I do not agree with you here. This is the real difference between proper human translation, and the crap machine "translations" churn out. A human translator writes output suitable for the understanding of a human reader.
Humans gave the "translation" of "Ao'bu'lai'en/Aobulaien". Machines gave the translation of "O'Brien". Guess which one was right.
More seriously, yes, machine translation isn't infallible, but it lacks the biases that humans have, and while a human who is fluent in the language being translated to/from is useful to ensure accuracy for MTL (which can honestly be done these days in most cases by sectioning out characters in batch and checking against jap-eng dictionaries and/or googling them to get a reference; it's only the more esoteric things that
require C1-level understanding that you might need more of a specific understanding for), while localizers are not useful at all. Seriously, MTL has come a long way from where it used to be. I've held conversations with Japanese and Chinese people on Twitter, Discord, and elsewhere, and I was able to understand them and vice versa just fine, using MTL to translate. Some words do trip it up (as I said, it's not infallible), and you can usually test this by feeding the translation back in but in reverse, but then it's just a case of choosing a synonym to get the exact meaning across.