I think something that's been somewhat overlooked in the Siri discussion is the expectation of usability. Our expectation relating to standard touch screen/keyboard/mouse usability has been fairly standardized by devices over the last ten to fifteen years. With Siri this expectation is not yet present, meaning we may have unrealistic expectations for the timeframe between command and action, and we will have far less tolerance for misunderstanding, mis-communication or failures with Siri than we might with traditional user interfaces.

The bottom line is that Siri better be basically perfect - if it's unreliable, even within what in engineering terms would be totally acceptable standards for such an advanced bit of software, people will just not bother and revert back to standard UI mechanisms. This is always the danger with any new user interface, but, I think, especially relevant here. After all, people tend to get pretty annoyed when you don't listen to them.