Great questions. Specifically for #2 it's hard to say at this point. At present (in it's very early state), I believe they just pick the first result. But the resolution process returns a "confidence" value. This is purely my speculation and I need evidence before I publish it in an article, but I believe that confidence here indicates the intent to use ML.

It can probably be assumed that you would be able to set a preference for one module over another, but I haven't found any proof of that yet.