> If you replace a computer with a piece of paper with a different name on the account, it doesn't make it any easier for a service representative to override.

It certainly does: * Before computers, most of the roles of the computer were not fulfilled by pieces of paper. It was the teller-of-the-past who was the human front-end and identity verifier/calculator/UI/running policykit instance/etc.

* teller-of-the-past was necessarily trusted to use their discretion to a greater degree than teller-of-the-present

* teller-of-the-past could cache the results of an identity verification on Tuesday to use as the identity verification on Wednesday

* supervisor could bless teller-of-the-past to teach teller-of-the-past2 how to do an end-run around policy foo for edge case bar, where edge case bar involves a specific human or category of humans. (As an aside, the supervisor could do this more quickly than a team of third-party software developers can code, test, and deploy an interface for the same human or small category of humans)

* piece of paper could not lock teller-of-the-past out of completing a transaction with the customer

* even if teller-of-the-past decided not to fulfill the transaction, the equivalent of the computer's identity verification algorithm was probably someone else standing in the same physical bank location who could be asked to help complete the task

> This isn't a computer problem. It has always been inconvenient to change your name.

I think what the author is trying to describe is a problem of lack of human discretion, which is something that wouldn't exist at the scale that it currently does without over-reliance on computers. Gmail and plenty of other services don't degrade to humans at all, so I don't think this is subtle stuff.