This article in the Washington Post notes that University of Maryland researcher Ben Shneiderman has a good reason to believe that voice interfaces for controlling computers are fundamentally flawed, and that reason is speaking uses auditory memory which happens to use the same part of your brain as short-term and working memory. Therefore, it's hard to speak and think at the same time.
Using a mouse or keyboard to execute commands is easier than verbalizing them because hand-eye coordination uses a different part of the brain.
I'm no HCI expert, but that sounds right to me.
It is part of the vanity of mankind that we want computers to imitate us. Sure, some people probably wish for Jude-Law-like robot companions as in A.I. for those lonely nights when they can't find a prom date, but that's just creepy. The utility of a robot that is passably human has always been somewhat dubious. When Big Blue beat Kasparov in chess, thankfully no one thought it necessary to have a humanistic robot sitting across the chessboard from Kasparov, with robotic arms to move the pieces.
Computers do some things far better than humans, and vice versa, and almost all of the productivity benefits I'll experience in my lifetime come from focusing on those. I don't want a robot to make me breakfast unless he can do it at least one order of magnitude better than I can. Far preferable to have computers continue to do things I have no chance of doing alone, as a human being.