Turing's Cathedral Page 16
The Bigelows lived in a hand-hewn eighteenth-century farmhouse with no electricity except for one circuit in the basement that powered a water pump. Julian surreptitiously installed an additional circuit that terminated, in his bedroom, at a single electric light. He entered the Massachusetts Institute of Technology at the age of seventeen, delivering milk in a Model T Ford to pay for his tuition and graduating with a master’s degree in electrical engineering in 1936. “When I was at MIT,” he remembers, “the electronics and radio work which had been going on was considered a rather suspect and maybe a frivolous thing. One should really be designing large generators, or at least a large arc discharge thyrotron or something.”2
Bigelow’s first job was with the Sperry Corporation, in Brooklyn, New York, building navigational gyroscopes and machinery for automatic detection of flaws in railroad tracks. Sperry later merged with office equipment manufacturer Remington Rand to become Sperry Rand, the early computer conglomerate that acquired the Eckert-Mauchly Electronic Control Company and unsuccessfully defended the ENIAC patents against Honeywell—after implementing a cross-licensing agreement with IBM. At the end of 1938, Bigelow left Sperry and joined IBM in Endicott, New York, as their first employee with the job title of electronic engineer. “At that time, IBM was a very mechanically oriented company, and the notion of electronic computing was almost repugnant,” he recalls.3
At the onset of World War II, Bigelow, an amateur aviator for most of his life, returned to MIT to retrieve his academic records and enlist in the navy as an aviation cadet. “But when I got up there,” he explains, “I had to go see my department head, and he grabbed me and said, ‘We can’t let you go, we need you. We’ve got this fellow, Norbert Wiener, going around saying he knows how to win the war singlehandedly, so to speak, with his intellectual ideas. Nobody can find out what he’s talking about, so we need you to work with him to see what there is to it.’ ”4
At the close of World War I, after leaving Oswald Veblen’s group at the Aberdeen Proving Ground, Wiener had secured a job at the Boston Herald, where his career as a reporter and features writer was short-lived. The problem, as he described it, was that “I had not learned to write with enthusiasm of a cause in which I did not believe.” After being fired from the Herald, he was hired as an instructor at MIT, his home for the next forty-five years. Wiener was “daring and uncautious, instinctive as often as logical, and utterly unsuited to meticulous step by step analytical-experimental staircase procedures,” Bigelow reported to von Neumann in 1946. “He has had sad experiences trying to work with large groups such as might be expected to carry out reliable experimental programs; he finds himself obliged to work with little funds and a few enthusiastic individual supporters.”5 Bigelow served as Wiener’s assistant from 1940 to 1943.
Wiener, whose nearsightedness had disqualified him from the infantry in World War I, decided to take on the antiaircraft fire-control problem: the most intractable targeting challenge of World War II. In 1940, German bombers were raining high explosives on Great Britain. U.S. targets might be next. The newly established National Defense Research Committee (NDRC) of the Office of Scientific Research and Development (OSRD) was fielding a wide range of proposals—with the Wiener-Bigelow collaboration being one of the longest shots. Wiener approached the problem from mathematical first principles, while Bigelow attempted to embody Wiener’s mathematics in an automatic antiaircraft fire director—dubbed the “debomber”—that was never built.
Wiener’s first suggestion, made to Vannevar Bush in September of 1940, was to circumvent the need for precision by “bursting in the air containers of liqu[e]fied ethylene or propane or acetylene gases so that an appreciable region will be filled with an explosive mixture [and] interdicted to enemy aircraft.”6 This unsportsmanlike proposal got no response from Vannevar Bush.
Wiener then approached Warren Weaver, who had been assigned responsibility for the antiaircraft effort in the United States, proposing to investigate “the design of a lead or prediction apparatus” which “anticipates where the airplane is to be after a fixed lapse of time.”7 Weaver, taking a wartime break from serving, as he described it, as the Rockefeller Foundation’s chief “philanthropoid,” awarded the requested $2,325 in December 1940, and the NDRC’s D.I.C. (Detection, Instruments, Controls) Project 5980 was launched. In 1940 an antiaircraft gunner facing high-altitude bombers had roughly 10 seconds to observe an approaching target before estimating its range, setting a timed fuse, and firing a 90-mm shell that would spend up to 20 seconds in flight. The job of the gunner was to guess where the airplane would be at the designated instant, while the job of the pilot was to guess where the shell would be at that instant—and to be somewhere else.
Wiener and Bigelow considered the observer, gun, airplane, and pilot as an integrated, probabilistic system. The odds favored the pilot: in 1940 only one out of about 2,500 antiaircraft shells scored a hit. In a preliminary report, they explained how they intended “to place the analysis of the problem of prediction upon a purely statistical basis, by determining to what extent the motion of a target is predictable on the basis of known facts and history, and to what extent the motion of the target is not predictable.”8
The predictable elements would indicate the most likely future position of the target, with the unpredictable elements determining the optimum “spread”—the degree to which the gunner would scatter fire because the exact position of the target was unknown. This distinction was equivalent to the distinction, in communications theory, between signal and noise. Similar ideas were formalized by Claude Shannon (working in consultation with Wiener) and Andrey Kolmogorov (working independently in the Soviet Union) at about the same time. “The transmission of a single fixed item of information is of no communicative value,” Wiener explained in his report to Weaver in 1942. “We must have a repertory of possible messages, and over this repertory a measure determining the probability of these messages.”9
Wiener had launched his mathematical career with a theory of Brownian motion—the random trajectory followed by a microscopic particle in response to background thermodynamic noise. He was thus prepared for the worst case possible: an aircraft that changes course at random from one moment to the next. Wiener’s theory, strengthened by Bigelow’s experience as a pilot, held that the space of possible trajectories (equivalent to the space of possible messages in communications theory) was constrained by the performance envelope of the aircraft and the physical limitations of the human being at the controls. Almost all combat flying, Bigelow observed, was composed of curves, not straight lines. Straight-line extrapolation of a flight path was a reliable predictor only of where the airplane would not be at any given future time.
Wiener was strictly a theoretician. Bigelow, “a quiet, thorough New Englander, whose only scientific vice is an excess of scientific virtue,” in Wiener’s assessment, was an engineer at home with machines. “For many years, Bigelow nursed a series of old and decrepit cars,” Wiener explained, “which, by all the canons of the motorist, should have been consigned to the junk heap years ago.” Alice Bigelow, Julian’s daughter, remembers “learning how to jump start a car while driving it, as soon as I was big enough to see, from age nine or something—because the cars were always broke. Dad would push it: ‘It’s fine, it’s fine, just a little push, and it will go.’ And in Princeton that was just unthinkable.”10
It was equally unthinkable, in Princeton, to live in anything except a conventional house. Bigelow, however, purchased a former blacksmith shop on Clay Street, in central Princeton, and in 1952 moved it to a vacant lot on Mercer Street, between the Battlefield and the Friends meetinghouse at Stony Brook. When negotiations with the Township and Borough of Princeton over the cost of moving overhead wires out of the way to allow passage of the building broke down, “he cut the thing in half like a layer cake, and then bolted it back together,” remembers Jule Charney, leader of the IAS meteorological group.11
Bigelow maintained a succession o
f small aircraft, including a Cessna he purchased damaged in Wyoming, restored to airworthiness, and flew home. A dismantled aircraft engine once occupied the Bigelows’ Princeton living room, concealed by a tablecloth when there were guests. Wiener, who had been unable to stay upright on a horse, was afraid of flying, but “was willing to take the chance and fly with me,” Bigelow recalls. “We flew from Framingham to Providence and back again. Inside the airplane there were some steel tubes which brace the windshield and his hands left fingerprints on them.”12
Asked why Wiener—who had the resources of MIT at his disposal, and whose interest in digital computing preceded von Neumann’s—did not build his own computer, Bigelow answered, “He wasn’t a man to do something that was practical. A computer has to work.”13 So did an antiaircraft director, and on October 28, 1941, Warren Weaver sent Wiener and Bigelow a list of questions centered on one uncertainty: Was Wiener’s theory going to amount to anything that could affect the outcome of the war?
On December 2, 1941, five days before the Japanese attack on Pearl Harbor, Bigelow responded with a fifty-nine-page letter to Weaver, marked “to be destroyed after reading,” reporting progress on the “debomber” so far. The goal, as Bigelow described it, was an antiaircraft director that kept the signal (the aircraft’s flight path) separate from the noise: noise introduced both by a pilot attempting to behave unpredictably, and by observation and processing errors along the way. “To re-separate the signal from the soup in these last two terms is no cinch, and in the case of random or Brownian noise with no simple spectrum it is quite impossible to do the filtering perfectly,” noted Bigelow. “Result: lost ground.”14
Bigelow compiled a list of fourteen “Maxims for Ideal Prognosticators” starting with MIP 1: “Make all observations in same coordinate system as will finally be used by the gun-pointer.” Maxims 2–4 advised separating the available information into that needed immediately and that needed later, while Maxim 5 added that “if noise is ever to be filtered from signal, it must be done at the earliest possible stage rather than after the two are tangled with other noises and signals, for the same reason that repeater stations are used on a signal line rather than filters and amplifiers at the ends.” Maxim 7 advised “Never estimate what may be accurately computed”; Maxim 8 advised “Never guess what may be estimated”; and, if a guess was absolutely necessary, “Never guess blindly” was Maxim 9.
Maxims 10 through 14 specified how to implement optimal prediction when the target “has the character of a Brownian motion impressed upon a resonator system.” Existing methods of tracking a target’s changing position “of necessity refers it to an irrelevant point of observation thus destroying its fundamental symmetry,” while an ideal predictor should assume that the target obeys the conservation laws of physics “upon which is superimposed a random modulation symmetrical in time.”15 The Wiener-Bigelow debomber would model the behavior of the airplane within the frame of reference belonging to the airplane, rather than referring it to that of the observer on the ground.
“We should clear any fog surrounding the notion of ‘prediction,’ ” Bigelow confessed. “Strictly and absolutely, no network operator—or human operator—can predict the future of a function of time.… So-called ‘leads’ evaluated by networks or any other means are actually ‘lags’ (functions of the known past) artificially reversed and added to the present value of the function.”16 Nonetheless, Bigelow’s strategy paid off. A proof-of-principle model was constructed, allowing an operator controlling a white spot of light to follow a red spot of light, driven by a modified phonograph turntable and representing an evasive target, around a darkened room. Wiener “was excited by the thought that his calculations were relevant and serviceable,” Bigelow recalls. “He showed it by puffing on his cigar in a violent way. The room would be full of smoke. He’d sort of jump up and down. He was a little bit too eager to accept the demonstration I produced as a proof that it would work.”17
Wiener “was really flying on Cloud 9,” adds Bigelow. “I did not want to put down the mathematical ideas he was talking about, but simply to realize those in a continuous and effective way in time for this war was probably wildly impossible.”18 As the chances of putting his ideas into practice diminished, Wiener pushed harder on the theoretical side. “I tried to work against time,” he explained. “More than once I computed all through the night to meet some imaginary deadline which wasn’t there. I was not fully aware of the dangers of Benzedrine, and I am afraid I used it to the serious detriment of my health.”19
On July 1, 1942, George Stibitz, chairman of the Anti-Aircraft Director Division of the NDRC, spent the day with Bigelow and Wiener, noting in his diary that “their statistical predictor accomplishes miracles.… For a one-second lead the behavior of their instrument is positively uncanny. Warren Weaver threatens to bring along a hacksaw on the next visit and cut through the legs of the table to see if they do not have some hidden wires somewhere.”20
The Wiener-Bigelow collaboration, although failing to produce a working debomber, was influential on other fronts. With neurophysiologist Arturo Rosenblueth, Bigelow and Wiener coauthored a 1943 paper, “Behavior, Purpose and Teleology,” that suggested unifying principles underlying purposeful behavior among living organisms and machines. “Teleology has been interpreted in the past to imply purpose and the vague concept of a ‘final cause,’ ” they noted, explaining that “we have restricted the connotation of teleological behavior by applying this designation only to purposeful reactions which are controlled by the difference between the state of the behaving object at any time and the final state interpreted as the purpose.” Teleology was thus identified with the Bigelow-Wiener definition of negative feedback, where “the signals from the goal are used to restrict outputs which would otherwise go beyond the goal.”21
This paper served as the namesake for the informal Teleological Society, whose inaugural meeting, hosted by von Neumann, was held at the Institute for Advanced Study on January 4–6, 1945. With the sponsorship of the Josiah Macy Jr. Foundation, a series of more formal conferences followed, and what came to be known as the Cybernetics movement took form. “Cybernetics came into its own,” explained neurophysiologist Warren McCulloch, “when Julian Bigelow pointed out the fact that it was only information concerning the outcome of the previous act that had to return.”22
In 1943, Bigelow left MIT, reassigned by Warren Weaver to the NDRC Applied Mathematics Panel’s Statistical Research Group. Under the auspices of Columbia University, eighteen mathematicians and statisticians—including Jacob Wolfowitz, Harold Hotelling, George Stigler, Abraham Wald, and the future economist Milton Friedman—tackled a wide range of wartime problems, starting with the question of “whether it would be better to have eight 50 caliber machine guns on a fighter plane or four 20 millimeter guns.”23 Bigelow was brought in to help with an automatic bomb sight being developed for high-speed dive bombers trying to hit fixed targets on the ground: the debomber problem upside down. He was promoted to associate director, and remained with the group for thirty-one months.
Back in Princeton, von Neumann was trying to get the Electronic Computer Project off the ground. Presper Eckert, who was expected to lead the engineering team, was reluctant to leave the Moore School for the uncertainties of the Institute, and sent his brother-in-law, mechanical engineer John Sims, instead. Sims was hired on January 18, 1946, and assigned to begin searching for tools, electronic components, and materials, becoming the project’s first employee. Herman H. Goldstine, awaiting release from the army, became the second employee, accepting a position (first offered on November 27, 1945) as associate director on February 25, 1946. His salary was set at $5,500—lower than an Institute professor, but higher than an Institute visitor, upsetting the distinction that had been in place since 1933.
As negotiations with Eckert stalled—grinding to a halt once Eckert and Mauchly decided to go into business for themselves—von Neumann began searching for an alternate chief engineer
. Wiener, asked for his recommendation, put Bigelow at the top of the list. “We telephoned from Princeton to New York, and Bigelow agreed to come down in his car,” Wiener recalls. “We waited till the appointed hour and no Bigelow was there. He hadn’t come an hour later. Just as we were about to give up hope, we heard the puffing of a very decrepit vehicle. It was on the last possible explosion of a cylinder that he finally turned up with a car that would have died months ago in the hands of anything but so competent an engineer.”24
Bigelow was hired on March 7, 1946, at a salary of $6,000, effective June 1, with an interim $25 per diem as a consultant until he could move to Princeton from New York. The Aydelottes offered Julian and Mary Bigelow temporary accommodation in Olden Manor, including “use of the kitchen for as many meals as Mrs. Bigelow feels equal to preparing.”25 There were several months of commuting while Julian completed his obligations to the Statistical Research Group and Mary, a psychologist, arranged to move her practice to Princeton from New York.
The Bigelows became pillars of the close-knit Institute community. Mary was a gifted therapist, and Julian was fluent not only in mathematics and physics, but in the undocumented practices that were required, in postwar New Jersey, to get anything built—or fixed. “I came to Princeton in fall 1948 with three year old Katharina, both of us quite lost in that vast new country,” Verena Haefeli, now Verena Huber-Dyson, remembers. “Everybody was friendly, something that did not just happen to you in Switzerland without the appropriate preamble of introductions. It was Mary Bigelow that managed to put me at ease, with her warm, naturally outgoing ways and her sensitive understanding of the human psyche. I remember Julian’s handsome good looks, imposing figure and especially his clear blue eyes. To me, fresh from mixed up Europe, he was the prototype of American uprightness and purposefulness.”26