In early 1948, Max Newman invited Turing to Manchester to work on the computer project there. Since Williams and Kilburn knew nothing about computers and nothing about Colossus, Newman and Turing had to communicate to them what a computer might do and how it might work without describing what they had accomplished at Bletchley Park. But the two engineers were too far along in the project to allow for much input from the two mathematicians—Newman and Turing were interested in theory, but the engineers were more intent upon producing a workable memory system. As with ENIAC and Colossus, time pressures were pushing the project forward in a way that didn’t allow for what Williams and Kilburn considered to be untested ideas, though the pressure this time came not from war, but from the fact that the British government already had a contract with a local weapons and electronics manufacturer to produce the machines once the prototype was built. And Tommy Flowers was having difficulties, too: even though he had invented Colossus, he could not get a computer job, and even though he had done a successful experiment with electronic telephone exchanges in 1939, he made no headway on that front, either.
In the spring of 1949, Atanasoff was invited by General Jacob Devers to leave the Naval Ordnance Lab and move to Fort Monroe, Virginia, as chief scientist for the Army Field Forces. Devers was a West Point contemporary of George Patton who, as an army administrator between the two world wars, had upgraded and reconceived the Field Artillery, then, as an administrator in London, had organized and trained many D-day divisions. His own Sixth Army Group had landed at Marseilles, and according to David P. Colley in the New York Times:
The Sixth Army Group reached the Rhine at Strasbourg, France, on Nov. 24 … His force, made up of the United States Seventh and French First Armies, 350,000 men, had landed Aug. 15 near Marseilles—an invasion largely overlooked by history but regarded at the time as “the second D-Day”—and advanced through southern France to Strasbourg. No other Allied army had yet reached the Rhine, not even hard-charging George Patton’s.
Atanasoff was eager to work with Devers, but the general, now sixty-two, retired at the end of September that year. Atanasoff’s new boss was General Mark Clark, who had run the Italian campaign. Clark had a reputation for being difficult and egocentric. One history relates that during the war, he had a rule that “every [press] release was to mention Clark at least three times on the front page and at least once on all other pages—and the General also demanded that photographs be taken of him only from his left side.” Clark killed several of Atanasoff’s projects, and in 1950 Atanasoff returned to the navy to run a program overseeing the development of artillery detonators. Also in 1949, Atanasoff and Lura were divorced, and Atanasoff married Alice Crosby, from Webster City, Iowa, whom he had met through her job in the publications department at the Naval Ordnance Lab.
By mid-1950, Atanasoff felt that his career with the military had reached a dead end, and he was disheartened, too, by the idea that all of his enterprise and inventiveness had gone into making weapons.
In the summer of 1949, Turing was interviewed by a newspaper in relation to a dispute between two other men about machine intelligence and the possibility of a machine having a sensibility. The two men were Norbert Wiener, who had just published Cybernetics, and a neurosurgeon, Geoffrey Jefferson, who gave a speech that attempted to debunk any ideas that a machine could have emotions or self-consciousness and could, therefore, be said to think in a human way (Jefferson was a pioneer of the frontal lobotomy). When Turing was interviewed by the Times (London), he declared that “the university [of Manchester] was really interested in the investigation of the possibilities of machines for their own sake.” This was an inflammatory statement on a sensitive topic, especially in light of the scarcity of government funding for research projects. Max Newman had to write to the Times and reassure readers that the Manchester computer then being developed was intended to have practical applications and was, therefore, both worth building and not intended to usurp human beings.
But Turing was not deflected by the outcry. For the next year, he discussed and pondered the question of thinking—how, indeed, could a machine be said to be “thinking”? How could a human interacting with a machine without knowing it detect whether he was interacting with a machine or with another human? The result was a paper, published in October 1950, entitled “Computing Machinery and Intelligence.” Turing proposed a thought experiment, a situation in which an investigator would question a man (A) and a woman (B) in order to determine which was the man and which was the woman. The man would be told to obstruct the investigator, and the woman would be instructed to help the investigator. They would supply their answers in written form. Once the reader has considered this situation, he is then asked to consider the same situation, but the man has been replaced by a machine. In this situation, Turing asks, will the investigator be able to solve the puzzle correctly more or less often if A is a machine or a man? In other words, Turing proposed, if a machine can imitate a man answering questions well enough so that there is no difference in the ability of the investigator to pass a given test, then the machine may be said to be thinking. Turing extrapolated from this game to a future date when computers would have sufficient memory storage so as to be able to appear to make decisions and best guesses—at that point, he thought, what they would be doing would be called thinking. More important than answering the question of whether machines might think, though, was the posing of the question. The job of science, Turing felt, was to conjecture, to not be shy about being “heretical.”
It was Max Newman who was deflected—for him, the media brouhaha was the beginning of his retreat from computers. According to his son, William Newman, he soon went back to mathematics and focused on his old love, topology. In later years, Max ascribed this withdrawal to the dominance of the engineers, but in addition to that and the public outcry, his son also suspected “that his decision was influenced by his opposition to using the Manchester computer in the development of nuclear weapons.” Given his connections to von Neumann, his suspicions were certainly well grounded because von Neumann, of course, was even more involved in the development of the hydrogen bomb than he had been in the development of the atom bomb. He firmly believed that the West had to stay ahead of the Soviet Union, remarking that “with the Russians, it is not a question of whether, but when.” According to Norman Macrae, he felt that “all those sitting around the Soviet decision-making tables should know that in the first few minutes of a nuclear war, a bomb would arrive where they were and personally kill all of them.”
1. McCartney says six.
2. From the Clay Mathematics Institute website: “If we stretch a rubber band around the surface of an apple, then we can shrink it down to a point by moving it slowly, without tearing it and without allowing it to leave the surface. On the other hand, if we imagine that the same rubber band has somehow been stretched in the appropriate direction around a doughnut, then there is no way of shrinking it to a point without breaking either the rubber band or the doughnut. We say the surface of the apple is ‘simply connected,’ but that the surface of the doughnut is not. Poincaré, almost a hundred years ago, knew that a two dimensional sphere is essentially characterized by this property of simple connectivity, and asked the corresponding question for the three dimensional sphere (the set of points in four dimensional space at unit distance from the origin). This question turned out to be extraordinarily difficult, and mathematicians have been struggling with it ever since.”
Chapter Nine
By the spring of 1947, Mauchly and Eckert had not yet filed the ENIAC patents, which their original agreement with the University of Pennsylvania had given them rights to. That April, they met with von Neumann, Goldstine, Dean Prender of the Moore School, and Irven Travis, the man who had first declared the new, more restrictive patent policy. Ostensibly, the meeting was to discuss potential EDVAC patents; von Neumann brought a lawyer with him. It was at this meeting that the university and Mauchly and Eckert learned for the
first time that von Neumann, according to Scott McCartney, “had met with the Pentagon legal department about the patent situation, and had filed an Army War Patent Form himself” on the basis of the “First Draft” document Goldstine had typed up in June 1945. The fact that hundreds of people had read the document constituted publication, as far as the army was concerned, and so the ideas in the document could not be patented. McCartney maintains that this argument on the part of the army was a surprise to von Neumann and Goldstine as well as to Mauchly and Eckert, but, given von Neumann’s connections and his habit of being “five blocks” ahead of the competition, it seems unlikely that his lawyer would not have informed him of this possibility before the meeting. The meeting served to spur Mauchly and Eckert’s own patenting efforts, and they filed their paperwork at the end of June 1947. According to McCartney, “The application was broad and unfocused and it attempted to make more than one hundred claims covering the computing waterfront.” Crucially for the future of computing, Eckert and Mauchly assigned the patent rights they claimed not to themselves, personally, but to their company, in order to lure potential investors and contracts.
Mauchly’s job was to manage the company and to find financing and contracts. Eckert’s was to oversee the building of their first machine, now dubbed UNIVAC (for UNIVersal Automatic Computer). By December 1947, the company had thirty-six employees, including several engineers and other technicians who had followed Mauchly and Eckert (or had been lured by them) out of the Moore School. Another was Grace Murray Hopper, who had worked for Aiken at MIT and later developed COBOL, the first data processing language that worked like English. Company culture was energetic and exciting—Eckert was an inventive dynamo who showed up late every morning, sometimes six or seven days a week, and worked until late in the evening. But without Goldstine’s discipline, Eckert’s ideas were not focused on building his machine in a progressive and productive manner—he tinkered with every part and redid everyone’s designs. And he did not care for disagreement. McCartney characterizes the engineering side of the company as a “dictatorship,” but it was a chaotic dictatorship, which turned out to be a bad form of organization, since the contracts Mauchly was procuring were fixed-price contracts, as if the products were ready, although they were only in development. Even though working on UNIVAC was exciting, cost overruns meant that contracts could not be fulfilled in a timely manner, and new projects had to be added in order to pay for old projects. Eventually, UNIVAC cost $900,000 to develop, though the contracts were worth only $270,000. Eckert and Mauchly were incapable of being frugal, and nothing in their experience at the Moore School had trained them to attempt such a thing. They were accustomed to both the stimulation and the chaos that large teams of inventors generated, but having always been administered, they did not themselves know how to administer. The number of employees crept upward, and at one point engineers were encouraged to purchase stock in the company for $5,000 just to keep the company afloat.
In the meantime, von Neumann took Goldstine and Arthur Burks to Princeton to work on a computer for the Institute for Advanced Study (though Burks left within a few months for a teaching job at the University of Michigan). In the book Colossus by Jack Copeland, photograph 50 is a picture of John von Neumann, standing beside the Princeton IAS computer. The picture is undated, but the IAS computer began to operate in the summer of 1951 and was officially operational on June 10, 1952. Along the bottom of the wall of hardware runs a row of shiny metal cylinders, their ends pointing upward at about a forty-five-degree angle (fifteen are visible in the photo). These cylinders are Williams tubes, and they constituted the memory of the IAS computer.
At this point, von Neumann had been organizing his computer project for at least seven years. Back in the summer of 1946, when Atanasoff was told that the navy computer project was off, he was not told why, but part of the reason was that in late 1945, the very well connected John von Neumann had entertained letters of interest from the University of Chicago and MIT, with further feelers from Harvard and Columbia. Von Neumann was drawn to Princeton even though, as the letter from Norbert Wiener of MIT (soon to get in trouble with Dr. Jefferson) predicted, the problem that would plague the development of the IAS computer was that at “the Princestitute [the Institute for Advanced Studies] … you are going to run into a situation where you will need a lab at your fingertips, and labs don’t grow in ivory towers.” Von Neumann got something that he considered more important from the Institute for Advance Study—$100,000 for development (equivalent to $1 million today), with another $200,000 readily available. Even $300,000 would not be enough, though, so von Neumann approached both the army and the navy. Something that von Neumann understood (and that, of course, Atanasoff had also understood) was the computing difficulties of solving nonlinear partial differential equations. But if Atanasoff, writing his dissertation on the dielectric constant of helium in 1930, was forced to grapple with the vast tedium of his equations, von Neumann, overseeing the mathematical side of the Manhattan Project, understood the difficulty even more sharply because he had a greater experience with what the military wanted to do with such equations. Though the equations he had worked out for the detonation of Fat Man and Little Boy were done to the best of the Manhattan Project’s mathematical ability, they did not prove as predictive as the army and air force had hoped they would. And von Neumann was also interested in the applicability of such equations to weather patterns and forecasting.
And so, in late 1945 and into 1946, von Neumann wooed both the army and the navy—to the navy, he promised analysis of explosions in water, weather prediction, and even weather control. According to Norman Macrae, von Neumann did not hesitate to threaten the navy with the idea of Josef Stalin using computer-driven weather control to launch a new ice age in North America (though there was no reason to believe that the Soviets were developing a computer and nothing of the sort has since come to light). The army and the navy both kicked in funds for von Neumann’s computer, and the navy ended Atanasoff’s computer project. To his credit, though, von Neumann understood that the army and the navy had to agree to the same terms in their contracts, so that the project would not be subject to cost cutting by one branch or the other, and he also insisted that the intellectual property that might come out of the project would neither be made top secret nor be patented, thereby ensuring that other projects could also emerge from the IAS project. He seems to have understood all along the implications of the fact that he would be building upon ENIAC, upon the “First Draft,” and upon EDVAC, that he would be recruiting to Princeton at least Goldstine and Burks, and that he would make use of his connections with Manchester through Max Newman, and through him to F. C. Williams and Thomas Kilburn. It is quite possible that he understood the relationship between Atanasoff’s ideas and what he intended to do, but there is no evidence for it one way or another, other than the fact that he did have conversations with Atanasoff at the NOL.
At Princeton, von Neumann, Goldstine, and, to some extent, Arthur Burks wrote the papers that codified and described the ideas about computer memory that von Neumann had introduced in the “First Draft.” According to Macrae, von Neumann described the ideas, Goldstine and Burks wrote them up, and von Neumann then rewrote them. The final draft was up to Goldstine, but it carried von Neumann’s name.
Von Neumann wanted Eckert as his engineer for the Princeton project. Eckert turned him down, according to McCartney, because he remained loyal to Mauchly and, according to Macrae, because he wanted to patent his inventions and profit from them. But Eckert and von Neumann also had a history of conflict, which might have played a part in Eckert’s decision. Von Neumann did not approach Atanasoff, although it’s hard to avoid the thought that his conversation with Atanasoff at the NOL constituted something of a job interview. Atanasoff found von Neumann congenial—but then, so did almost everyone else. At any rate, the team von Neumann set up did not include Atanasoff. Kirwan Cox maintains that Atanasoff was known at Iowa State for being
abrupt and hard to get along with—he had a disconcerting habit of turning away in the middle of conversations: “People thought he was walking away in anger, but he was just finished with the conversation in his own mind. He was tough on people.” It may be that von Neumann recognized that Atanasoff was not a team player and that in any project Atanasoff might be involved in, he would insist on calling the shots.
The memory system Eckert was developing was, in the eyes of John von Neumann, one of UNIVAC’s main drawbacks. This system, called a mercury delay line, owed something to Eckert’s radar experience. The UNIVAC mercury delay line required an array of horizontal cylinders filled with liquid mercury through which electrical impulses could travel rather slowly. The memory worked by recycling the electrical impulses through the mercury over and over, using quartz transducers.1 Mercury delay line memories had an advantage in that the acoustic conductivity of quartz and mercury were about the same, but they also had serious drawbacks—the architecture of each cylinder was very particular and they were easy to damage. The word “unwieldy” doesn’t even begin to describe a mercury delay line memory—for UNIVAC, the memory required its own room, in which stood seven memory units, each composed of eighteen columns of mercury. This room could store 15,120 bits of memory (equivalent to 1,890 bytes, or not quite 2 kilobytes, although bytes and bits of memory were not standardized at the time—in the UNIVAC I, a byte was 7 bits, not 8). Added to that was the weight and the toxicity of mercury, which in itself limited the general usefulness of the UNIVAC, as well as its potential commercial appeal. And the UNIVAC was a decimal machine, making it even more unwieldy.