In the realm of NLP, there are some companies that offer a human in the loop service. So to the outside, they offer human quality results. And on the inside, they use their machine learning technology, supervised by humans. One way to do it, I guess, is to especially watch the results where the model is least confident. And of course, you feed back all of these corrections back into the model, so it becomes better over time.
And those objections you refer to, I am quite sure that people like journalists would’ve had the same objections to machine learning transcription services just a few years back. Now, of course, they all use otter.ai and the like and cannot imagine ever going back to slow human transcriptionists.
It is not possible to "hide" the fact that it is done by software. As there are some differences. The software is stronger in some aspects of the tasks and weaker in other aspects. Overall, the software is superior. But it is noticeably different to work with the software than with a human.
Think Robotaxis: Superhuman reaction time. Never sleepy. But bad at interpreting unusual situations. Overall, they will be 10x safer than human drivers at some point. But you will still notice if a human or a bot drives the car.