When Computers Were Human (31 page)

Karl Pearson and Frances Cave-Browne-Cave had used correlation analysis to search for connections between weather patterns on the Atlantic Ocean. Wallace borrowed their methodology in order to connect fifty years of climate data with planetary data from nautical almanacs. He began with the ephemerides, punching one card for each of the 18,261 days in the fifty-year period of his study. Each card contained a code for the date, the positions of the moon and the major planets, and a blank spot for weather data, such as the daily rainfall in Des Moines, the high temperature in Cedar Rapids, or perhaps the wind direction in Ames.

Wallace had become a prominent political figure, though he had not quite reached the national stage. When his political opponents learned of his weather research, they mocked it as “weather astrology,” a form of research that properly belonged in the misguided world of Jonathan Swift's Laputa.
³⁶
The study was probably naively conceived and certainly belonged to a branch of meteorology that contributed little to the understanding of the weather, but its scientific validity is less important than the scale of the research. To compute a single correlation statistic, Wallace would have to duplicate his set of cards, punch his weather data on each of the 18,261 cards, compute the basic statistics for the correlation, and give those statistics to a human computer to finish the calculation. In this process, the final step was insignificant. The few multiplications and the single division needed to compute the correlation were dwarfed by the other activities. With machines of the 1930 era, a skilled operator would take three hours to create a duplicate set of cards, forty working days to punch the weather data onto the duplicates, and then twenty-one days to compute the basic parts of the correlation. The human computer would need a mere fifteen minutes to complete the task.

Wallace attempted to entice the Iowa State Mathematical and Statistical Service to continue the research. “This study seems to be of such unusual fundamental value that I cannot help but feel if would be a splendid thing if you people at Ames could take over these cards,” he wrote to George Snedecor. Busy with his new hybrid seed company, Wallace offered to donate his 18,261 cards to Iowa, suggesting that the database was of “unusual value.”
³⁷
Snedecor declined the offer, perhaps sensing that the project was not founded on a firm understanding of meteorology and certainly knowing that such a demanding project would have easily overwhelmed his modest staff and his relatively simple punched card office. His computing laboratory ranked among the most powerful scientific computing facilities of the age, even though it consisted of no more than seven or eight people: Mary Clem, a small staff of computers, and an operator for the tabulator. They could process far more numbers than Pearson's Galton Laboratory with its collection of Brunsviga calculators. Wallace's weather calculations, even if they were based upon valid science, would have consumed all of their time and effort.

The computers at the American Telephone and Telegraph company were veterans of the First World War in the sense that the computing division of the company began to take shape during the summer of 1918. The company had expanded at the start of the war in order to provide the army and navy with radios and telephone equipment. It had a long history of scientific research that could be traced back to the original telephone of Alexander Graham Bell (1847–1921) in the 1870s. Company scientists studied a variety of problems that were related to telephone services. Chemists studied new materials for insulating wires; physicists looked at the propagation of radio waves; and statisticians evaluated different designs for operator stations. The computing division was an offspring of the transmission section, the group that was working to develop reliable and efficient long-distance telephone lines.
³⁸

The calculations of telephone transmission imposed special demands because they utilized a form of arithmetic that involved complex numbers. Each complex number consists of two values. For historical reasons, one value is call the “real part” and the other is called the “imaginary part.” Because of these two parts, complex arithmetic requires more labor than ordinary arithmetic. The sum of two complex numbers requires two ordinary additions. A complex multiplication requires seven ordinary operations: four multiplications and three additions. The most taxing operation, complex division, requires sixteen steps: eight ordinary multiplications, six ordinary additions, and two ordinary divisions.

Though complex numbers increase the amount of calculation, they actually simplify the analysis of electrical circuits, particularly the analysis
of vacuum tubes. The vacuum tube amplifier had proven to be the key technology for long-distance transmission. American Telephone and Telegraph had built a prototype amplifier in 1912 and had demonstrated a transcontinental circuit at the 1915 San Francisco World's Fair.
³⁹
In 1916, the company hired a mathematician, Thorton Fry (1892–1991), to assist the electrical engineers with their analyses. In the first year of the war, Fry hired a computer, Clara Froelich (b. 1892). Froelich was a graduate of Barnard College, the women's school affiliated with Columbia University. From what we know about her early years, Froelich was a reserved woman, a student of mathematics who had complained about being isolated among the social circles of Barnard.
⁴⁰
She shared the computing duties with two other computers, but she proved to be the only permanent member of the computing staff. Fry preferred to hire recent graduates of women's colleges, and few of these workers stayed at American Telephone and Telegraph for more than a year or two.

27. Thornton Fry and computer at American Telephone and Telegraph

In 1922, Fry, Froelich, and the other computers were removed from the long-distance transmission division and given their own office, the division of mathematics.
⁴¹
Their office occupied one corner of a yellow brick building that sat on the Manhattan riverfront about two miles north of the New York headquarters of American Telephone and Telegraph. As was common in office buildings of the day, no partitions divided each floor to separate the work spaces. The computers could watch the staff of
electrical engineers studying the behavior of a vacuum tube or the properties of cables. They could smell creosote drifting up from the laboratories responsible for preserving the wood of telephone poles. From their desks, the computers could look south toward the skyscrapers of Wall Street or watch ships struggling up the Hudson River to the west. In 1925, the building, the computers, and the rest of the researchers were transferred to a new organization, named Bell Telephone Laboratories. Bell Laboratories was the largest of the many industrial research facilities formed in the 1920s. By the end of the decade, “more than 1,600 companies reported that they supported research laboratories,” wrote historian Robert Reich, “employing nearly 33,000 people in all.”
⁴²

The mathematics division of Bell Telephone Laboratories “does not regularly supply computing services to other departments,” wrote Fry in 1925. He explained that computation was usually “performed by special groups of calculators in the departments where their services are required.”
⁴³
Froelich and her peers worked with Fry to develop new methods for calculation, advised other researchers on computational issues, and helped instruct the computers of other departments. Like many scientists of the age, Froelich studied the operation of IBM tabulating equipment, hoping that it might be adapted to the calculations of the company's engineers. The mathematics division did not have its own machines, so she was required to spend evenings in the company's accounting office with the tabulators that were employed for business transactions by day. She “made valiant efforts to use [punched card tabulators] for more purely mathematical problems,” reported Fry, “but with little success.”
⁴⁴

Froelich had greater success with the new desk calculating machines. These machines were the direct descendants of the geared adding machines of the 1890s. In the intervening years, they had acquired electric motors and a fuller set of operations. By 1925, the computing staff could claim to be experts on two machines, one called the Millionaire calculator and the other the Mercedes. The Millionaire had a mechanical multiplication table, while the Mercedes could “perform automatic division, requiring only that the operator set up the divisor and the dividend in proper registers, but not requiring any further supervision.” All of these, she found, could be used for the two-part operations of complex arithmetic.
⁴⁵

The individual who acquired a reputation for adapting commercial business calculations “with little or no change in construction—to scientific computing” was L. J. Comrie.
⁴⁶
Though Comrie was intrigued with the idea of building special-purpose machines for scientific problems, he argued that there were many benefits to be gained from commercial machines. First, they were generally flexible. Second, they were usually quite reliable, the product “of groups of experts,” rather than a “single and
perhaps not too experienced designer.” Finally, he argued that they were “economical as compared with the overhead costs of design and construction on a small scale.”
⁴⁷

After leaving Karl Pearson and the Galton Laboratory, Comrie had moved to Cambridge and had studied astronomy. Slightly older than most students and calloused by his brief experience at the front, he did not easily fit into the society of graduate students. Rather than focusing on the narrow subjects of astronomical theory and the automatic collection of data, he looked to the larger astronomical community and attempted to make a mark. Before completing his doctorate, he had organized a computing section for the British Astronomical Association, a loose network of twenty-four volunteer computers that prepared special tables for the association.
⁴⁸

During the early 1920s, Comrie emerged as a restless and ambitious individual. Little in the scientific world met his standards. The computers at the Greenwich Observatory were the first to feel the lash of his criticism. After spending two months in the observatory computing office, Comrie concluded that the computers used inferior methods in their work. Believing that the observatory staff were not listening to his remarks, he made his complaints publicly.
⁴⁹
In the years that followed, he was often sharp and occasionally angry. While teaching at Swathmore College in the United States, he complained about his teaching duties. When he moved to Northwestern University in Illinois, a school that seemed to be more open to him, Comrie felt that he was unappreciated and underpaid. “I feel that my qualifications and experience entitle me to a better position than the one that I now hold,” he wrote to the university president.
⁵⁰
He found satisfaction only in calculation and looked for a time when he might return to London and take charge of the most sophisticated computing office in England, the computing floor of the
Nautical Almanac
. In 1925, anticipating that the almanac director would soon retire, the British Admiralty offered him a position in the almanac office.
⁵¹

At the almanac office, Comrie experimented with punched card tabulators. His initial experiments, though promising, were not as important to him as a test of a new accounting machine. This device, called the National Accounting Machine, had multiple registers, sets of gears that would store several numbers. These registers were used to keep different sums for accounts. A single action might post a number to the general ledger and accounts receivable. In these multiple registers, Comrie saw a machine that could be operated as a difference engine. It can “be called a modern Babbage machine,” he wrote, for “it does all that Babbage intended his difference engine to do and more.”
⁵²
For a century, Babbage's difference engine had been a distant but desirable goal, a machine that
promised to simplify many almanac computations. Babbage, of course, had failed to complete such a machine. The difference engine built by George and Edvard Scheutz had been sensitive, prone to failure. In a moment, Comrie had discovered a difference engine that was robust, was mass produced, and had “spare parts and expert service … readily available.” “Above all,” wrote Comrie, “others interested may purchase the machines at a moment's notice and at prices that are economical.”
⁵³

The economics of the National Accounting Machine was important to Comrie, as the British Nautical Almanac Office had little money for computing machinery and no funds for developing new computing devices. He would recall the almanac office as a place of “politics and red tape,” but this judgment was colored by his later experiences.
⁵⁴
During the 1920s, he was able to find enough funds to outfit the computing room with ordinary mechanical calculators for production work and was able to purchase a National Accounting Machine with money he acquired from the Mathematical Tables Committee of the British Association for the Advancement of Science. The British Association had been founded in the 1830s by Charles Babbage and his friends as an alternative to the Royal Society. The Mathematical Tables Committee had been formed in 1871, the year of Babbage's death. It consisted of a small group of mathematicians, numbering between six and twenty, who prepared mathematical tables. Since its founding, it had published about fifty tables in the reports of the British Association for the Advancement of Science.
⁵⁵