NLP will only continue to improve in the years ahead. I've got a few echo's around the house, and yes the current implementation of NLP is not perfect. It requires some modification of how the user speaks, but the first time you say "Alexa what's the weather today?" and Alexa comes back with the day's forecast is a jaw dropping moment (at least for me it was). I believe voice activated devices will ultimately be placed all over the house and we'll communicate with our computers for a lot of various tasks through voice - it's just natural for humans. Think of all the mundane tasks you do typing into a computer - setting calendar appointments in outlook for example - and then think how many mouse clicks and typing are required. It's so much easier to just tell a computer using a voice command.

The privacy issue is easily solvable IMO. Google already requires a physical mute button on their devices to disable the mics in hardware not just software. It will require either user trust (admittedly in short supply with tech companies lately) or some government regulations but there is no reason even today for devices to have to "record" your conversations in order to process. As long as you require a "wakeword" to tell the device to start listening to your next commands the device doesn't need to record all audio all the time.