Skip to content

The roads must roll

In a past life, I worked on one of the first AI/ML/NN [1] processors for postal machines (also built one of the first CCD cameras for them, and one of the first NN ASICS [2]).

The most impressive part wasn’t the handwriting recognition (which wasn’t at that time any better than the hand-built emperically determined resistor network [3]), or putting said network in an FPGA (which was a new thing at the time, along with the onions in our belts).

No, the most impressive part were the huge machines that sorted the mail according to whichever camera and processor was bolted to the back to make decisions. These could be upgraded in perpetuity. The mechanical behemoths that sorted billions of pieces of mail, the rollers and belts and pulleys – those were engineering marvels. Hard to see how those could be replaced. Built by companies that no longer exist, by dudes (sadly, all old white dudes) who have long since retired or died.

To see them get thrown out on the street, never to be resurrected, was heart-breaking. Those will never get rebuilt, replaced, with something better.

They could be, of course. But like new roads, or keeping restaraunts, gyms, and theatres viable during a completely unforeseen pandemic [4], it’s not in the best interests of the deadenders to have a Post Office, as specified in the Constitution. Maybe there will always be something called the Post Office in the future. But a system that could deliver a handwritten envelope across a continent in a few days, for a pittance? That’s gone now, forever, because it being gone is in the best interests of a few.


[1] It’s AI when you’re selling it, ML when you budget for it, and linear algebra when you implement it, as the now old joke goes.
[2] Probably? 1986-ish
[3] But I guess that was pretty good for a first try. Problem was, it never got any better on the second, fifth, or tenth try. Then I went to physics grad school. I didn’t want to wait around for Moore’s law for 20, wait, no 30 years to get to the point where it was useful. But how useful is it if it takes the same amount of energy to solve a problem as it does to build a car?
[4] Oh wait, it was completely foreseeable, because it had happened before. And government plans existed to deal with it. And previous administrations implemented teams to plan for it. ¯\_(ツ)_/¯