In my experience, the class of user that both turns off their contributions to improvements and also remains satisfied simply doesn’t exist.

People will turn off the server-side logging because an article told them to, then they’ll complain that Siri sucks, and the service won’t get better.

Therefore, I recommend that people who don’t want to help improve the services through the logging process simply turn it off altogether. That is the safest way to make sure their utterances are not going someplace they are not comfortable with.

In the current scenario, all server-side logging is on by default and there's no way to turn it off. In fact the devices don't even tell you when or what they're logging, so it can be assumed to be ever-present.

However, if you turn it off and then ask people for their data during specific time periods, (EG can I log your voice commands for the next 24 hours?) then people are aware of what you're doing, there is a limited period where they're giving up their privacy, and it's based on informed consent.

What these companies are actually doing is highly unethical from a scientific perspective. They're using people to train their models without even informing the experimental subjects that their data is being collected. There's no debrief for the subjects so they can understand what was learned.