>i'm not sure if has to be in real time. if you think about it, people use all different ways to fill up some time before they finally, after innumerable little pauses, sidebars and parentheticals (like this) they get to the point

Definitely. What I'm saying is, if we want full immersion, that at the very least it will need to be able to respond as fast as a human. And that is often nearly instant in natural conversations.

And of course even when it gets to the point where it can have instant responses, to keep the realism it will have a logic system where it decides how long it should pretend that it's "thinking" before it starts voicing out its response, according to the nature of the conversation and how long a regular human would need to think to respond to a particular statement.