Tuesday, February 01, 2005

CBS, UNIVAC, and Credibility Issues of 1952

CBS, UNIVAC, and Credibility Issues of 1952

From William Gibson's blog comes the Christie's "Origins of Cyberspace" auction, featuring among other interesting material this Lot.

Pull quote:
On election night programmers fed early return data from eight key states into UNIVAC 1 to be compared with voting patterns from past elections. On the basis of this comparison, the computer predicted a landslide victory for Eisenhower: 438 electoral votes versus 93 for Adlai Stevenson. This contradicted the polls, which had called for a close race, and nervous officials from both CBS and Remington Rand insisted that the parameters of the computer's calculations be adjusted to better match the pollster's predictions. After some tinkering, the UNIVAC was made to produce odds of 8 to 7 for an Eisenhower victory, and CBS used this revised prediction in its election broadcasts. However, when the final returns were in, UNIVAC's initial forecast of a landslide turned out to be correct: Eisenhower ended up with 442 electoral votes, while Stevenson had only 89. The computer had predicted the election's outcome with over 98 percent accuracy, and on the air that night CBS confessed to its mistake in not trusting the UNIVAC's original numbers.

Note for the record that this didn't destroy CBS's credibility.


Post a Comment

<< Home