by Kurt W Beyer
Kahrimanian’s paper, titled “Analytical Differentiation by a Digital Computer,” described a differential compiler that represented the fulfillment of the type-B compiler that Hopper had written about in “The Education of a Computer.”48 In view of the difficulty Hopper experienced solving for differential equations on the Harvard Mark I, including von Neumann’s famous implosion problem, Kahrimanian’s differentiator seemed too good to be true. The user had only to provide a specific function in the pseudo-code and indicate the number of n derivatives, either complete or partial, to be generated. The operator routine of the differentiator would then guide the computer to produce the subsequent formulas for the nth derivative. The formulas derived by the operator routine could be immediately compiled to construct a program for numerical evaluation. Comparing the work of Koss and Kahrimanian with other papers presented at the ACM conference, and considering the nature of the discussions that followed, Hopper came to this conclusion: “It was evident that Remington Rand’s Programming Research Group had progressed considerably further in the development and application of automatic programming techniques than had any other single or combined effort in this field.”49
MAKING CONVERTS
One of the non-Remington Rand participants impressed by the A-2 and the differentiator was Carl Hammer. Hammer had first been exposed to data-processing equipment as a research associate in the T. J. Watson Scientific Computing Laboratory at Columbia University. During the fall of 1953 he was employed by the Franklin Institute to study the industrial applications of computers. Hammer successfully used the differentiator and the A-2 compiler to solve a difficult differential equation concerning roller bearings and oil viscosity. Much to his amazement, he solved the problem in a day. “I wrote twenty lines of coding for an A-2 compiler to solve the whole problem,” he recalled, “and here they’d been working for months and years sometimes on these things.”50
Hammer was so impressed, in fact, that he wrote an article for an engineering magazine describing the A-2 compiler and its revolutionary consequences for engineering and science. “I tried to point out to the people that this was a remarkable thing,” he recalled. “We could speed up industry, science, all the applications we could learn more, and all that from the power of this machine.”51 In 1955, Hammer quit his job and joined Remington Rand. From 1955 to 1957, he headed the European UNIVAC Computing Center.
Not all programmers were as enthusiastic as Hammer about the potential of automatic programming. Herb Grosch,52 a notable computer pioneer who headed up computer operations at General Electric (a UNIVAC customer in 1953), became a vocal opponent of automatic programming. Grosch and his fellow “Neanderthals” (as they came to be called as the debate intensified during the 1950s) continued to argue that programming took far too much creativity and dexterity for the human being to be replaced by the very machine he was manipulating. Furthermore, most commercial programs were task-specific or company-specific, so it made more sense for a team of experienced programmers to design and implement handcrafted codes that met the particular needs of the client.
Reflecting on the negative reactions of some of her fellow programmers, Hopper expressed the belief that arguments focusing on “efficiency” and “creativity” covered far baser motivations: “Well, you see, someone learns a skill and works hard to learn that skill, and then if you come along and say, ‘you don’t need that, here’s something else that’s better,’ they are going to be quite indignant.” In fact, Hopper felt that by the mid 1950s many programmers viewed themselves as “high priests,” for only they could communicate with such sophisticated machines. They served as the intermediaries between user and computer, and automatic programming jeopardized their exclusive position.53
Hopper was not the only one who came to this conclusion. John Backus, developer of Speedcode and later of FORTRAN, was conscious of the programming community’s reaction to his contributions: “Just as freewheeling westerners developed a chauvinistic pride in their frontiersmanship and a corresponding conservatism, so many programmers of the freewheeling 1950s began to regard themselves as members of a priesthood guarding skills and mysteries far too complex for ordinary mortals.”54 But the more the likes of Backus and Hopper preached the benefits of automatic programming, the more concerned the programming priesthood became about the spreading technology.
WINNING THE HEARTS AND SOULS OF MANAGEMENT
If convincing fellow programmers of the benefits of automatic programming was challenging, persuading senior management at Remington Rand to invest money in the new technology was nearly impossible. For many executives in the 1950s, the very notion of the computer as a business technology itself was hard to grasp, aside from the question of how these machines worked. “It was totally hopeless,” Hopper recalled, “to explain to anybody that you thought computers could put together programs because it was obvious to everyone that computers could only do arithmetic; they couldn’t write programs. And no matter how much you explained they weren’t really writing a program, they were only piecing one together, people just didn’t understand that.”55
By December 1953, Hopper felt it was critical to make management understand the importance of automatic programming Until then, Hopper and her network of programmers had worked on compiler technology during their spare time. But the success of the prototypes and the growing demand required full-time resources and personnel. On 31 December 1953, Hopper drew up a report aimed at convincing Remington Rand to invest in her concept of automatic programming. Her “plea for a budget,” as she referred to it, not only chronicled the development of compiler technology but also addressed the commercial potential and implications of automatic programming in far more depth than any of her writing up to that point.56
Grace Hopper. Courtesy of Library of Congress.
Hopper’s report highlights a growing reality that few in or out of the computer industry comprehended as of 1953: the cost of programming, operating, and maintaining computers was increasing and would continue to do so at an alarming rate. These rising costs resulted from a variety of factors. First and foremost, the accepted method for constructing programs was labor-intensive. Creating a basic commercial application such as a payroll routine required a team of mathematically trained programmers. By 1953 they were in short supply. Those with experience commanded high salaries, and those without experience required expensive and time-consuming training. As computer hardware proliferated, the shortage of skilled labor would only intensify.
Worse, Hopper reminded her audience of executives, there was no such thing as a static commercial program, for changing business conditions required continuous modifications. “A change in government regulations, city ordinances, union contracts, or requirements of company management, even a simple change, such as income tax exemptions, may invalidate not only parts of runs, but whole runs in such a base problem,” she wrote.57 The problem was exacerbated by the fact that most commercial program rewrites had to be made on extremely short notice, thus requiring companies that operated computers to maintain permanent teams of programmers. The way out of this costly conundrum, according to Hopper, was through the implementation of compiler techniques. A pseudo-code for payroll could be updated quickly, and the run-program could be recompiled at a moment’s notice. The changes were checked automatically and human error was eliminated, thus ensuring the accuracy of the computed results. Hopper concluded that the reduction in staff would result in a significant reduction in costs.
The most striking and innovative aspect of Hopper’s report, however, concerned a radically new way to think about executive decision making:
In any business or industry a large quantity of information exists on paper. This information in random, unprocessed form is of little or no value to management as a basis for decision making. When information is organized and processed, it becomes “intelligence.” Effective management, then, relies on the ability to process information to be used in decision making.58
Not
only did the information have to be processed accurately; it had to be processed quickly. Hopper noted that there existed a whole set of statistical studies and tabulations—known to be useful to management—that were not typically made. Even with the aid of a computer and a programming staff, this set of problems could not be calculated rapidly enough to supply the needed intelligence before a decision had to be made. Compiling routines, combined with the speed of electronic circuits, opened unexplored intellectual territory not only for scientists and engineers but also for business leaders. Automatic programming could precipitate a revolution in decision making.59
“A monumental effort,” Hopper told Remington Rand management, “is being expended by IBM, RCA, Raytheon, CPC (National Cash), Consolidated Engineering, Burroughs and all their users and prospective users, and by their affiliated university installations, to develop similar [automated programming] techniques for use with their digital computers.” Though she did not reveal her sources, Hopper said: “Through personal contacts, it has been determined that our closest competitors are making an all-out effort to approach the level that Remington Rand has attained and to surpass us.”
In her final call to arms, Hopper demanded that “personnel be devoted 100% of their time to the development of the compiler techniques and particularly to the development of the A-3 compiler.”60 Risking her position at Remington Rand, Hopper was ultimately successful in creating a separate department to further explore automatic programming. In a letter to the Army Map Service dated 10 February 1954, the new Director of Automatic Programming thanked her distributed development team and shared with them her success in institutionalizing automatic programming at Remington Rand:
The fact that there is today some hope of building an automatic programming group is due to your efforts on A-2. I cannot tell you how grateful I am. If I had not been able to report the success of your efforts, automatic programming would have been a very dead duck in Remington Rand.61
On the eve of 1954, Hopper believed Remington Rand had the ability to seize and maintain the lion’s share of the computer industry, a market that appeared to be on the verge of rapid expansion. Remington Rand held a substantial lead in both hardware and software technology. But its closest competitor, the International Business Machines Corporation, was demonstrating a growing interest in computer technology.
9 IBM ANSWERS REMINGTON RAND’S CHALLENGE
In the summer of 1949, when Grace Hopper joined the Eckert-Mauchly Computer Corporation, the International Business Machines Corporation controlled 90 percent of the market for mechanical calculators and punch-card machines, with annual revenues of $300 million. Apart from experimental work on the Harvard Mark I and the Selective Sequence Electronic Calculator, IBM leadership firmly believed that the company’s future growth would be driven by mechanical punch-card technology. But within 10 years IBM was the undisputed leader in electronic computers, with more than 70 percent of the global market. IBM’s dominant position in the marketplace helped to drive annual revenues to $1.8 billion by 1960.1
How IBM came to control the early computer industry has been thoroughly explored by economists and historians. What is more important here is how IBM’s success, as well as Remington Rand’s loss of market share, affected the direction of programming’s development during the 1950s. In particular, how did Remington Rand react to the IBM challenge, and in what ways did this reaction influence Grace Hopper and her work?2
IBM: HARD-WORKING UNDERDOG, OR RUTHLESS MONOPOLIST?
On 17 January 1969, the U.S. Department of Justice filed an antitrust lawsuit against International Business Machines. The IBM lawsuit became the longest and most arduous court case in U.S. history, coming to a sudden end in 1982 when the Reagan administration withdrew the antitrust charges. In hindsight the case appeared to have wasted both IBM’s and the government’s money and manpower, but for business historians the trial offered a documented history of the computer industry. Secondary literature published in the 1980s relied heavily on the thousands of papers made public and the hundreds of testimonies generated during the 13-year trial. These first attempts to capture the industry’s early history tended to follow closely the opposing positions outlined during the antitrust trial. Authors aligned with the Justice Department’s perspective depicted IBM as a ruthless monopolist bent on controlling the computer industry. Cash flow generated by the company’s punch-card monopoly permitted it to underprice the 700 series and other early computers. According to the economist-turned-historian Richard Thomas DeLamarter, IBM’s pricing policies hurt Remington Rand’s bottom line while expanding IBM’s market share: “Remington saw its share of the expanding market plunge irretrievably during the next few years as customers flocked to the IBM 701 and 702.”3 The economists Franklin Fisher, James McKie, and Richard Mancke, on the other hand, asserted that IBM was just one of many companies that had the potential to enter the new industry during the 1950s. The knowledge needed to build computers was widely dispersed, thus creating a level playing field on which IBM competed. According to Fisher, McKie, and Mancke, IBM became dominant because the company’s leadership consciously committed more resources and personnel to the new industry. This risky investment translated into better products and better service, and ultimately IBM was rewarded with market leadership.4
REMINGTON RAND’S MARKETING COUP
Neither DeLamarter’s account nor that of Fisher et al. pays much attention to developments outside IBM. Interviews with Grace Hopper and with 30 other people associated with UNIVAC during the 1950s suggest that IBM did not win the computer market so much as Remington Rand lost it. The difficulties of merging Remington Rand with EMCC and ERA during the first part of the 1950s kept Remington Rand from capitalizing on its technological lead. As Remington Rand stumbled, IBM’s far more sophisticated sales and customer support system quickly came to dominate the fertile market. This is a tidy summary, but further inspection reveals a far more complicated story.
In November 1952, on the night of the Eisenhower-Stevenson presidential election, Remington Rand seized the imagination of the country and positioned itself to become, in the eyes of the American public, the undisputed leader in computer technology. A mock-up of a UNIVAC machine had been installed in the headquarters of the CBS television network, next to the desk of CBS’s anchorman, Walter Cronkite. In the days leading up to the election, Hopper’s programming team had entered demographic and voting data from previous elections into a working UNIVAC. On election night, these data were associated with real-time voting results. With only 5 percent of the votes tallied, the UNIVAC predicted an Eisenhower landslide and a final electoral-vote count of 438 to 93. Because UNIVAC’s prediction differed so markedly from the findings of the Gallup and Roper opinion polls, executives at CBS decided to inform Americans that UNIVAC predicted a close race, with Eisenhower slightly in the lead. In the end, Eisenhower defeated Stevenson soundly.5
The uncanny accuracy of UNIVAC’s prediction during a major televised event sent shock waves through the nation. At the conclusion of the election coverage, a Remington Rand spokesman appeared on CBS and informed viewers that UNIVAC’s original prediction had been suppressed. As Thomas Watson Jr. recalled, “millions of people were introduced to the UNIVAC by Edward R. Murrow, Eric Sevareid, and Walter Cronkite, who called it ‘that marvelous electronic brain.’ ”6 In the months that followed, “UNIVAC” gradually became the generic term for a computer.
J. Presper Eckert Jr. (center) discussing UNIVAC output with Walter Cronkite. U.S. Census Bureau.
UNIVAC’s election-night heroics had significant ripple effects. Morale was boosted within the Eckert-Mauchly division, and Remington Rand management began to take the machine more seriously. Furthermore, the demand for UNIVACs swelled considerably in the months after the election, despite Remington Rand’s lack of investment in sales and marketing. More and more, members of Hopper’s team found themselves fielding sales inquiries from potential customers rather than programming. In fa
ct, one of the programmers (George Danehower) later recalled that the vast majority of UNIVAC contracts were generated this way: “The machines were not really sold. Certain companies had certain kinds of problems . . . they heard about the computers, so they made inquiries.”7 The ultimate sale was actually made through the process of responding to inquiries. In 1953 and 1954, Remington Rand received more than 40 UNIVAC orders.8
Two of the more significant commercial orders of this period came from the General Electric Company and from the Pacific Mutual Life Insurance Company. Both companies conducted feasibility studies and concluded that the UNIVAC was superior to IBM’s comparable machine, the 701/702. Pacific Mutual’s reasons were highlighted in a paper given by Wesley Bagby, the company’s controller, to the Controller’s Institute of America in November 1955:
There were three items of difference . . . which influenced our final decision. Remington Rand’s UNIVAC could be leased or purchased outright. IBM’s 702, which they were competing against, could only be rented. Second, UNIVAC, one of the specific models we were considering, had been thoroughly field-tested. The other, IBM 702, had not yet been used by any customer. Third, it appeared to us that UNIVAC had some technological advantages.9
According to Ken Garrison, a member of Pacific Mutual’s evaluation team, those technological advantages centered on the reliability of UNIVAC’s mercury delay memory and the machine’s ability to buffer input and output, which permitted offline tape processing and printing.10