It's probably a mix. The computational power to run a dialog system is one thing but it's also just more convenient in terms of maintainance to have the system be completely server side.

Of course, getting your sweet voice for future training also helps!

>Of course, getting your sweet voice for future training also helps!

There is a setting to disable having your voice saved along with options to automatically delete it after a number of months.

Ok, but then unfortunately we have the trust issue.

Do we trust them, to:

a) do it and not just lie about it totally or partially, and we find out later on when somebody does eventually report it/find out/investigate, if ever happens.

b) even if they intend to really do it, that they do it properly.

If it doesn't reach them, you are 100% sure the data is not there and not at risk (from their end of course). Otherwise is just let's hope for the big corporation with optimizing for most profitability for their stackholders as main objective to do "their best at protecting it's consumers data and privacy".

Which sure... hurting consumers and getting fines aint' great, but not always ends up in less profitability than doing the right thing from the beginning.

Of course... this is not a big deal compared with other stuff, there are alternatives and it's not something you really depend on day to day. Compared with other stuff that is for sure.

We have become absolutely spoiled, now talking with a machine feels like something an arm chip should be able to handle

You can preprogram super cheap chips to do voice commands. They come with limitations but they don't wire tap your home to sell private conversation data to advertising companies to later water board your kids into purchasing thing's on social media

You might write those limitations off as an asterisk , but they are probably orders of magnitude of differing quality.

You don't make a state of the art product by shipping some code and never getting any direct feedback on it.