See, but you're still thinking in current voice-interface terms. If I want to edit something that I "misspelled" (if this will even be a problem, since you are

See, but you're still thinking in current voice-interface terms. If I want to edit something that I "misspelled" (if this will even be a problem, since you are _talking_) I'll just tell my digital secretary "errr, I meant <x>" and it will know what to do. Same for turning off the lights: Sure, it's faster if you only consider the flicking of the switch. It's another thing if you also incorporate the time it takes you to get up from your couch/bed/wherever you are without a light switch in arm's reach. You definitely have a point with privacy though. Also, the volume of all people on a train talking to their smart assistants (though the question remains if this is really so much different from people talking to other people on a train). EDIT: I misread your point about mistyping vs. misspeaking. Still, the interface I'm talking about does not work in modes. It's able to truly understand you and interpret your commands appropriately.