An article in The Register:

Why machine-learning chatbots find it difficult to respond to idioms, metaphors, rhetorical questions, sarcasm.

The way I see it, if AI finds it hard to understand everyday language, (a) I don’t think there’s much danger of it taking over the world just yet and (b) if it tries to we can probably fool it in some way.

On the subject of language, there is quite a bit of difference between English English and American English. Actually, the words sound the same, but the cultural context is often quite different. This can be a minefield, as I discovered: see Language Barrier.

As for robots becoming masters of the universe, I’m running a series (sporadically, it has to be said) about predictions made by science fiction authors. See Dystopian Visions.