When you learned idioms, you probably looked at them from the encountered point of view. Meaning: I'm reading this Russian text and this idiomatic phrase occurs. Oh, that's the equivalent of saying this in my native tongue.

But it is trickier to be looking at an idiomatic native text and intuit: I must replace this with some equivalent when translating this to Russian.

My favorite book about the vagaries and pitfalls of translation is Douglas Hofstadter's "Le Ton beau de Marot: In Praise of the Music of Language". It covers the difficulties of translating idioms, puns et al. And those don't even cover whether translating a poem means "word-for-word", "rhyme-scheme", "thematically" or some other criterion.

So, maybe idioms are not a difficult test of an AI's competency. But I still feel the fundamentals of natural language will be part of the determination.

Edited: to remove a redundant phrase (idiomatic native idiom).