Much of the necessary knowledge to complete tasks is not written down but 'stored' in the organisation completing the tasks explicitly including the people actually doing it.
I think this is a common case but not the only one.
The counter-example that came to mind is franchises. When you buy a franchise (e.g. a McDonalds) you do not only get a brand name, suppliers, etc. You get an explicit set of written-down business processes (the "three-ring binder" in Snowcrash terms). Part of the whole point of franchises is that you can "complete tasks" in a reasonable manner when starting from scratch, without any intangible institutional memory or knowledge in the heads of long-term workers.
I don't know how explicit running a McDonalds is but this is a very interesting point. This lends credence to my belief that it is not out of some fundamental principle that necessary knowledge to run a plant is not written down.
Here's an account by a retired engineer of what happened when his old company wanted to streamline a process in the factory where he used to work.
People only knew how to keep the factory going from one day to the next, but all the documentation was lost-- the factory had been sold a couple of times, and efforts at digitization caused more to get lost. Even the name of the factory had been lost.
Fortunately, engineers keep more documentation than their bosses allow them to. (Trade secrets!) And they don't throw the documentation away just because they've retired.
I've been concerned about infrastructure neglect for a while, and this makes me more concerned. On the other hand, instead of just viewing with alarm, I'd like to view with quantified alarm, and I don't have the foggiest on how to quantify the risks.
Also, some of the information loss is a result of a search for efficiency. How can you tell when you're leaving something important out?