefulness of being able to link around the world through commercial telecommunication links. By the early 1980's the system had grown to such an extent that the US military decided that there were too many other users connected, and they divided the network into two parts: a dedicated military network and the old ARPAnet for research sites. The two systems were however linked, and other smaller networks began joining through gateways. As the advantages became more apparent to purely civil computer users, new networks were developed. Those which used the Internet Protocol (IP) that had been developed for military use flourished because of the ease of connection. In 1986, five supercomputers in the United States were linked in the National Science Foundation Network (NSFNET). They started by building on the ARPANet, but the following year they broke away and gave the development to the commercial and purely academic world. The NSFNET grew as it offered access to any institution, and in March 1990, ARPANet closed down. Schools began to connect to the network and this was followed by individuals and businesses. (12).

Public awareness of the Internet grew enormously through 1995 as modem speeds increased and costs decreased. The difficulties of navigating this network of millions of computers were eased by the rapid development of easy to use software using hypertext links to move around the World Wide Web. The nuclear war data communication system had become a worldwide information system for every individual. Its resilience has had other profound effects. Governments can no longer easily control information. A dissident can write in one country and be read anywhere in the world. Cutting off nodes in the system has no long term effect: the data packets will re-route as necessary.

The growth of the Internet is a sobering case study for the futurologist. The technology, sponsored by military funds, has existed for 30 years, it was well known in scientific research circles, and it was in daily use. Yet there was no expectation of the exponential growth in the last 5 years. This growth stemmed from the combined effect of mass market personal computers, fast cheap modems, and easy to use software. We are seeing new global information relationships established, which coupled with global economic activity, will have very significant implications for the role of the state in the future. The ability to exchange data with any computer system in the world has profound security implications. Information warfare will require new defensive and offensive techniques.

Already the computer is an essential component of every aspect of war. From the digital clock in the terrorist bomb to the supercomputer simulating the nuclear explosion in a new warhead, modern warfare depends on computer devices. For the gathering of intelligence, computers can assess information, compare multiple sources, analyse it and present it in time for it to be of use. Computer guidance of weapon systems allows navigation systems to work to accuracy measured in centimetres over ranges measured in thousands of kilometres. No longer is accuracy a function of range. In communication systems, computers have brought security through encryption. We will discuss the question of codes more deeply in Chapter 14. In military aircraft computer control allows designs that are inherently unstable and thus offer great agility in combat. Artillery computers can control army firepower. At sea, the ship's captain can fight from deep within his vessel using computer generated information from all his sensors. Under the sea, the strategic submarine depends on computers to run the nuclear power system, and each missile must be linked into the navigational computer. When the missile is fired the on-board computer will control its flight path to strike its target. In the cruise missile, the computer is able to compare the ground over which the missile is flying with its expected contours, and update the navigational information accordingly. The list could go on endlessly. There is no doubt that computers are important to military technology. The question for the future is what areas of research in this field will have the greatest potential to improve military capability, what vulnerabilities result from the universal application of computers, and how can such vulnerabilities be exploited?

The advent of more powerful systems will be particularly important in the acquisition, analysis and dissemination of timely information. Today the cycle time for reconnaissance, analysis, mission tasking and execution is measured in hours, whether one considers space based sensor systems, aircraft or even foot patrols. The ability to data link various sensor systems through a powerful computer, and retransmit the analysed information to the firepower system in a timely way is one of the most urgent requirements. Such technology, while complex, is designed to ease the decision making problem of the soldier. If he has a hand held air defence missile, he does not wish to know the speed, temperature, altitude, radar shape, exhaust characteristics, or heading of the aircraft in his sight. He does not need to know the larger airspace management picture, and how his target fits into it. The computer can look at all the evidence from many sources and determine the probability of the target being an enemy aircraft. For the soldier in the field, his display need be no more than a green or red light to indicate friend or foe. Computer power and artificial intelligence have much to offer in this area.

Weapons have become more accurate and effective as sensor systems have become small enough to include in the weapon. As the computer power that can be included in a missile increases so the weapon needs to depend less on the human operator. It is sometimes suggested that this offers the prospect of the automated battlefield. At its most dramatic, this automation was assumed in the nuclear deterrence posture of launch on warning. Concerns over the possibility of losing retaliatory capability to a pre-emptive strategic attack led to the suggestion that missiles could be launched when a massive attack was detected, but before the enemy missiles arrived at their targets. The short time of flight of ballistic missiles mean that such a system would require computer analysis of the sensor data indicating a massive enemy launch of offensive missiles. If the analysis confirmed such an attack then the computer would activate the retaliatory strike. Such a Doomsday Machine was always unlikely to find favour whatever the improvements in sensor technology, computer power and machine reliability. But is it possible than an analogous automatic system could be used on the conventional battlefield? One can conceive of an intelligent reconnaissance drone searching an area for enemy tanks, activating a surface to surface missile with multiple terminally guided munitions, and hence destroying the tanks. The drone would carry-out the post attack analysis and decide whether further attack was necessary. The possibilities of such systems would depend crucially on the computer's ability to recognise decoy schemes and counter them. The predictability of the system which lacks human intervention is currently a great weakness. Nevertheless, this type of warfare may become possible, and research will certainly be important.

Moving from the use of computers for the tactical battle to their application at the strategic and policy level, wargaming on computers is already an important tool for both planning and training. If the most powerful supercomputers can model the workings of the Earth's atmosphere, will they not also be able to predict the right strategy and tactics to win wars? Here the prospects seem less certain. The difficulties of modelling human qualities, political will, public opinion and the many subjective factors which affect the outcome of a conflict, make the prospects for computer prediction rather less likely than accurate economic forecasting. As a training aid in war, from the individual simulator for a soldier operating an anti-tank missile, through the air-to-air combat simulator, to the full command and control war simulator, increasing computing power is important.

The very usefulness of the computer brings with it the penalty of dependence. The fineness of the component integrated circuits make them sensitive to their environment. In particular excess voltages can destroy the circuitry efficiently. Of particular concern is the electromagnetic pulse (EMP) which accompanies any nuclear explosion. During the Cold War, military electronics were routinely, but at considerable expense, protected against possible EMP levels. Administrative systems and civilian computers were never afforded such protection. Yet the destruction of such systems would have a crucial effect on how well a modern nation could survive. Both the military machine and the civilian community have moved towards a total dependence on complex computer systems. This suggests that a productive line of research may be in anti-computer weapons and their counter defences. Attacks on computer systems can be against the hardware, the software or the data. All are vulnerable.

Hardware attacks can use old or new weapon systems. The terrorist bomb placed in the commercial centre of a city can have severe economic effects if computer systems are not protected, and data backed up and removed to remote locations. At the high technology end of the spectrum, it may be that directed energy weapons will provide a method of destroying integrated circuits unless critical system processors are appropriately protected.

The public is already aware of the vulnerability of systems to attack through software. Every computer user has experienced the unexplained system crash caused by errors in software. These 'bugs' are inevitable as software becomes more complex. For critical weapon and safety systems, assurance of low risk of errors comes at a high price. Research into techniques for software testing can pay high dividends. Software can also be attacked. Computer viruses are instructions for computers to carry out some task which the user had not intended. They come in many forms and may harmlessly display a message saying "Merry Christmas" each year, or alternatively may delete all the data and program files within a networked system. To sabotage a computer system through a virus requires the unfriendly sequence of code to be introduced into the system. This can be done when loading a programme, when entering data, when exchanging data over a network (either internal or external), or when manufacturing the components of the system. The attacker may be a disgruntled employee working on site, or an unfriendly state on the other side of the globe attacking through the internet. Protective firewalls to prevent unauthorised access are a first defence. Accepting the drawbacks of lack of connectivity is also a defensive measure for key systems. Supervision of all computer related activity is also vital. Nevertheless, there remain real vulnerabilities. The development of critical weapon systems requires particular attention. Specialist chips will come from many sources, and an integrated circuit containing 10 million transistors could hide a piece of code which might render the system vulnerable.

The final question to be considered is the division of effort in computer research. The military needs have been important in the history of computer development so far. Commercial exploitation has followed and reduced unit costs. Pure research in mathematics, physics, biochemistry and computer science is likely to produce useful progress with applications to the military. There will however remain the specialised needs of the military fields, which will require specific defence research and development. The area to give the greatest return will be in exploiting an enemy's reliance on computers, and countering our own vulnerabilities.

NOTES ON CHAPTER 6

1. The Penguin Computing Book by S.Curran and R Curnow, Penguin, London 1983 p 83.

2. 'Computers' by S.M.Ulam in Scientific American, September 1964, pp203-6

3. 'The Fundamental Physical Limitations of Computation' by C H Bennett & R Landauer in Scientific American July 1985 p45.

4. 'Supercomputers' by R.D.Levine in Scientific American January 1982 pp 112-124.

5. 'The Connection Machine' by W.D.Hillis in Scientific American June 1987, pp86-93.

6. For a discussion of the approaches to parallel processing see 'Advanced Computer Architectures' by G.C.Fox and P.C.Messina in Scientific American October 1987.

7. 'The CRAY J90 Series' information sheet from Cray World Wide Web page May 1996.

7a. 'The Future of the Transistor' by R.W.Keyes in The Computer in the 21st Century. 1995 published by Scientific American. pp 90-95

8. 'The Superconducting Computer' by J Matisoo in Scientific American May 1980 pp38-53

9. 'Tripping the Light Fantastic' in Scientific American August 1986 pp57B-58.

10. 'Collective Computation in Neuronlike Circuits' by D.W.Tank & J.J.Hopfield, in Scientific American December 1987, pp62-70.

11. 'Computer Software for Intelligent Systems' by D.B.Lenat in Scientific American September 1984 pp152-160.

12. A description of the development of the internet appears in the opening section of 'The Internet Starter Kit' by A.C.Engst - Hayden Books 1993.