Alan Turing: The Enigma The Centenary Edition
Page 40
His liaison job accomplished, Alan left Washington at the end of December. He had been working at the nerve-centre of the alliance, at its point of equilibrium. The British contribution had not yet been overtaken by the American. The Casablanca conference, from 14 to 24 January, saw Churchill as Roosevelt’s equal. For the last time, the Americans supported a British strategy to regain the Mediterranean. For the first time, Britain was to act as an American base. It was also the equilibrium point of the war. It was taking far longer than expected to clear North Africa, Montgomery missing some ultra-fine chances, with disastrous consequences all over the globe. The Russian front was still undecided. Nothing was clear, despite the demand for ‘unconditional surrender’. The crude ‘strategic bombing’ was endorsed, for want of anything better. But the Atlantic battle, agreed at Casablanca to remain the top priority, had taken a turn. For the first time, new Allied ship construction was exceeding losses.
Alan went to Saunderstown, Rhode Island, to visit Jack and Mary Crawford again, as he had done from Princeton. But Jack had died on 6 January, a few days before Alan arrived. His widow asked him to stay on a few days nonetheless. Then his direction turned too. He went down to New York City, arriving4 at the Bell Laboratories building on West Street, by the piers, on the afternoon of 19 January 1943. And for two months he soaked himself in the electronic technology of speech encipherment.
Like most organisations devoted to secret work, Bell Labs operated in a cellular fashion so that people never knew what was happening outside their own department. Alan, however, was free to move into any ‘cell’ he wished, although he had to be careful not to transfer information himself. It came through to the Bell engineers with whom he worked that his ‘clearance’ had come not from the Army or the Navy, but from the White House itself. Most of his time, however, was spent in one particular ‘cell’ which had the responsibility of trying to crack speech encipherment systems which had been proposed. He made an impression from the start, for within an hour of arrival he had solved a problem. It involved a scrambling system in which time segments were permuted, by means of nine magnetic heads simultaneously reading a magnetic tape. ‘That ought to give you 945 codes,’ said Alan when this was explained, ‘It’s only 9x7x5x3.’ It had taken one of their technicians a week to work it out.
For his first week, Alan acquainted himself with all the projects on which they were working, and became particularly keen on taking up one of them himself. This was the challenge presented by an RCA engineer, who had devised a system in which a speech signal would be multiplied by a key signal. It presented quite an unusual problem. On 23 January, Alan announced he had begun to think of a way of attacking it, and then he came in after the weekend convinced of its possibilities. His idea involved the use of the Vocoder.
Alan had probably already learnt about the Vocoder when in Britain, since Dollis Hill had received information on it in 1941. It was a very advanced piece of communications technology, which had been patented by the Bell engineer H. W. Dudley in 1935 and developed since then at the Bell laboratories. The idea of the Vocoder was to abstract the essential elements of speech, throwing away much of its redundancy, and conversely to reconstruct the speech signal from its essential components. One way of thinking of this process was to regard it as reducing the bandwidth, or frequency range, of a speech signal.
Any Bell Labs engineer would be familiar with the idea of reducing the frequency range of speech, since the telephone cut off sound above 4000 Hz. The resulting lack-lustre tones were still perfectly comprehensible, the point being that higher frequencies were redundant in ordinary applications. But to reduce that frequency limit much further would produce a sort of miserable grunting, which would not do at all. The Vocoder did something far more sophisticated. It collected information about the amplitude of the speech signal at each of ten frequencies up to 3000 Hz, and also took an eleventh component which coded either the fundamental pitch of the sound or (during unvoiced sounds like ssss) an absence of pitch. Each of these eleven signals required a frequency range of only 25 Hz. In this way, sufficient information was abstracted for intelligible speech to be reconstructed, and yet the total bandwidth was confined to less than 300 Hz.
Alan had already suggested that the principle of the Vocoder, taking samples at ten different frequency levels, could be applied to an attack on the time-segment permuting type of speech scrambler – perhaps with the idea of recognising the neighbouring segments automatically. His idea for applying the Vocoder to the RCA multiplying speech cipher was something much more sophisticated, and he said it would require at least a week of computational work to see if it was feasible. In his second week at Bell Labs he settled down to this work, which involved calculations with Hermite polynomials, and in his third week he had some assistance with the calculations.
But Alan was also involved in a quite separate ‘cell’, which was devoted to creating the world’s first totally unbreakable speech encipherment system. This was the most advanced work in progress at Bell Labs, and its best guarded secret. The original goal had been that of finding a way to encipher speech on the Vernam principle, so that if a one-time key were employed, the result would be as unbreakable for speech as it was for telegraph signals. With this end in view, they had attacked the quite novel problem of representing speech by the discrete 0’s and 1’s used in a Vernam cipher system.
They had begun in 1941 with the Vocoder, and tried to adapt it to their purpose by approximating its eleven outputs as being either ‘on’ or ‘off’. This, however, resulted in a ‘badly mutilated’ speech signal. Accordingly they had abandoned the simple binary ‘on or off of the Vernam cipher, and instead approximated the Vocoder outputs not by two possible levels, but by six. The eleventh signal required a finer tuning than the others, and was allowed thirty-six levels. The effect was to encode the speech-signal as a total of twelve streams of ‘base six’ digits, like 041435243021353…. Each such stream would then be added in modular fashion* to a similar but random key sequence, and the result transmitted. At the other end the identical key would be subtracted, and the speech reconstituted. The speech signal was to be sampled for its ‘levels’ fifty times a second, which meant that the transmission was roughly equivalent to sending 300 teleprinter characters a second. They had succeeded in devising the equivalent of a one-time pad system for speech.†
The development was given the intriguing name of ‘Project X’ or the ‘X-system’. By November 1942 an experimental model had been installed at New York and tested with5 ‘a synthetic set of signals from a signal generator that had previously been sent to England’. In January 1943 they were beginning to assemble the first model intended for operational use. There were tremendous technical obstacles. Not only was the basic Vocoder already very complicated, but it required a large number of further components for taking the discrete (or ‘quantised’) levels. It also required the allocation of seventy-two different frequencies, for the twelve streams of digits were to be played like music, with a different frequency, not a different amplitude, for each possible digit. The system also required perfect synchrony between sender and receiver, and had to allow for the fading and time delays in the Atlantic ionosphere.
The result was a roomful of electronic equipment at each end of the system:
A terminal occupied over 30 of the standard 7-foot relay rack mounting bays, required about 30kW of power to operate, and needed complete air conditioning in the large room housing it. Members working on the job occasionally remarked about the terrible conversion ratio – 30kW of power for 1 milliwatt of poor-quality speech.
But it worked, which was the main thing. For the first time, secret speech could cross the Atlantic. Alan’s inspection of the apparatus on behalf of the British government preceded a formal Anglo-American agreement on the subject. The somewhat disgruntled minutes6 of the meeting of the Chiefs of Staff Committee of the War Cabinet, on 15 February 1943, explained the position:
THE COMMITTEE had before them
a Memorandum by the British Joint Communications Board on a proposal for the installation by the Americans of a highly secret apparatus for telephone communication between the United States and London.
THE COMMITTEE had before them a Memorandum by the British Joint Communications Board on a proposal for the installation by the Americans of a highly secret apparatus for telephone communication between the United States and London.
THE COMMITTEE were informed that Major Millar, a US officer specially sent over for the purpose of installing the apparatus, had now arrived. His instructions were to place it in some building where it would be exclusively under American control, though it could be used by high officials in the British Government. There were only two other sets, one of which was being installed in the White House, and one in the War Department in Washington. No more sets could be produced for eight or nine months.
The following were the main points in the discussion:
a) Security. It was noted that the only Englishman who had yet been able to examine the apparatus was Dr Turing of the Government Code and Cypher School. In view of the fact that conversations relating to British operations would undoubtedly take place over the secret telephone, we had a legitimate interest in finding out whether the new apparatus could really be considered one hundred per cent secure. It was thought that this could best be cleared up by the Joint Staff Mission in Washington, where ample technical talent was available.
b) Site for installation.* In view of the fact that the apparatus would undoubtedly be used by the Prime Minister, and that no extension telephone involving an outside line could be permitted, it seemed that the only practicable site for the installation would be the new Government Office building, Great George Street. It was noted that the Americans hoped to have the installation complete by the 1st April.
(c) Control of the Apparatus. Although the secrecy being observed by the Americans about the apparatus, and their desire for exclusive control, might be open to criticism, it was thought better not to raise objections at this stage.
The Committee went on to tell the Joint Staff Mission in Washington to ‘approach the Americans with the object of making a thorough examination of the new secrecy apparatus, so that we could be satisfied that its security could be relied upon.’ Alan left Bell Labs for a week in Washington from 17 to 25 February, and this might well have been in connection with these negotiations. Apparently he found room for improvement, since according to a later minute of the Chiefs of Staff:7
LT. GENERAL NYE recalled that Dr Turing had not been completely satisfied as to the security of the equipment and had suggested certain alterations.
Meanwhile Alan’s work on the RCA cipher seemed to show that his method would not work. He joined in the work of the ‘cell’ on another approach to the problem. Despite the great technical secrecy, there were enough straws in the wind for his colleagues to realise that he was doing other, top-level, work. Thus it was noted that while speaking with H. Nyquist, one of the top Bell consultants working on the X-system, Alan had met William Friedman, who was the chief American cryptanalyst. It got back to the ‘cell’ that Alan was ‘the top cryptanalyst in England’. One of his colleagues there, Alex Fowler, heard this and saying, ‘Oh, you can help me,’ produced a newspaper puzzle. ‘That’s one of those Herald Tribune cryptograms,’ replied Alan, ‘I’ve never been able to do those.’ He sometimes mentioned his previous period in America, and his connection with Church, and some of the mathematicians at Bell were aware of the Turing machine. But he still found it difficult to adjust to American civilities. New acquaintances at Bell Labs complained of Alan giving no sign of recognition or greeting when he passed them in the halls; instead, he seemed to ‘look straight through them’. Alex Fowler, who was an older man of just over forty, was able to take Alan to task. He was abject, but made an explanation hinting at why he found so many aspects of life difficult. ‘You know at Cambridge,’ he said, ‘you come out in the morning and it’s redundant to keep saying hallo, hallo, hallo.’ He was too conscious of what he was doing, to slip into conventions without thinking. But he promised to do better.
There was no time for social relaxation. It was the peak of the war effort, and they were all working for up to twelve hours a day. Alex Fowler would have liked to find the time and energy to entertain Alan, but it was out of the question. Like many people, he was also afraid of boring him. Alan meanwhile was accommodated in a hotel. He told a joke about how he tried going to read in the toilet when the black-out was on, but found to his chagrin that the lights went out there too.
Greenwich Village in 1943 was, perhaps more exciting than Princeton in 1938. Alan later told a story about a man in the hotel having made a sexual approach, amazing him by its casualness. No hint of anything like this was heard in Bell Labs, though Alan once said, ‘I’ve spent a considerable portion of time in your subway. I met someone who lived in your Brooklyn who wanted me to play Go.’ Another time he said: ‘I had a dream last night. I dreamt I was walking up your Broadway carrying a flag, a Confederate flag. One of your bobbies came up to me and said, “See here! You can’t do that,” and I said, “Why not? I fought in the War between the States”.’ Alan’s curious English voice, like the X-system encoding his information by frequency rather than by amplitude, made a vivid impression on his temporary colleagues.
By the end of February Alan had gained more familiarity with the electronic equipment that was used in the laboratory. Although his work was primarily theoretical, he asked many questions about oscilloscopes and frequency analysers, such as they were using for breaking speech encipherment systems, and left them impressed with the amount of knowledge he had picked up. He also took advantage of the theoreticians at Bell Labs, for instance learning from Nyquist his theory of feedback, which was a new departure making use of the complex numbers.
But another significant interaction of his visit was the one he made every day at teatime in the cafeteria. Here he met a person who had been able to take the part of an academic, philosophical engineer, the role that Alan might have liked had the English system allowed for it. This was Claude Shannon, since 1941 working for Bell Laboratories and producing ideas of a breadth which would find scant encouragement in any British company. While Friedman was Alan’s opposite number in terms of direct responsibility for cryptanalytical work, Friedman was an older and more old-fashioned figure: a code and cipher fanatic, rather than someone who had looked at cryptology through the eyes of modern science as Alan had. In intellectual depth it was Shannon who was Alan’s opposite number, and they found a good deal in common.
People had thought about machines since the dawn of civilisation, but Computable Numbers had come up with a precise, mathematical definition of the concept of ‘machine’. People had thought equally long about communication, but here again it needed a modern mind, in this case Claude Shannon’s, to provide a precise definition of the concepts involved. These were somewhat parallel developments. Shannon had completed his first paper8 in this direction in 1940, and by 1943 his fundamental ideas were beginning to be used in Bell Laboratories, in whose mathematical department he was now employed. He was consulted about the design of the X-system, which posed some of the questions answered by his work.
The transmitter, the ionosphere, and the receiver, were a communication channel, in his terms; a channel with a limited capacity, and a channel plagued by noise. Into this channel had to be squeezed a signal. Shannon found ways to define channel capacity, noise, and signal, in terms of a precise measure of information. The problem of the communication engineer was that of encoding the signal in such a way as to make best use of the channel, and to prevent it being distorted by the noise; Shannon found new precise theorems which placed limits upon what could be achieved.
There was not only a parallel between his work and Alan Turing’s; there was a sort of reciprocity. On his side Alan, although his main strength was in the logic of machines, had dipped into the study of information. Not only was this true
in general terms of all his cryptological work, but there was an even more specific point of contact. Shannon’s measure of information was essentially the same as the Turing ‘decibans’. A ban of weight of evidence made something ten times as likely; a binary digit or bit of information made something twice as definite. There were fundamental connections between the theories, although they were not free to discuss them. Shannon only knew by implication why Alan was at Bell Labs at all.
Then on his side, Shannon had also independently thought about logical machines. From 1936 to 1938 he had been working on the differential analyser at MIT, and had designed a logical apparatus with relays in connection with a particular problem. This in turn had led him to write a paper9 in 1937 which drew the connection between the ‘switching’ operations of electromagnetic relays, and Boolean algebra – hence doing this just as Alan was designing his electric multiplier at Princeton.
Alan showed Computable Numbers to Shannon, which he read, immediately impressed. They also discussed the idea implicit in Computable Numbers, an idea of which they were independently convinced. Shannon had always been fascinated with the idea that a machine should be able to imitate the brain; he had studied neurology as well as mathematics and logic, and had seen his work on the differential analyser as a first step towards a thinking machine. They found their outlook to be the same: there was nothing sacred about the brain, and that if a machine could do as well as a brain, then it would be thinking – although neither proposed any particular way in which this might be achieved. This was a back-room Casablanca, planning an assault not on Europe, but on inner space.