Voice is horriblly slow medium of transfering information. I read because it's faster than listening to an audiobook. It's not scannable. You can't skip through the unimportant parts with one thought as you can do when you look at things.
You can listen to a single voice stream at a time so when AI talks to you you are more cut off from the people around you than when you look at our phone. ...unless exchanging glances is more important than what people are actually trying to tell you when you happen to look at the screen.
Well that's a shortsighted way to view voice. You don't have to use audible feedback alone. Combine aural, visual and tactile feedback and you have a much more efficient means of control.
For example, in the carputer I designed years ago, I could see the information on a small LCD screen, I could hear it with text-to-speech, and I could feel it with vibrations from motors in the steering wheel. I could also give feedback by speaking commands or entering them on a small keypad. So I could switch between fast visual feedback or multitask my visual road input and audible computer input.
Not to dispute your comments, I fundamentally agree, but it is possible to speed up voices to the point where you are getting information at near reading speeds. The ear is also remarkably good at separating information at different frequencies (like how you can pick a single type of instrument out of an orchestra), so it may be possible to encode multiple streams of information that way as well.
I'd imagine AI with voice interfaces would sound like chipmunks but their users would be able to understand them as they would grow up with them.
Still AI would have to understand you and monitor you and your surroundings so that it won't give you information you don't want and not to interrupt anything that has higher priority for you. Like your wife asking you a question. On the other hand AI could repeat and rephrase your wives question in its chipmunk voice if by observing your body it detected that you missed it or misunderstood it.
Ear adjusted to listening AI could have problems with understanding what humans say.
I can talk much faster than I can type, and read at about the same speed as listening. You can also speed up audio to inhuman speeds, it just takes a little bit to get used to.
It's also a lot more handy than pulling out a keyboard (especially a mobile or touchscreen one) or screen.
Power users would increase the speed of audiobook to a point where it's barely comprehensible to someone who is not used to it. It's pretty much as fast as reading. But yes, you can't skim easily.
You can listen to a single voice stream at a time so when AI talks to you you are more cut off from the people around you than when you look at our phone. ...unless exchanging glances is more important than what people are actually trying to tell you when you happen to look at the screen.