I think the automatic translation is a bad example, because it requires you to solve several problems, which are not exactly related to AI. First of all you need to parse the text(syntactically) in the original language which is a hard problem because although languages are thought to be context free, context-free grammars are not known for most of them and there are few suitable and complete data sets(corpora). Second problem is semantics and it is related to AI. Third problem is again constructing a morphism from the original language with the semantic information attached to the target language, which again requires a good corpus and a good algorithm which takes both syntax and semantic into consideration - AI is partially involved here.