So you’re not keen on uninspired, all-the-same-looking 2D/3D art, but uninspired, all-the-same-sounding writing and VO is fine?

First, doesn't have to be if done right.

VO issue can be solved by training with the speech of different professional speakers (controversial topic, I know...) And when feeding the base LLM generating the dialog with a suitable amount of hand-written, high-quality NPC-specific examples, the same sounding writing issue can also be tackled.

But secondly, and most importantly, the choice is not between fully interactive realtime conversation using Ai or the same but human generated,
but rather between fully interactive realtime conversation using AI or non of that at all.

That's the main difference between this specific use and, say, the graphics art example.
There the alternative to AI never is "no graphics at all"...

That's the reason why I would accept AI use for this specific purpose, as it delivers something completely new and unachievable by other means, and is not only used to substitute existing human labor.