Page 2 of 3
Over the years to come Hollerith was always generous in attributing the basic idea of using punch cards to Billings and even offered "Chicken Salad" as the two-word explanation of why he invented automatic data processing!
However having the basic idea of using punch cards was a long way from making the machines that processed them fast and reliable. In 1882 Hollerith took a job at the Massachusetts Institute of Technology teaching engineering. He used his spare time to build his census machine. After only a year he moved back to Washington to work at the Patent Office. Then he set himself up as a patent expert - knowledge of the patent system is something that he would need in the coming years!
In 1884 he took out his first patent - "Certain new and useful improvements in the Art of Compiling Statistics". He took a part-working machine to the census office but before they put money into the project they wanted a working demonstration. Hollerith borrowed $2500 from his brother and became the first of the "garage" startups in the history of computing!
He had considered all sorts of ways of recording the data so that it could be processed. Paper tape seemed like a good idea. It was already in use as part of a high speed telegraph system and so there were some machines that processed it. He abandoned this idea because he realised that the tape format made it difficult to get at individual records. Hollerith recalled seeing a railroad ticket collector using a "punch photograph" a ticket that was punched to indicate what the passenger looked like - light, hair dark, large nose, etc - and he decided that punch cards were the thing to use.
At first his punch cards only had holes around the edge because his punch couldn't reach the centre but it didn't take long for him to solve this problem. He had re-invented the punch card of Babbage and the Jacquard loom but it is not clear if he knew of these earlier efforts. As his brother was in the silk business it is possible that he heard about card controlled looms from him.
Hollerith looked around for ways of proving that his machines worked and so he offered to automate the Baltimore health records which were in a complete mess. Hollerith did much of the punching of the cards himself, a not inconsiderable feat using a manual punch at 1000 cards per day each with ten or more holes. Each card contained the data for one patient and once punched his tabulating and sorting machines made it possible to get answers to questions that previously seemed impossible.
An early Hollerith card
The machinery for handing the cards may have been mechanical and it may not have been a stored program computer but to the people involved with the medical records its impact was just as great and just as remarkable. The age of automatic data processing had arrived.
Even so it is important not to imagine that the card handling machines were anything like the high speed paper handling marvels that were developed in the 1960s. Hollerith's first tabulator used a set of steel spring loaded pins - one for each hole position on the card. If a hole was punched then the pin passed through and completed a circuit by dipping into a small depression filled with mercury. The operator had to pick up a card and place it into a frame over the mercury filled depressions. By closing the lid which held the spring-loaded pins the data on the card was registered on the counters. This may not seem like a much faster process then manually tallying the cards but each hole position on the card had a pin and an electromechanical counter - so the entire card was processed in one operation.
The automation of the health records was a complete success and Hollerith rented machines to the Health Department - his first commercial success. His next project was the automation of health records kept by the War Department. They were also willing to rent his machines but in this case the problem was more difficult and needed more data stored on each card.
Instead of simply increasing the size of the card Hollerith decided to allow combinations of holes to mean something. This was of course the start of the use of coding that would eventually lead to binary and ASCII codes. Of course all of this success was nice but Hollerith's eye was firmly on the coming census of 1890. He was offered a job with the census office but he turned it down. He wanted to sell his machines, not be part of the census!
To select the machinery to be used to a contest was arranged between three competing systems - one using coloured slips of paper, another using poker chips and Hollerith's cards. The test involved the data for 10,000 people and Hollerith's machines finished the job in just over 3 days. Most of the time was spent punching the cards, the actual processing took just 5 hours - the second fastest method took 4.5 days.