UNIVAC Called the Election Correctly — But CBS Hid the Result
Zusammenfassung
On election night, November 4, 1952, the UNIVAC computer predicted that Dwight D. Eisenhower would win in a landslide with 438 electoral votes — hours before the returns were in. The human experts dismissed this as impossible; pollsters had predicted a close race. CBS decided not to broadcast the prediction. Only after Eisenhower’s actual landslide victory did CBS confess to viewers that UNIVAC had been right all along. The final electoral vote: Eisenhower 442, Stevenson 89. UNIVAC had been off by four.
The Prediction Nobody Believed
Remington Rand, UNIVAC’s manufacturer, offered CBS the use of the computer to predict the 1952 presidential election. Statistician Max Woodbury and a team of mathematicians programmed UNIVAC with voting data and historical patterns. Early in the night, with approximately 3 million votes counted out of a projected 61 million, UNIVAC produced its prediction: Eisenhower would win decisively, 438 electoral votes to Stevenson’s 93, with 100-to-1 odds in Eisenhower’s favor.
The CBS team was alarmed. Pollsters and pundits had unanimously predicted a tight race. Walter Cronkite, covering the election for CBS, was skeptical. The UNIVAC team itself, doubting their own result, re-weighted the algorithm to produce a more “reasonable” prediction: 8-to-7 odds in Eisenhower’s favor. CBS broadcast this adjusted, hedged prediction instead.
The Confession
When the actual results came in — a crushing Eisenhower victory — CBS went back on air to acknowledge what had happened. The computer had been right from the start. The humans had suppressed a correct prediction because it contradicted their assumptions.
The original UNIVAC prediction, made with a tiny fraction of the final vote counted, was within 0.9% of the actual result. It was the first time a computer had predicted a US presidential election — and the first time humans had overridden a computer prediction and been wrong to do so.
The incident established the pattern for all subsequent election-night computer projections: the machines would be given the data, and humans would be expected to trust the results.