Exactly, certain ideas seem perpetually "futuristic" but never seem to catch on: voice recognition ("Imagine dictating your letters!" => Really annoying to make corrections), gestures ("Imagine waving instead of using a mouse!" => Gorilla Arm), video chat (I want to wear my pajamas and not pay too much attention), and so on.

That said, I don't want to discount some new innovation taking over :).

Dictating letters is still niche and will likely remain so, except for people who have severe RSI or other reasons to go through the training required. If I'm spending enough time at one sitting to do a letter, and if I have the investment in learning how to touch type, then I'm a lot faster at typing than dictating. Even if the dictation is perfect.

On the other hand, voice recognition for specialized vocabularies is happening. I've noticed a progressive replacement of "push 1 for X, push 2 for Y" with voice interfaces in support calls. My car has voice recognition for the sound system that actually works, e.g. "Play Track Shelter" plays a song of that name. While there are still freaky behaviors, these work well in that they don't waste my time when I'd rather be doing something else -- like having my problem fixed or listening to Icon of Coil. Because they are for specific narrow domains, these applications of voice recognition coexist with my typing, touch, et al.

Put another way, new UIs and input devices seem to work best when they have a path for incremental adoption. It's rare to see a complete overnight change. Even the iPhone, as amazing and revolutionary as it is, built on pre-existing experience with touch screen phones. (Maybe "reacted against" is a better term than "built on.") Plus the iPhone still has a keyboard, even if it's a neat touchscreen one instead of a hardware one.