I am not proposing sucking in all the data, I mean an awareness of its environment so it knows when voice is appropriate. For instance if it can detect me driving in my car (due to speed) with no passengers (noise it has heard) maybe it can understand it should use voice.

You do realize that in order for the assistant to know you're in a car, and that you're "at speed" it would likely need access to your location information, right? Aggregated over time (even if only when the assistant is "on" - ie, the iOS "while using" setting), that's a pattern of your movements.

Either Apple/Google/etc and granulize the permissions such that things like "boolean: user is in vehicle at speed" are addressable separately (possible but unlikely) or such knowledge comes with a cost - the app knows where you are and have been.

Same applies doubly for "noise it has heard".

Consequently my decision is: no assistant, thankyouverymuch.

Ok, so how exactly would assistants do that? They'd need to take environmental inputs such as GPS data to determine if you were travelling in a car. It would need to have consistent access to your microphone and speakers so that it could gather and compute info relating to your environment (noise levels, voice recognition, etc). Combine that with the fact that most phones already track your location fairly precisely. There exists almost zero incentive for Google/Fb/Apple/Amazon/MSFT to NOT collect and aggregate massive amounts of user data.

How will it learn to compare all these different environments? ML obviously; which pretty much entails leveraging the consumer data to provide a better product. The cost of usable, voice-driven assistants is giving up a vast trove of personal data and abandoning the concept of privacy completely, just so I can order some paper towels from an Internet of Shit connected device instead of walking three blocks to a store.

Your position is akin to a Louis CK joke(not offense meant): "Oh boy! I hope he doesn't do what he's going to do!"