UNIVAC Becomes a Household Word
A reading from Harry Wulforst's 1982 bookBREAKTHROUGH TO THE COMPUTER AGE
Scanned by jds 1991 on Logitech 4" hand held scanner
then OCRed with bundled software
all at a cost of about $400.
The red light on the television camera clicked on. Charles Collingwood, seated in a CBS studio in New York City, smiled confidently at his unseen audience across America as he revealed the network's plans for broadcast coverage of the 1952 presidential election: "Here at CBS we are making elaborate preparations to bring you the results of the night of November 4th, just as quickly and as accurately as is humanly possible. Matter of fact, we want to bring them to you faster and more accurately than is humanly possible, so we have enlisted the aid of Remington Rand's UNIVAC.... If UNIVAC behaves the way we think it will, we'll all know the winner long before the final votes are counted."
During the weeks preceding Collingwood s announcement, political analysts, statisticians, and computer programmers had huddled in meetings in Philadelphia conjuring up ways to endow UNIVAC with sufficient expertise to predict the winner of the Eisenhower‑Stevenson race after only a small portion of the votes had been tallied. Months earlier, when UNIVAC was first brought to the attention of CBS, the experts at Remington Rand had blithely assumed that programming the system for analyzing voting patterns across America would be a routine undertaking. However, as the weeks rolled by and the election night deadline loomed ever closer, the experts thought that they had bitten off more than they could chew.
It had started as a simple business barter arrangement. In April 1952 a representative of the network had come to Remington Rand with a proposition. The office equipment manufacturer would be given nationwide television exposure at no cost in return for the temporary use of a hundred or so typewriters and adding machines. The quid pro quo was to be part of the programming: As television cameras panned around the huge studio (which would be filled with workers diligently tapping the keys of adding machines and typewriters), a camera would zoom in over someone's shoulder from time to time to focus briefly on the Remington Rand logotype emblazoned above the keyboard. The deal was sealed with a handshake, but before the CBS man reached the door, a suggestion by a Remington Rand publicist stopped him in his tracks.
Sustaining viewer interest during the tedious hours of reporting the vote, precinct by precinct and state by state, was a tough challenge for broadcasters. Break the monotony, said the publicist, by predicting the winner with an electronic computer and viewers will stay with you all night to see if the computer is right or wrong. This extra bit of show biz, which might possibly add some life to a slow-moving story, was enthusiastically endorsed by CBS management, but professional newscasters did not know whether to take the matter seriously or not. Their general uneasiness was evident in a response by Walter Cronkite (then chief Washington correspondent for CBS) to a question by Dorothy Fuldheim during an evening news broadcast by WEWS‑TV in Cleveland, Ohio:
DOROTHY FULDHEIM: Tell me, Walter, what are you going to do to report this very historic election?
WALTER CRONKITE: Well, this year we've got the same basic formula that we had before, which is, of course straight reporting of how the returns are coming in. However, we do have a little gimmickry this year which I think is most interesting, and may turn our to be something more than gimmickry. We're using an electronic brain which a division of Remington Rand has in Philadelphia.
DOROTHY FULDHEIM: What does it do?
WALTER CRONKITE: It's going to predict the outcome of the election, hour by hour, based on returns at the same time periods on the election nights in 1944 and 1948. Scientists, whom we used to call long hairs, have been working on correlating the facts [for these predictions] for the past two or three months. . . . Actually, we're not depending too much on this machine. It may just a sideshow... and then again it may turn out be of great value to some people.
At the time Cronkite was telling Dorothy Fuldheim that UNIVAC might be of great value, the long hairs in Philadelphia were fearful that its widely publicized debut on television would be a flop. Before UNIVAC could be programmed to rapidly analyze the election night returns, mathematical equations had to be formulated describing trends and voting patterns for thousands of political sub‑divisions throughout the United States. This massive undertaking was rendered more difficult by the fact that there was virtually no precedent for an analysis of this magnitude. The statisticians, labouring over reams of data covering the two previous presidential elections, had no existing body of expertise to guide them.
Finally, a workable method for processing the data gradually took shape. Now the experts could direct their attention to the practical business of getting ready to run the problem when the returns started coming in. Because the only available UNIVAC systems were in Remington Rand's factory in Philadelphia, a special Teletype line was reserved for transmitting the vote counts from CBS election night headquarters in New York. Three UNIVAC systems figured in the plan: One to actually process the data and be seen on television; a second UNIVAC, behind the scenes to carefully check the output of the first; and a third, on standby, in the event of an emergency.
called out on a printer. Then the latest total for that precinct was located on the original Teletype and the correct count re-entered into the system, which produced a validated listing on a fourth reel of magnetic tape.
While the most recent tallies were being transcribed on the fourth tape, all of the precinct totals were sorted into a predetermined sequence. At the same time, the data were subjected to three additional tests. First, the number of districts reporting was checked against the total number of districts in the area under study. Second, the reported vote on any given pass had to be at least as high as the previously reported total. And finally, the major party votes in each district were compared with similar data compiled for that constituency in 1944 and 1948.
The vote totals were then in the proper format for entry into the primary UNIVAC system. For this phase, composite analyses of returns in the 1944 and 1948 elections had been stored in the computer's memory in addition to a comprehensive history of state‑by‑state voting trends dating back to 1928. From this information district profiles were charted that defined the relative strengths of the major party registrations and the so‑called independent vote.
The district totals fed into UNIVAC's processor from the fourth tape were compared to the historical records of those districts in the computer's memory. Then a final vote probability was determined for each locality. These voting patterns became the basis for a general preliminary estimate of the total national vote for each candidate. Adjustments were then factored in at prescribed intervals to correct any discrepancies that arose between the actual vote counted for a district and the voting pattern selected for that district. In this way as the evening wore on and larger percentages of the total vote cast could be fed into the system, each subsequent projection could be based more on hard facts and
less on assumptions.
By 9:00 p.m., with early returns streaming in from the eastern and central time zones, the huge CBS election night headquarters in New York City was buzzing with activity. Telephones jangled. Teletype machines clacked noisily. Scribbled figures on scraps of paper were passed hastily to toteboard operators. Then the director in the control room, scanning an array of monitors, barked an order. Instantly the face of Charles Collingwood flashed on screens in living rooms across the nation as the comforting voice of Walter Cronkite told viewers what was going on.
WALTER CRONKITE: And now to find out what perhaps this all means, at least in the electronic age, let's turn to that electronic brain, UNIVAC, with a report from Charles Collingwood.
COLLINGWOOD: UNIVAC, our fabulous mathematical brain, is down in Philadelphia mulling over the returns that we've sent it so far. A few minutes ago, I asked him what his prediction was, and he sent me back a very caustic answer. He said that if we continue to be so late in sending him results, it's going to take him a few minutes to find out just what the prediction is going to be. So he's not ready yet with the predictions but we're going to go
to him in just a little while.
As Collingwood was telling his audience that UNIVAC was not ready for a prediction, it had in fact already made one. The business about needing more time was a cover‑up. Unknown to Collingwood, the folks in Philadelphia had fabricated that story to save face.
A few minutes earlier, with only three million votes counted, an astounding forecast rolled off the electric typewriter that functioned as the computer's printer. UNIVAC gave 43 states and 438 electoral votes to Dwight D. Eisenhower. Adlai E. Stevenson would capture only 5 states and 93 electoral votes. The odds for victory by the Republican candidate were predicted by UNIVAC as 100 to one or better in his favour ‑ an Eisenhower landslide.
Computer programmers huddled around the printer in shocked silence. Throughout the campaign, pollsters and political analysts had been predicting a close election that would not be decided until the wee hours of the next morning. Yet with only 7 percent of the vote counted, UNIVAC had gone way out on a limb. Too far, it seemed to those watching the state‑by‑state breakdown emerge from the printer. What was this? Several southern states going Republican. That hadn't happened in seventy‑two years! Something was wrong.
A murmur in the crowd: There must be a glitch in the program. Then a mad scramble. Everyone, grabbing code books and programming records, frantically flicked through reams of data, hoping, by some miracle, that any error that had escaped notice would now surface and be recognized. But after several minutes of fruitless page turning, punctuated by answering telephoned appeals from New York to get UNlVAC's act together, the search was called off.
Arthur F. Draper, Remington Rand's director of advanced research and the man in charge of election night operations in Philadelphia, agreed with his advisers that something drastic had to be done to bring UNIVAC back to its senses. They decided to go right to the heart of the matter and arbitrarily change the factor‑so carefully fine tuned through months of preparation‑that extrapolated the number of returns actually received into estimated final totals for each state.
Fortunately, this was a simple procedure. One merely had to run the program to the breakpoint where the critical factor was computed, stop the run, type in a new figure from the supervisory control desk, and resume processing. Within two minutes, a new set of totals began rolling off the printer. A chastened UNIVAC reported 18 states and 317 electoral votes for Eisenhower. Much better, but not good enough for the thoroughly shaken crew in Philadelphia.
Still hedging, they tweaked the formula again. This time UNIVAC called the election a toss‑up. It gave twenty‑four states to each candidate with Eisenhower leading in electoral votes by a scanty margin of 270 to 161. Breathing easier, and wiping perspiration from foreheads, the computer people in Philadelphia released these figures to CBS, which broadcast them on the network at 10:00 p.m.
By 11:00 p.m., however, Eisenhower votes were rolling in like a tidal wave, and UNIVAC, shrugging off the dampening influence of the twice‑revised formula, swung back again to the original prediction of 100 to one odds in favour of the general. At midnight, in his recap of the evening's coverage, Collingwood asked Draper what went wrong:
COLLINGWOOD: An hour or so ago, UNIVAC suffered a momentary aberration. He gave us the odds on Eisenhower as only eight to seven . . . but came up later with the prediction that the odds were beyond counting, above 100 to one, in favour of Eisenhower's election. Let's go down to Philadelphia and see whether we can get an explanation of what happened from Mr. Arthur Draper. Art, what happened there when we came out with that
DRAPE.R: Well, we had a lot of troubles tonight. Strangely enough, they were all human and not the machine. When UNIVAC made it's first prediction, we just didn't believe it. So we asked UNIVAC to forget a lot of the trend information [concerning previous elections], assuming that it was wrong.... [but] as more votes came in, the odds came back, and it is now evident that we should have had nerve enough to believe the machine in the first place.
With less than 7 percent of the vote tallied at 9:00 p.m on election night, UNIVAC gave 438 electoral votes to Dwight D. Eisenhower and 93 to Adlai E. Stevenson. When the final count was in and the electoral college convened several weeks later, the official total was 442 for Eisenhower and 89 for Stevenson.
Many who had believed resolutely in the superiority of man's intellect now harboured doubts. A machine, a computing machine, had confounded the experts And a new word, UNIVAC, which when uttered would conjure up fear, awe, or disdain, had become a prominent fixture in the American vocabulary.