Forecast factory
Lewis Fry Richardson's 1922 thought experiment: 64,000 people in a circular hall, each responsible for one square of the globe, together computing the weather faster than it happens.
During the First World War, between shifts as an ambulance driver, Richardson computed a six-hour forecast for a single point in Europe by hand from 1910 observations. It took him six weeks and produced a pressure change of 145 hectopascals — catastrophically wrong, which much later turned out to stem from unfiltered oscillations in the initial data rather than from the method. He published it anyway, in Weather Prediction by Numerical Process (1922), and let the reader see the error.
The final chapter sets out his answer to the speed problem. A hall shaped like a theatre, its walls painted as a map of the world, 64,000 human computers in the galleries each minding one grid square and passing results to their neighbours. A conductor stands in the middle keeping time with coloured beams of light. Read today it is a specification for a parallel computer: processing elements, nearest-neighbour communication, and a clock. It took until 1950 for ENIAC to run the same calculation — Charney, Fjørtoft and von Neumann needed roughly a day to predict a day, which was precisely the threshold Richardson had been aiming at.