Natural Intelligence

Before handwriting recognition, bar codes, and national zip codes, the Post Office had people. Brooklyn Postal Station S, at 1075 Lafayette Avenue:

and the Vanderveer Postal Station at 2319 Nostrand Avenue:

What you’re looking at are, in modern language, the work stations for mail sorting clerks. In the second photo, you’ve got a couple of men at work on the left, taking stacks of mail and sorting it, by hand, to the cubbyholes that represent different addresses. This wasn’t the last time a pair of eyes would be on each piece of mail: the carriers would pack bags from the cubbyholes for delivery and would look at the address at least once in the process of packing and walking their routes. But this was the primary sorting.

Note that the workstations (a) have stools and (b) have top shelves high enough that there was no way you could sit on your stool the entire time you worked. Also, I used a stool just like those, sitting at a drafting table at my first job, and they are uncomfortable for any extended period of time.

It’s not surprising that the Post Office looked to automate this task, as it had been a bottleneck in the distribution of mail. Good optical character recognition for handwriting is a twenty-first century invention, but if you can narrow down the possibilities, even bad recognition works well. The Post Office introduced two-digit location codes in big cities in 1943, as postal clerks were being drafted and an aid for their inexperienced replacements was needed; the five-digit ZIP codes were introduced in 1963 along with the two-letter state abbreviations. The use of automated sorting in the US began tentatively in the 1950s and really picked up speed in 1965. The use of two-letter, capitalized state abbreviations greatly simplified reading the state for the machines; the use of numerical zip codes simplified getting the general area right and also simplified identifying the address. If your OCR has to compare a sloppily-written street address to the entire country, it’s a difficult computing task; if it only has to compare to the street names in a specific zip code, it’s much easier. And finally, people tend to write block-print capital letters and numbers more legibly than they write longer addresses.

In other words, the use of artificial intelligence to replace people was successful based on simplifying the task to something that the machines could handle. Those same simplifications made the task easier, albeit more boring, for people.