Nervous System: UNIVAC Predicts the Next President

David Kalat examines CBS’s television coverage of the 1952 presidential election, when the UNIVAC computer predicted seemingly unbelievable results.

Download the article.

As the United States experiences another “Super Tuesday” of primaries, it is interesting to look back at the curious history of televised election predictions.

On the night of the 1952 presidential election, CBS’ Walter Cronkite welcomed computer scientist J. Presper Eckert to explain to viewers how programmers had been feeding voting statistics from past races into a computer called UNIVAC. Periodically through the broadcast, Cronkite and his fellow journalists would turn to Eckert to ask if the machine had a prediction yet—each time, the teletype machine that was rigged to UNIVAC to serve as its output remained silent. The event had been conceived as a publicity stunt, and CBS had anticipated a dramatic brush with the high-tech future, but as a piece of television the prediction was a disaster. Only after Dwight D. Eisenhower was declared the winner did CBS’s news team sheepishly admit to the viewing public that in fact, earlier that day, with just a sliver of the votes counted, the computer had correctly predicted the final Electoral College result within 99 percent accuracy. CBS had buried the result, thinking the prediction was a mistake.

UNIVAC was a mainframe computer engineered by Eckert and his partner John Mauchly. A few years earlier, Eckert and Mauchly had created ENIAC for the US government, but a quirk of patent law deprived them of any ownership stake in the computer that helped win World War II. The two scientists had then struck out on their own to create computers for civilian use. UNIVAC was the first American nonmilitary computer, but an eight-ton device the size of a garage with a 1951 price tag of over a million dollars was a tough sell to anyone who did not need it to win a war. Eckert and Mauchly sold their company to Remington Rand, which struggled to find buyers for UNIVACs.

Hoping to promote sales through televised exposure, Remington Rand approached CBS in the summer of 1952. Their suggestion was to have the computer predict the next president, live on the air. CBS was skeptical of the machine’s abilities but interested in the gimmickry of the idea, and agreed.

Due to the enormous size and unwieldy nature of the computer, a mockup was built for the cameras, to give Eckert something concrete to demonstrate. Meanwhile, back in the lab, mathematician Grace Hopper and statistician Max Woodbury built an algorithm to derive a prediction by comparing early returns against a database of past election results.

Shortly after the broadcast began, UNIVAC’s teletype printer started working. The paper printout read, “It’s awfully early, but I’ll go out on a limb,” predicted Eisenhower would win with 438 electoral votes to Adlai Stevenson’s 93, and capped it off by claiming the odds were 100 to 1 in Eisenhower’s favor.

It was generally assumed that whoever won, it would be a tight race. To the extent either candidate had a lead, opinion polls tended to favor Stevenson. When the computer declared that the underdog would win by a landslide, CBS’ news chiefs refused to air it, fearing the television equivalent of a “Dewey Defeats Truman” headline.

The computer’s prediction seemed unlikely. The machine had leapt to its conclusion far earlier in the night than anticipated and had announced an astonishing level of confidence. As shown in the printout (now preserved in the Computer History Museum), UNIVAC actually claimed that the odds were 00 to 1for Eisenhower, because UNIVAC’s programmers never expected the machine to need to report better than double-digit odds.

After CBS buried the prediction, Eckert and his team regrouped hastily. Predicting the next president was not the real goal—the winner of the election would be known in a few hours. The point of the exercise was to impress the general public about UNIVAC’s competence and utility. The engineers fed UNIVAC additional data, hoping to “improve” its performance. When the machine spat out a new prediction with more modest 8–7 odds in favor of an Eisenhower win, CBS correspondent Charles Collingwood felt comfortable repeating that part of the prediction on air. Meanwhile, however, the UNIVAC programmers realized they had input some of the recent returns incorrectly. When the error was fixed, UNIVAC reverted back to its original forecast—an Eisenhower landslide with 438 electoral votes, and a 100 percent certainty of an Eisenhower victory. CBS did not update its reporting.

The final tally showed Eisenhower with 442 electoral votes to Stevenson’s 89. UNIVAC came within one percent of the correct answer.

In a moment of television almost as stunning as Eisenhower’s victory, CBS’s Collingwood told the viewing audience the truth. The computer had not been sitting silent all night as viewers had been led to believe—the humans in the newsroom had suppressed the fact that an electronic brain had called the outcome correctly hours earlier.

Remington Rand could not have hoped for a more dramatic and memorable public relations win. UNIVAC became a household name.

The views and opinions expressed in this article are those of the author and do not necessarily reflect the opinions, position, or policy of Berkeley Research Group, LLC or its other employees and affiliates.

Read the full article. (subscription required)