Could someone please explain me the following?

I've read the whitepaper and everywhere throughout it there is a talk about removing the redundancy that exists in current centralized solutions.  And that the work of a worker will be verified only with some probability.

But as far as I understand, the redundancy plays important role in establishing the solution's accuracy which is close to 100% in redundant case.

I mean, if the work is performed by a single worker then the human factor plays a big role. Even the big specialist will make mistakes time after time. A mediocre specialist might make mistakes say 3-5% of the time. And given that the verification of their work is probabilistic, most of the unintented mistakes will be presented to requester as a proper answer (which is not).

On the other hand, the redundant systems don't have such a problem.

Am I getting something wrong?