by George Dyson
In 1941, German telecommunications began to be encrypted with much faster, digital equipment: the Geheimschreiber, manufactured by Siemens, and the Schlüsselzusatz, manufactured by Lorenz. These devices, known collectively as Fish, and derived from automatic Teletypewriter equipment, produced a sequence of 0s and 1s (the key) that was then added to the binary representation of an unenciphered (plaintext) message and output for transmission as ordinary 5-bit Teletypewriter tape. The machine’s 12 code wheels, of unequal length, were circumscribed by a combined total of 501 pins that could be shifted between 2 positions, giving the system 2501 (or about 10150) possible states. The key was added modulo 2 to the plaintext message (counting by 2 the way we count hours by 12, so that 0 + 1 = 1 and 1 + 1 = 0), with 1 and 0 represented by the presence or absence of a hole in the tape. Adding the key to the enciphered text a second time would return the original text.
Fish traffic was beyond the reach of the electromechanical bombes. Electronics was the only hope of catching up. A series of machines named “Heath Robinson” were built on the principle that by simultaneously scanning two different (and relatively prime) lengths of punched paper tape as continuous loops, all possible combinations of the two sequences could be compared. Based on standard teleprinter tape and standard 5-bit teleprinter code, but running at high speed through photoelectric heads, the Heath Robinsons used electronic circuits to compare the two sequences, but it was difficult to maintain synchronization between two tapes.
It was then proposed by Thomas H. Flowers, an engineer working for the British Post Office’s telecommunications research station at Dollis Hill, to eliminate one of the tapes by reading its sequence into an internal store consisting of fifteen hundred vacuum tubes, or, in British terminology, valves. This internal memory could then be synchronized to the sequence of pulses read from the other tape, which could be run without sprockets at much higher speeds by friction drive. “The tapes were read at 5,000 characters per second, [which] implies a tape speed of nearly 30 miles per hour,” recalled Jack Good. “I regard the fact that paper teleprinter tape could be run at this speed as one of the great secrets of World War II!”32
With practice, it was possible to run loops of tape as much as two hundred feet in length. The new machine, code-named Colossus, was constructed under the supervision of Flowers and operated and programmed under the direction of Max Newman, who had set all these wheels in motion when he sparked Turing’s original interest in the Entscheidungsproblem in 1935. Colossus was an electronic Turing machine, and if not yet universal, it had all the elements in place.
Colossus was so successful (and subspecies of Fish so prolific) that by the end of the war ten Colossi were in use, the later versions using 2,400 vacuum tubes. They were programmed by a plugboard and toggle switches at the back of the machine. “The flexible nature of the programming was probably proposed by Newman and perhaps also Turing, both of whom were familiar with Boolean logic, and this flexibility paid off handsomely,” recalled I. J. Good. “The mode of operation was for a cryptanalyst to sit at Colossus and issue instructions to a Wren [Women’s Royal Navy Service] for revised plugging, depending on what was printed on the automatic typewriter. At this stage there was a close synergy between man, woman, and machine.”33 As a step toward the modern computer, Colossus represented as great a leap as the ENIAC, and was both running and replicated while the one-of-a-kind ENIAC was still being built. Each Fish was a form of Turing machine, and the process by which the Colossi were used to break the various species of Fish demonstrated how the function (or partial function) of one Turing machine could be encoded for execution by another Turing machine. Since the British did not know the constantly changing state of the Fish, they had to guess. Colossus, trained to sense the direction of extremely faint gradients that distinguished enciphered German from random alphabetic noise, was the distant progenitor of the search engine: scanning the Precambrian digital universe for fragments of the missing key, until the pieces fit.
It was the alumni of Bletchley Park who were first to demonstrate a working stored-program computer (the Manchester Small Scale Experimental Machine, which ran its first program on June 21, 1948) and first to construct a kilobit-scale electronic memory (the electrostatic Williams tube). But the driving force behind computer development had shifted across the Atlantic, from the logical puzzle of cryptanalysis to the numerical design of hydrogen bombs. When Bletchley Park disbanded, the Official Secrets Act handicapped those who could not refer openly to their wartime work. The ENIAC was publicly unveiled in February of 1946, while the existence of Colossus would not be officially acknowledged for thirty-two years.
The extent, if any, of direct collaboration between Turing and von Neumann remains unknown. The British had a substantial nuclear weapons research group and, in consultation with von Neumann, made important contributions to Los Alamos. The Americans had a substantial group of cryptanalysts and, in consultation with Turing, contributed to the effort at Bletchley Park. Turing was in the United States between November 1942 and March 1943, and von Neumann was in England between February and July 1943. Both visits were secret missions, and there is no record of any wartime contact between the two pioneers.
“There was some hold-up about his job, which involved a useless period of idling in New York,” says Sara Turing about Alan’s three-month visit to the United States during the war. “He seems to have taken the opportunity to visit Princeton and probably saw something of the progress of computing machinery in the States. He returned in a destroyer or similar naval vessel and experienced a good tossing on the Atlantic.” Jack Good remembers that upon Turing’s return from the United States in March of 1943, he discussed “a problem about bags of gunpowder at the points in a plane with integer coordinates. Given the probability that the explosion of one bag will cause adjacent ones to explode, what is the probability that the explosion will extend to infinity?”34 If you had to characterize the problem of determining the probability of a nuclear chain reaction without mentioning fission cross-sections, bags of gunpowder on the integer plane is a good mathematical fit.
J. R. Womersley, superintendent of the Mathematics Division of the National Physical Laboratory, who had read “On Computable Numbers” and become interested in Turing machines before the war, had been sent to the United States in the spring of 1945 to survey the latest (and still-secret) computer developments, including the Harvard Mark I tape-controlled electronic calculator, which he described in a letter home as “Turing in hardware.” Womersley reported to Douglas R. Hartree, who reported to Sir Charles Darwin, director of NPL and grandson of the Charles Darwin. “JRW sees ENIAC and is given information about EDVAC by von Neumann and Goldstine,” Womersley noted in 1946.35 In June of 1945, Womersley met with Max Newman, and asked to meet Turing. “Meets Turing same day and invites him home,” Womersley noted. “Shows Turing the first report on the EDVAC and persuades him to join N.P.L. staff, arranges interview and convinces Director and Secretary.”36 In September of 1945 Turing was assigned to study the ENIAC and EDVAC reports. Womersley notes, “Turing decides that mechanisms proposed for EDVAC are appropriate to his ideas.”37
“I am of course in close touch with Turing,” Max Newman wrote to von Neumann in early February 1946, explaining that “about eighteen months ago I had decided to try my hand at starting up a machine unit when I got out.” Since technical details about the ENIAC were still restricted, and the existence of Colossus was not even acknowledged, a personal visit was arranged. “What I should most like is to come out and talk to you (for one thing I am still a bit cramped in discussing the past, and have to ask you not to put 2 and 2 together too accurately, and not to pass it on if you do.)”38
Von Neumann secured an Institute stipend for Newman to visit Princeton. In January 1947, Turing himself visited, reporting that “my visit to the U.S.A. has not brought any very important new technical information to light, largely, I think, because the Americans have kept us so well informed during the last
year.… The Princeton group seem to me to be much the most clear headed and far sighted of these American organizations, and I shall try to keep in touch with them.”39
The war had scrambled the origins of new inventions as completely as a message passing through an Enigma machine. Radar, cryptanalysis, antiaircraft fire control, computers, and nuclear weapons were all secret wartime projects that, behind the security barriers, enjoyed the benefit of free exchange of ideas, without concern for individual authorship or peer review. Von Neumann served the role of messenger RNA, helping to convey the best of the ideas—including the powers of Turing’s Universal Machine. Among the bound volumes of the Proceedings of the London Mathematical Society, on the shelves of the Institute for Advanced Study library, there is one volume whose binding is disintegrated from having been handled so many times: Volume 42, with Turing’s “On Computable Numbers,” on pages 230–65.
Turing and von Neumann were as far apart, in everything except their common interest in computers, as it was possible to get. Von Neumann rarely appeared in public without a business suit; Turing was usually unkempt. “He tended to be slovenly,” even his mother admits.40 Von Neumann spoke freely and with great precision; Turing’s speech was hesitating, as if words could not keep up with his thoughts. Turing stayed in hostels and was a competitive long-distance runner; Von Neumann was resolutely nonathletic and stayed in first-class hotels. Von Neumann had an eye for women, while Turing preferred men.
When von Neumann spoke about computing, he never mentioned artificial intelligence. Turing spoke about little else. Turing and von Neumann designed different styles of computer and wrote different styles of code. Von Neumann’s design was captured in the “First Draft of a Report on the EDVAC” of June 30, 1945, and the “Preliminary Discussion of the Logical Design of an Electronic Computing Instrument” of June 28, 1946. Turing’s design was captured in his “Proposed Electronic Calculator,” written for the National Physical Laboratory in the brief interval between being shown the EDVAC report in September 1945 and the end of the year. He delivered a complete description of a million-cycle-per-second Automatic Computing Engine (ACE), accompanied by circuit diagrams, a detailed physical and logical analysis of the internal storage system, sample programs, detailed (if bug-ridden) subroutines, and even an estimated cost of £11,200.41 As Sara Turing later explained, her son’s goal was “to see his logical theory of a universal machine, previously set out in his paper ‘Computable Numbers,’ take concrete form.”42
After a comparison of available forms of storage, ranging from punched paper tape through “cerebral cortex” to electrostatic storage tubes, Turing specified mercury-filled acoustic delay lines for high-speed storage. He estimated cost, access time, and “spacial economy” (in digits/liter) for all forms of storage, putting the cost of a cerebral cortex at £300 per annum—his King’s College fellowship for the year. Viewed as part of a finite-state Turing machine, the delay line represented a continuous loop of tape, 1,000 squares in length and making 1,000 complete passes per second under the read/write head. Turing specified some 200 tubes, each storing 32 words of 32 bits each, for a total, “comparable with the memory capacity of a minnow,” of about 200,000 bits.43 Taking ten pages of the proposal to do so, Turing worked out the storage capacity, attenuation, noise, temperature sensitivity, and regenerative requirements, all from first principles.
Turing’s ACE became bogged down in postwar bureaucracy and, like Babbage’s Analytical Engine, was never built. In May 1950 a partial prototype (the Pilot ACE) was finally completed and “proved to be a far more powerful computer than we had expected,” wrote J. H. Wilkinson, even though its mercury delay lines held only 300 words of 32 bits each. “Oddly enough much of its effectiveness sprang from what appeared to be weaknesses resulting from the economy in equipment that dictated its design.”44
Turing, having had a taste of wartime “action this day” at Bletchley Park, grew impatient with the NPL administration, and the administration grew frustrated with Turing’s tendency to leap ahead. “Our big computing engine … has now got to the stage of ironmongery,” Sir Charles Darwin wrote to his superiors in July 1947, explaining that Womersley and Turing “are both agreed that it would be best that Turing should go off it for a spell.”
“He wants to extend his work on the machine still further towards the biological side,” Darwin continued. “I can best describe it by saying that hitherto the machine has been planned for work equivalent to that of the lower parts of the brain, and he wants to see how much a machine can do for the higher ones; for example, could a machine be made that could learn by experience?” Darwin finally got to the substance of his request. Turing, he explained, “would be content with something like half-pay … and indeed said that he would really prefer it, because if he were earning full pay, he would feel that ‘I ought not to play tennis in the morning, when I want to.’ ”45
Turing took a leave of absence from NPL, returning to his King’s College fellowship for a year before resigning from NPL in May 1948 to join Max Newman’s computing group at Manchester University, where his interests in mechanical intelligence found free rein. “An unwillingness to admit the possibility that mankind can have any rivals in intellectual power,” Turing wrote in his sabbatical report submitted to the NPL in 1948, “occurs as much amongst intellectual people as amongst others: they have more to lose.”46
Turing’s approach to machine intelligence was as unencumbered as his approach to computable numbers ten years before. He continued, once again, from where Gödel had left off. Does the incompleteness of formal systems limit the abilities of computers to duplicate the intelligence and creativity of the human mind? Turing summarized the essence (and weakness) of this convoluted argument in 1947, saying that “in other words then, if a machine is expected to be infallible, it cannot also be intelligent.”47 Instead of trying to build infallible machines, we should be developing fallible machines able to learn from their mistakes.
“The argument from Gödel’s and other theorems rests essentially on the condition that the machine must not make mistakes,” he explained. “But this is not a requirement for intelligence.”48 Turing made several concrete proposals. He suggested incorporating a random-number generator to create what he referred to as a “learning machine,” granting the computer the ability to take a guess and then either reinforce or discard the consequent results. If guesses were applied to modifications in the computer’s own instructions, a machine could then learn to teach itself. “What we want is a machine that can learn from experience,” he wrote. “The possibility of letting the machine alter its own instructions provides the mechanism for this.”49 He pointed out that “paper interference” with a universal machine was equivalent to “screwdriver interference” with actual parts. In 1949, while developing the Manchester Mark I (prototype for the Ferranti Mark 1, the first stored-program electronic digital computer to be commercially produced), Turing designed a random-number generator that instead of producing pseudo-random numbers by a numerical process included a source of truly random electronic noise. This avoided von Neumann’s “state of sin.”
Turing also explored the possibilities of “unorganized Machines … which are largely random in their construction [and] made up from a rather large number N of similar units.”50 He considered a simple model with units capable of two possible states connected by two inputs and one output each, concluding that “machines of this character can behave in a very complicated manner when the number of units is large.” He showed how such unorganized machines (“about the simplest model of a nervous system”) could be made self-modifying and, with proper upbringing, could become more complicated than anything that could be otherwise engineered.51 The human brain must start out as such an unorganized machine, since only in this way could something so complicated be reproduced.
Turing drew a parallel between intelligence and “the genetical or evolutionary search by which a combination of genes is looked for, the criteri
on being survival value. The remarkable success of this search confirms to some extent the idea that intellectual activity consists mainly of various kinds of search.”52 Evolutionary computation would lead to truly intelligent machines. “Instead of trying to produce a programme to simulate the adult mind, why not rather try to produce one which simulates the child’s?” he asked. “Bit by bit one would be able to allow the machine to make more and more ‘choices’ or ‘decisions.’ One would eventually find it possible to program it so as to make its behaviour the result of a comparatively small number of general principles. When these became sufficiently general, interference would no longer be necessary, and the machine would have ‘grown up.’ ”53
Turing gave provocative hints about what might lie ahead. “I asked him under what circumstances he would say that a machine is conscious,” Jack Good recalled in 1956. “He said that if the machine was liable to punish him for saying otherwise then he would say that it was conscious.” Lyn Newman remembers long discussions between Max Newman and Turing over how to build machines that would modify their own programming and learn from their mistakes. “When I heard Alan say of further possibilities ‘Wh–wh–what will happen at that stage is that we shan’t understand how it does it, we’ll have lost track’—I did find it a most disturbing prospect,” she reported in 1949. Jack Good would later explain that “the ultraintelligent machine … is a machine that believes people cannot think.”54