How to train AI Skills Extraction

I don't see any information in the documentation about how to train the AI Document Extraction process.  I thought that invoices were already trained, but that doesn't seem to be the case. Additionally how would we train a new document class, like insurance claims for example. I see doc'm about how to setup the fields, but how do we train for locate/extraction rules?   Do we just run a bunch of docs thru manual reconcile?  Is there a way to clean out bad documents/trainings? How do we get more insight into what's going on behind the scenes with extraction?

Thanks in advance!