Imagine if Math calculators were just subtly wrong some percentage of the time for use cases that people use dozens or hundreds of times a day. If you could punch in the same math formula 100 times and get more than 1 answer on a calculator, most people wouldn't trust those for serious work.

They probably wouldn't think that the calculator makes them faster either

If calculators did work that way, I'm afraid that people would nevertheless take them up because "it saves so much time", and would develop fancy heuristics to plausibility-test for errors.