We found this to be a cogent take on the issue of Big Tech listening in on voice recordings. The author makes the perhaps subtle but still important distinction between having concern that Big Tech is listening to voice recordings in order to improve NLU and how these companies use and profit from our personal data.
Meredith Whittaker from the NYU AI Now Institute makes the point clearly that, right now, we need humans to do a lot of the heavy lifting of AI, and that includes making sense of spoken language in ways that software cannot yet.
While, as the article mentions, there is exploration in how to get around the review of recordings, it remains "the cheapest and most possible way to do this". We have not heard of the simulation method being used successfully, however something like that could aid the effort to improve systems while keeping private conversations private.