The Universal Machine

Transcription

1 The Universal Machine The End of Certainty Technological Progress The slow start of chemistry was overcome with the work of Antoine Lavoisier ( ), who showed that chemicals actually gain weight from the air during combustion, and proposed the existence of oxygen. By 1869, understanding of the basic properties of the atomic elements had progressed far enough to produce Mendeleev s periodic table of elements and soon afterwards Maxwell (1873) proposed the existence of electro-magnetic fields and Stoney (1874) proposed the existence of electrons. It was at this time that Thomas Edison began his work on the development of the light bulb and on the generation and transmission of electrical power. By 1882 Edison s first power station started supplying electricity to New York, beginning the rapid transformation from reliance on gas and steam power to the domestic use of electricity and electrical devices. Edison also pioneered the development of the phonograph, with his first successful recording of the human voice in 1877, and his laboratories produced one of the first motion picture machines in In parallel, the telecommunications industry started to expand with the invention of the telegraph by Samuel Morse in 1837, the telephone by Alexander Bell in 1876 and Marconi s first radio transmissions in The late 1800 s also saw the invention of the first petrol powered vehicles, heralding the dawn of the automotive age. However the vast majority of the population were working hard, often in appalling conditions, for little reward and enjoying only the most basic conveniences of life. In most countries, power was still in the hands of an aristocratic elite whose perspectives on war and international politics had changed little since the time of Napoleon. Into this world, science and technology had delivered the machine gun and high explosive artillery shell, providing the means to kill millions of soldiers, and the new steam railways provided the means to deliver these millions onto the appointed battlefields. The Seeds of Materialism In 1859, Charles Darwin put forward the idea that the evolution of the species could be explained entirely by the physical processes of natural selection (i.e. survival of the fittest) and random mutation. 1

2 This famous theory provided a coherent explanation of the development of life on earth without the need of divine intervention, proposing that it was all a matter of chance operating over geological periods of time. Most shocking of all was the suggestion that the human race evolved from the apes, and the implication that we have an animal, rather than a divine, nature. While the great majority of people in the Western world were still believers in Christianity, many scientifically educated people were finding it difficult to reconcile the findings of science with the accepted teachings of the Church. This crisis was summed up by Nietsche s famous phrase God is dead from the Gay Science of 1887 (Nietzsche, 1887/2002). Religious conceptions of human nature and human history were also being challenged. By 1900, Sigmund Freud proposed the existence of an unconscious mind filled with repressed instinctual drives (Freud, 1911/1994) suggesting that all the energy used to build and maintain civilization was obtained as a result of the sublimation and rechanneling of these drives. This marked the beginning of a more unromantic and scientific view of the human condition. Karl Marx also set out to provide a materialistic explanation of human society and history, which he saw as an unremitting story of exploitation and injustice perpetrated by one class upon another (Marx & Engels, 1888/2002). As the century progressed, communist and socialist movements began to demand political reform and revolution to improve the conditions of the working classes rather than relying on the spirit of Christian kindness. The erosion of a shared European belief in Christianity contributed to the development of a fervent nationalism, which found expression in the colonial rivalries of the Great Powers (Britain, France, Germany and Russia). The growing industrial strength of Germany and the entrenched self-interest of the older powers, meant that some form of conflict was inevitable. What the world did not foresee were the horrific consequences of warfare in an industrial age. The Great or First World War, pitched millions of ordinary citizens into a military conflict that was to result in four years of continuous slaughter on an unimaginable scale. The battlefields of France represented the end of an era of optimism and faith in the value of European culture. No longer could the white European truly believe that he was bringing something good into the world. The Science of Uncertainty In 1900 Lord Kelvin commented that There is nothing new to be discovered in physics now, all that remains is more and more precise measurement. 2

3 Just five years later, Albert Einstein proposed his theory of relativity, and the modern era of mathematical physics began. Newton s gravitational theory was replaced with the four-dimensional abstraction of the space-time continuum, and thought experiments were now concerned with travelling at the speed of light where explanations of physical phenomena became primarily mathematical (and therefore much harder to understand from the perspective of sense perception). By the 1920s, with the development of quantum mechanics, this mathematical abstraction reached the point where matter itself passed beyond the grasp of the senses, with the paradox of matter simultaneously existing as both particle and wave. Quantum theory also proposed that the behaviour of matter is best explained by the laws of probability, rather than by the fixed deterministic laws of Newton s physics. This basic uncertainty as to where a particle will appear was formalised by Heisenberg s uncertainty principle which proposed the impossibility of fully specifying all physical attributes of a given particle at a given time Within our theme of the increasing abstraction of human mathematics and understanding, quantum physics represents a culmination of abstraction, where the universe is essentially understood as a mathematical entity. The Mathematics of Uncertainty In 1884, Gottlieb Frege developed the first satisfactory definition of number using the idea of the set of all sets having two elements to define the number two, and the set of all sets having three elements to define the number three, and so on. In 1900, David Hilbert proposed a concentrated program of research aimed at providing mathematics with a firm logical foundation. This was to be achieved by identifying a minimal set of fundamental assumptions or axioms, and then, using these axioms and clearly specified rules of proof, to logically deduce the rest of mathematics. In this way, mathematics was to be made certain, by reducing it to a logical system that was both complete and consistent (Complete means being able to prove any true statement using the axioms, and consistent means not being able to prove a statement that is inconsistent with the axioms). This program was not to survive the realities of the new century. Firstly, in 1902, Bertrand Russell pointed out certain contradictions that can arise when considering sets that contain themselves as members (known as Russell s Paradox). However, Russell was able to put the mathematical program back on track by defining a new set of contradiction free axioms, which he published in collaboration with Alfred North Whitehead in their famous Principia Mathematica of Then, in 1931, Kurt Godel published the first of his two papers that conclusively demonstrated the impossibility of creating a complete and consistent system of mathematics. Godel s incompleteness theorems ended the quest for logical certainty in mathematics, and had important implications for the subsequent study of computability. 3

4 One of the key unanswered questions posed by Hilbert in 1900 was whether there was an effective method for solving Diophantine equations. This stimulated a more general investigation of the decidability of mathematical questions, which was taken up by the English mathematician Alan Turing. In order to formalise exactly what is meant by an effective method, Turing developed the idea of a Turing machine in and used this idea to define computability (i.e. something is computable if it can be computed using a Turing machine). Using the same approach as Godel, Turing went on to prove that the problem of deciding whether a Turing machine will halt, given a particular procedure and set of arguments, and considering all possible values of these arguments, is itself undecidable. This came to be known as the Halting Problem (Turing, 1992). Turing went on to propose a Universal Turing machine that could take the description of any particular Turing machine as input and simulate its operation. Almost as a by-product of his central concern with the Halting Problem, Turing had defined the meaning of computation and proposed the first universal computing machine. The Mechanisation of Abstraction The Punched Card Machines At the turn of the 20th century, the most sophisticated commercial calculating devices were the punched card accounting systems, which combined the mechanical components of a calculating machine with an ability to electronically read information stored on punched cards. One of the first applications of these machines was in the US census of 1890, which used the Hollerith Electric Tabulating System developed by Herman Hollerith ( ). This machine worked using metal pins that would fall through the holes in census cards, generating numerical values using electrical signals, which were then automatically added to mechanical accumulators. In this way, census totals could be reliably and quickly calculated. The machine also allowed the census cards to be sorted by automatically opening an appropriate sorting bin after each read. By the 1920s, a wide range of punched card machines had been developed, and the International Business Machines Corporation (IBM) had emerged as one of the main players in the industry. In 1932, L. J. Comrie persuaded the Hollerith company to modify one of their machines, so that the contents of one register could be automatically transferred to other registers. By suitably wiring a plugboard, Comrie used the machine to simulate the operation of Babbage s Difference Engine, and was able to produce about 1000 lines of table per hour (Williams, 1997). 4

5 In 1935, IBM produced their first multiplying punch, which used telephone relay technology and could perform a single multiplication in about one second. The Zuse Machines It was the German Konrad Zuse who was credited with the development of the first fully automatically controlled computer. Zuse developed a revolutionary binary mechanism for storing numbers that was easier to build than a base-ten machine and also made it easier to treat numbers, logical operators and program instructions in the same way. Zuse implemented his ideas in the Z1 machine using a mechanical binary memory, made up of movable pins that could sit on either side of a slot. Hence each pin could represent a zero/one, yes/no binary number. The pins were then set and read using cleverly designed moving plates or mechanical logic gates, grouped together to form an arithmetic unit. Control input for the machine was encoded using holes punched in cinematographic film and output was displayed using electric lamps. Zuse s next machine, the Z2, retained the mechanical binary memory but used telephone relays for the arithmetic and control units. Although relay technology was relatively slow (as an electromagnetic field was used to operate a mechanical switch) and switches were fairly sizeable devices, they had the advantage that they could be controlled by electric current. This meant the physical location of the switch did not matter, it simply needed to be connected to the right wire. Zuse s assistant, Helmut Schreyer, became interested in using the newer vacuum tube technology as an alternative to relays to control the flow of current and eliminate the need of moving parts or switches and so dramatically increasing the speed of computation. This idea was not taken up and Zuse set about building a new machine (the Z3) entirely out of relay technology. When the Z3 was found to operate fairly reliably, the German military agreed to fund a larger machine capable of solving aviation problems involving large systems of linear equations. This final wartime machine (the Z4) reverted to a mechanical memory and incorporated several lookahead features to speed up calculation. The Bell Relay Machines In the late 1930s, George Stibitz and S. B. Williams designed a large relay-based calculator for the Bell Telephone Company. The resulting Complex Number Calculator could add, subtract, multiply and divide complex numbers entered by a human operator, using fixed point binary arithmetic but was not programmable and could not perform floating point arithmetic. In 1943, Bell and Stibitz developed the Relay Interpolator to simulate the motions of a target aircraft which was able to execute 31 different instructions and automatically accept input from punched paper tape. This programmable feature of the machine meant 5

6 it could perform more general purpose calculations, although it was limited to performing addition and subtractions. Stibitz s next machine, the so-called Ballistic Computer was developed in conjunction with the Interpolator, to calculate shell trajectories. This machine performed multiplication and division using floating point values and could accept input from 4 different tape readers, one of which would search through ballistic table values. Several more Bell machines were constructed during and after the war, all based on relay technology and decimal number systems. As with Zuse s machines, the slowness of the relay technology meant that the newer fully electronic machines finally replaced the Bell technology. Binary Logic Machines In 1869, William Jevons produced his first logic machine, consisting of keys, levers and pulleys, which calculated using the language of Boolean algebra. In 1881, Allan Marquand produced a similar but smaller machine, and proposed an electrical circuit design that would perform the equivalent operations. In 1936, the American psychologist, Benjamin Burack, succeeded in constructing an electrical logic machine, but did not publish his work until In 1938 Claude Shannon made an explicit connection between Boolean algebra and the representation of binary states in electrical circuits, thereby establishing the theoretical underpinnings of modern digital logic circuitry. Shannon worked with Stibitz on the Ballistic Computer at Bell Laboratories and became famous after the war for his landmark work A Mathematical Theory of Communication (1948), where he reduced the concept of information content to an analysis of bits or binary variables. The Harvard Machines During World War Two, Howard Aiken at Harvard University worked in collaboration with IBM and with existing IBM mechanical accounting machines. Using IBM s mechanical accumulator registers, Aiken developed a system of electronically controlled magnetic clutches that would engage number wheels from the registers with continuously rotating shafts (powered by a small motor). The longer a wheel was engaged with a shaft the further it would turn and hence the larger the number that was encoded. The machine resembled a semi-electrified and motorised Babbage engine, using the same idea of decimal number wheels arranged in columns and the same basic method of addition and propagating carries. Control was effected using standard IBM paper tape readers that converted punched holes into electrical signals. Also following Babbage s lead, later versions of the machine were able to switch between input tapes according to whether the value in a certain register had passed zero, thus introducing some ability to perform conditional branching. 6

7 The Harvard Mark I machine was completed in 1944 and immediately enlisted in the war effort, then in 1945 Aiken built a second machine running on the same basic principles, but constructed using electromagnetic relays. Aiken went on to build a Mark III (1949) and a Mark IV (1952) machine, both of which used relay and vacuum tube technology and had internally stored programs. In addition, Aiken developed the idea of storing the address of a number in special registers before executing an instruction which allowed much faster processing of arrays and loops. The First Electronic Machines In 1938 John Atanasoff and Clifford Berry had produced the basic circuit designs for a vacuum tube calculator, and by 1942, with some help from the US government, the pair had almost produced a working calculator. This machine consisted of an arithmetic unit made up of about 300 vacuum tubes, with a further 300 tubes used for control and memory purposes (see lecture slide 11). Atanasoff and Berry had also pioneered the development of a regenerative memory built from capacitors mounted on two rotating drums. Unfortunately, the Atanasoff-Berry Computer (or ABC) was abandoned. ENIAC Analog Beginnings In 1934 the Moore School was called in to build a copy of a mechanical differential analyser developed by Vannevar Bush at MIT. This machine was able to perform integration by converting the continuous movement of an operator, tracing a curve with a mechanical plotter, directly into a measure of the area under the curve. Until now, the machines we have considered have operated in discrete steps. From Schickard s calculator onwards, cogwheels, relays, or electrical circuits have been used to represent discrete states upon which calculations have been based. The use of discrete states means these machines are categorised as digital. Devices that convert one kind of continuous movement into another, such as Vannevar s differential analyser, fall into the category of analog machines. These machines also have a long history, and it is only since 1945 that the digital machine has come to prominence. Research into Electronics As a result of using the Moore School s differential analyser, the US army was able to reduce the average time to calculate a single artillery shell trajectory to 20 minutes. While this was a significant improvement, with the outbreak of the Second World War, the demand for new ballistic tables soon outstripped the analyser s ability to produce them. 7

8 At the same time, the Moore School was involved in the development of wartime radar, and had attracted a small group of talented researchers. One of these, John Mauchy, met with Atanasoff and Berry in 1940, and went on to inspect the early construction work on the ABC. As a result of this, Mauchy started discussing the idea of building a vacuum tube machine with the electrical engineer Presper J. Eckert at the Moore School. These discussions produced a short report in 1942 entitled The Use of High Speed Vacuum Tube Devices for Calculating, which predicted that such technology could reduce the time to calculate a trajectory to around 100 seconds. Initially the report attracted little interest, but, by 1943, the production of ballistic tables had fallen so far behind schedule that the U.S. army decided to take a chance with Mauchy s new idea. At first the machine was to be called the Electronic Numerical Integrator, as integration was to be its main task, but after recognition that it would have more general purpose abilities, the words and Computer were added, hence the ENIAC project was born. The Machine ENIAC was first commissioned in 1946, taking three years to complete, and arriving too late to make a difference to the outcome of the war. It finally cost the U.S. military $486,804 (six times over the original budget), using 18,000 vacuum tubes, 1500 relays, 70,000 resistors and 10,000 capacitors. It weighed over 30 tons and stood 8 ft high, 3 ft wide and 100 ft long, consuming 140 kilowatts of power and requiring two 12 hp blowers to keep it cooled. Nevertheless, at the time it was a marvel of technology, successfully performing all its calculations electronically, and only relying on mechanical devices for input and output. While ENIAC was programmable, it did not have a general purpose memory, and so did not accept program input in the way we have come to expect. Instead, program data was entered via an IBM card reader, and program execution was controlled by physical rewiring (using plugboards), and by a master programmer switching device, which automatically controlled iteration and sequencing operations. Individual ENIAC modules could also be set to perform particular calculations within a given master program sequence. In terms of intellectual heritage, the influence of Atanasoff and Berry s work on ENIAC has already been mentioned. In addition, many of the basic ideas behind the functioning of ENIAC can be seen as electronic analogs of Aiken s work on the Harvard Mark 1 (Williams, 1997), which in turn was influenced by Babbage s pioneering engines. For example, ENIAC used an electronic implementation of Babbage s anticipating carriage mechanism, which Babbage himself thought was one of his most important inventions. So, although Babbage was almost completely forgotten in his own century, there is evidence his ideas survived to influence our modern world. 8

9 Colossus The Background During and after the First World War, Allied forces had more or less successfully cracked existing German military ciphers, which were based on substitution techniques developed in the 19th century. However, with the development of electromechanical technology, it became possible to build machines that could perform far more sophisticated encryption, while still being able to decipher messages with relative ease. The most famous of these machines was the Enigma, developed in Germany by Arthur Scherbius. After the Germans adopting these machines, attempts to decipher military communications became less and less successful, until only the Polish remained serious about breaking the Enigma. The First Breakthrough In 1931, thanks to a disaffected employee working for German military intelligence, the French secret service were able to obtain documents describing exactly how the military Enigma machine was constructed. The French passed on their information to the Poles who were able to construct their own replica machine. However, having a replica was not enough. The Enigma machine worked by setting three rotors, each having 26 possible positions, and then typing in a letter from a keyboard. According to a system of wires, the signal from the keyboard was passed through the rotors and encrypted to another (different) letter. Then, one rotor was moved round one position before entering the next letter, causing a change in the wiring. This meant that the same alphabet would be encrypted differently for each possible position of the rotors (= = 17,576 possibilities). All that was needed to send and receive messages was that two Enigma machines were synchronised by having the rotors in the same starting positions. In addition, a plugboard was mounted on the front of the machine which allowed the values of six letters to be swapped before passing through the rotors, and the rotors themselves could be placed in different orders. Putting these features together meant there were over 10,000 million million possible ways to encrypt a particular message. Lecture slide 18 shows an Enigma machine with the three rotors removed and the plugboard partially obscured by the wooden panel at the front. To set up the machines, the German military sent a special message header specifying settings for the rotors. This sequence of three characters was then repeated. It was the Polish mathematician, Marian Rejewski who realised that this repetition of the rotor positions provided a door into the cipher. By knowing that for all messages the 1st and 4th, 2nd and 5th, and 3rd and 6th characters were the same, and by having reconstructed an actual machine, Rejewski was able to catalog each possible setting of the rotors for a given input, independently of the plugboard settings. 9

10 As there were 6 possible orderings of the rotors, this meant there were 17,576 6 = 105,456 possible rotor settings. After a year of work, each one of these settings was calculated and indexed, and the problem of breaking the day code became a matter of finding the correct plugboard settings. As only 12 letters were swapped by the plugboard, many clues were available from the sense of the partially decoded message, and the final decryption was relatively simple. Unfortunately, the Germans made some minor changes in the way the message header was transmitted and the Polish catalogs became useless. To counter this, Rejewski devised an automatic system of checking rotor positions using electromechanical relay devices called bombes. Each bombe was a reconstruction of an Enigma machine, with electrically driven rotors, that could check all possible settings of the rotors within 2 hours. If a correct day-key setting was found, the machine would automatically stop and illuminate a light bulb. By constructing six bombes (one for each possible ordering of the rotors) a complete search for the day-key could go on in parallel, and be completed within 2 hours. Then, at the end of 1938, the Germans developed two more rotors for the Enigma and changed the number of plugboard swaps from 6 to 10. Breaking this cipher would have required another 54 bombes, costing several times more than the annual budget of Rejewski s department. On 16th August 1939 two Enigma machine replicas and blueprints for the bombes were smuggled to Britain. On the 1st September, Germany invaded Poland and the Second World War began. Bletchley Park British code-breaking activity during World war Two was located in Bletchley Park in Buckinghamshire. It was here that Alan Turing, and many other brilliant mathematicians, linguists and engineers, were brought together to work on Enigma and the later Lorenz ciphers. Turing, started where Rejewski had left off and devised more sophisticated electromechanical search engines known as Turing bombes. With the stimulus of war, Turing bombes were soon built in large enough numbers to quickly solve the new 60 rotor setting problem The combined result of this work was that nearly all intercepted German army and airforce communications were successfully deciphered throughout the duration of the war. 10

11 The Lorenz Cipher The German High Command communicated using the Lorenz cipher, which encrypted teleprinter messages using the 5 bit Baudot Code and so did not share any features of the Enigma. The first Lorenz message deciphered in August 1941, when a German operator keyed in a 4,000 character message twice using the same encryption settings. By tracking small differences between the first and second messages, another Bletchley Park cryptoanalyst, John Tiltman, was able to fully decipher both texts. This information was passed on to Bill Tutte s team, who spent two months reconstructing the operation of the machine that must have produced it. This machine was then physically reconstructed, in early 1942, using relay technology at the Post Office Research Labs at Dollis Hill in North London. At this time the mathematician Max Newman became involved in the Lorenz project. He suggested that the double-delta method (devised by Bill Tutte to find the start positions of the Lorenz wheels) could be automated using an electronic vacuum tube machine. The designs were made and Dollis Hill was again commissioned to build what came to be known as the Heath Robinson. This machine showed in principle that the Lorenz cipher could be broken automatically, but was still too slow and had problems synchronising the high speed reading of two paper tape inputs. In March 1943, Newman went back to Dollis Hill and asked Tommy Flowers to design and a faster more reliable machine. Flowers, a talented electrical engineer, envisaged an ambitious new machine, using 1,500 vacuum tubes that would do away with the second tape input by generating wheel patterns electronically. There was scepticism that such a large device could work reliably, and consequently Flowers and the team at Dollis Hill went on to build the machine on their own. The project was completed within 9 months, and by January 1944 a working version of the Colossus Mark 1 was successfully deciphering Lorenz code at Bletchley Park. This machine was able to read Baudot Code at the rate of 5,000 characters per second, and times for decipherment were reduced from days to hours. Various versions of Colossus went on to decipher key messages from German High Command that determined the course of the Allied invasion of Normandy later in As that invasion continued and German landlines were destroyed, Hitler s strategy became an open book to the Allies and the course of the war was considerably shortened as a result. 11

12 TheWorld s First Computer? For many decades ENIAC was considered to be the world s first electronic digital computer. However, it was not until 1970 that the British government admitted to the existence of Colossus and only in the 1990s did details of its operation come to light. Since then much debate has centred on whether Colossus was really the world s first electronic computer. As with all such questions, the answer depends on a definition of terms. In this case, given that Colossus was specifically designed to process Lorenz cipher code, can it really be considered as a general purpose machine in the modern sense? The answer to this must be no. However, Colossus was programmable to the extent of executing various crosscorrelation algorithms controlled by switches, and in an experiment after the war, it was rewired to perform simple multiplication. Colossus also contained and and or gate logic circuits which could be plugged together in any combination, with counter and counter control circuitry, sophisticated photo-electric paper tape readers (which also controlled the clock cycles), shift registers and a master control panel. It was a symbolic/logic computer, rather than an ENIAC style numerical calculator, and as such had features that ENIAC lacked. Alternatively, if we look at ENIAC s claim to be the first electronic computer, it was certainly able to perform a wider range of tasks than Colossus, but it was also completed considerably later. Again, ENIAC was not programmable in the sense we understand it today. Like Colossus, it was only rewirable and reswitchable. Neither machine treated program instructions and program data in the same way or possessed a general purpose memory. In terms of the theme of this Chapter, we have not yet arrived at a Universal Computer in the sense that Alan Turing envisaged. Instead we have the potential for such a machine, and the electronic circuitry to implement it. The missing ingredient is the general purpose memory of the stored program computer. The Stored Program Computer The Universal Turing Machine The idea of a universal computing machine was first formalised by Alan Turing in Turing imagined a machine that could perform a small set of basic tasks: firstly, it could assume various states, secondly, it could read and write symbols to and from squares on a paper tape, and thirdly, it could move the tape one square to the left or right. The machine then operated according to a set of internal rules which specified, for each possible state of the machine and corresponding symbol on the tape, what new state the machine would assume, what symbol it would write to the tape and which direction it would move the tape (left or right). Using these simple definitions, Turing demonstrated that his machine could execute any effective method. 12

13 In modern terms, the Turing machine, given the right set of state and symbol transformation rules, is able to simulate the execution of any computer program (although translating a program into a set of instructions for a Turing machine can be quite awkward). Turing went on imagine a Universal Turing machine, capable of reading a symbolic description of any particular Turing machine, and then simulating the operation of that machine, for all subsequent input. In this single step, Turing foresaw the development of the modern computer. Although modern machines do not read from paper tape, they do read in the description of a program and store it internally. They also use the same symbolic language to represent program instructions, and all forms of data. While the physical realisation differs, the fundamental ideas are the same, and with hindsight we can see the Universal Turing machine is the true forerunner of the modern computer age. The Memory Problem At the end of the war, neither Colossus nor ENIAC had the ability to simply read in a program and execute it. Vacuum tubes were capable of performing up to a million switching operations per second, where as the fastest relays could only manage 100 to 200 switches per second. This meant that reading in and storing program instructions using paper tape or relay technology was not practical, as all the speed benefits of electronic computation would be lost. Consequently, the early machines used plugs and switches to set up a particular program, controlling which operation will happen next, when to exit a loop, etc. What was needed was a general-purpose memory store that could be erased and rewritten at electronic speeds and where data could remain for long periods of time without degrading. This memory had to be fairly inexpensive as it was needed in large quantities. Therefore, although the ideas were in place for a general purpose machine, various technical hurdles had to be overcome before such a machine could be constructed. The von Neumann Architecture In the U.S.A., Mauchy and Eckert, had already seen the need for a stored program machine and in conjunction with John von Neumann, the team at Moore produced a famous paper in 1945, entitled First Draft of a Report on the EDVAC. This report described the von Neumann architecture for a stored program computer, a term which has now become synonymous with the standard architecture of all modern machines. 13

14 Controversy ensued, as von Neumann was the only named author of the report, although Mauchy and Eckert had developed many of the ideas independently.. Most probably, von Neumann gave a certain rigour and formality to the existing ideas of the Moore school, making the EDVAC (Electronic Discrete Variable Arithmetic Computer) architecture general and abstract enough to still be relevant today. The original report divided the computer system into four main parts: the central arithmetical unit, the central control unit, the memory, and the input/output devices. Selfevidently, the arithmetic unit was to perform arithmetic and the control unit was to control the sequence of program execution. Then the memory was used to store both numerical data and numerically coded instructions, with the input/output units acting as the user interface. The key element was the explicit separation of memory from processing, and the use of that memory to encode both numbers and instructions. This simple scheme was to act as the blueprint not just for the EDVAC machine, but for the rest of the mainstream computer revolution. The First Stored Program Computers In terms of our historical study, the invention of the stored program machine represents the first computer in the modern sense of the word. According to the standard version of the story, von Neumann s report in 1945, and the subsequent course offered at the Moore School in 1946 led directly to the British using these ideas to build the world s first stored program machines. With hindsight, we can see that the secrecy surrounding the development of Colossus obscured the facts. In Britain there was still controversy between Cambridge and Manchester Universities as to who really built the first stored program machine. While the Manchester Baby was working by June 1948, it was only really a prototype machine, whereas Cambridge s EDSAC (Electronic Delay Storage Automatic Calculator, completed in May 1949) was the first fully functional machine capable of solving realistic problems. The Manchester team relied on a high speed digital CRT (Cathode Ray Tube) memory device, developed in October 1947 by Freddie Williams and Kilburn. With the memory problem solved, the construction of a working prototype proved (fairly) straightforward. The Cambridge team based their work more closely on the American design for EDVAC and built their memory out of mercury delaylines that stored information acoustically. Australia was also one of the first countries to design and build a stored program computer. This machine, the CSIRAC (Council for Scientific and Industrial Research Automatic Computer), was designed and built in Sydney and was operational by November The project was headed by Trevor Pearcey and Maston Beard for the CSIR (now CSIRO). Pearcey and Beard had both worked on British radar systems during 14

15 the war and were in contact with the teams working at Manchester and Cambridge. The CSIRAC was the fifth stored program machine operational in the world at that time. However, the British and Australian lead in computer technology was relatively shortlived and was partly explained by the acrimonious break-up of the American Moore School team. Mauchy and Eckert became involved in a lengthy dispute over patents for their work on ENIAC and EDVAC, and split with von Neumann over his taking over of their ideas. For these and other reasons, the EDVAC project was not completed until American power and wealth were to prove decisive, with American companies, particularly IBM, soon coming to dominate the industry. More generally, the end of the Second World War marked the end of the preeminence of European culture and science. From now on, the centre of scientific research was to be in America, and American popular culture was to have the major influence on the people of the world. Even more fundamentally, the Western free world now depended on America for protection from the communist block. It was the new era of superpowers, masscommunications and consumerism in which we still live. As far as the computer was concerned, few people at the end of the war could see beyond a machine performing more and more complex mathematical calculations. The development of computer languages and compilers, the transistor, the silicon chip and reliable, cheap, high speed memory were still to come. Nevertheless, with the advent of the stored program computer, the basic idea of computation and what a computer is had been settled. References Freud, S. (1911/1994). The interpretation of dreams. Barnes and Noble, Jamesburg, NJ. Marx, K., & Engels, F. (1888/2002). The communist manifesto. Viking Penguin, London. Neumann, J. von. (1958). The computer and the brain. Yale University Press, New Haven. Nietzsche, F. (1887/2002). The gay science. Cambridge University Press, Cambridge. Singh, S. (1999). The code book. Fourth Estate, London. Turing, A. M. (1992). Mechanical intelligence. Elsevier Science Publishing Co., New York. Williams, M. R. (1997). A history of computing technology. IEEE Computer Society Press, Los Alamitos, CA. 15