Oh man… PDP-11. It’s scary to ponder how much critical infrastructure runs on obsolete systems that a vanishingly small number of people understand.

Contemporary with the PDP-11 is the IBM Series/1 which runs the nuclear forces of the USA. The 8" floppies were retired in 2019 though. Progress? I guess.

Much older than that are the IRS Master files: the US tax system runs on IBM 7074 assembler although I think the actual 7074 hardware is long gone.

Presumably for something as expensive as the US nuclear weapons program, they could afford to hire and fully train 10x as many people as they need to be experts in externally-obsolete software/hardware and it just have the effect of a rounding error in their overall budget?

If that were cheaper, or considered more reliable or less hackable, than upgrading the systems to run on more modern technology then why not, as long as they look ahead far enough to make sure they fund the required tutoring to get the knowledge being passed down a generation or two, and do it before getting to the point where not having enough young experts actually has become a problem.

Can you give an example of modern technology that you believe would be supported or maintained in 50 years?

The PDP-11 was unusually well suited for that. They designed it so you needed less skilled workers to assemble it which meant a lot of tolerance in components which leads to much easier maintenance. For example, the backplane is wire wrapped so there's no soldiering there which can eventually tire out. If things move around, oh well.

Also the PDP-11 was designed at the time boundary of core vs semiconductor memory with the clear intent to be able switch to semiconductor which led to an unusually flexible design around memory (and also affected I/O design). Check this 1975 paper https://dl.acm.org/doi/pdf/10.1145/800110.803541

The tax system might soon be running on something more modern. Supposedly they are nearing the finish line on porting the whole stack to Java.

Nearing the finish line? https://crsreports.congress.gov/product/pdf/IF/IF12525

> One of the plan’s priorities is the retirement of the Business Master File by FY2027 and the Individual Master File by FY2028

But it's not at all clear whether there's funding available to execute this plan.

I mean, yes, it's been 15 years and now only about five remains (because these projects never meet deadlines, let's face it) so they are nearer the finish line ... but it's very far still.

Wow I remember my grandmother telling me about 8” floppies. Literally, my actual grandmother. Amazing that they were still in use up to 2019.

My grandfather worked with Marcell Jánosi https://www.inventingeurope.eu/knowledge/floppy-contradictio...

> Much older than that are the IRS Master files: the US tax system runs on IBM 7074 assembler although I think the actual 7074 hardware is long gone.

There HAS to be a project to feed the Master file(s) thru a modern AI to analyse and document the file(s).

You want to write documentation of something with significant legal repercussions using a probabilistic analyzer?

Your Honor, my client could not have committed tax evasion because these were the most likely numbers based on millions of tax reports analyzed.

Please, please do that in front of a judge, I will bring the popcorn.

An interesting scenario.

But I refer to the IRS's core executable code. Isn't the current state of affairs that it is code that is not well-documented, or not even documented at all ?

That's the impression I have gotten. Correct me if I am OTL.

What's the difference between AI documenting the system that makes decisions on the tax code and using AI to make decisions about the tax code? Nothing. I used this hypothetical scenario to highlight the absurdity and absolute idiocy of what you are suggesting.

I think number of such machines in developing countries is much larger then in developed.

In 80s, people said "it is good, that we don't have western infrastructure, because when become democracy and need it, will build all new from zero".

What really happened then, few waves of upgrades at West, when obsolete systems exported to developing countries very cheap.

But this is not whole picture. As First World is just about Billion humans (from more than 6 Billion at about Y2K) and have very limited land, number of obsolete systems exported to developing countries is just not enough for their size, so developing countries could not build whole infrastructure from just obsolete systems from West, but have to do (or to buy) something other.

So now, developing countries usually have mixed infrastructure, partially already obsoleted when imported systems, partially modern systems (many made in China), and partially some custom, even literally hand-made or garage-production.