Tags and Classification

Problem: DT has a powerful AI mechanism for classification and for suggesting relationships. However, as it stands, the mechanism appears useless to me because: (1) my basic file structure is complex and deliberately redundant, (2) the bulk of the items in my data bases are large, as with long papers or books, and (3) some of my higher-level file-system “boxes” cover many subjects (e.g., Government Reports).

I understand that the AI concept was developed with different users in mind, but it seems to me that it COULD be valuable to me nonetheless, if only…The question, then, is what’s the “if only” and whether the DT folks could/would make that happen.

Question 1: Have others solved this problem with clever tricks?
Question 2: Why can’t DT’s AI be given the optional characteristic of checking tags AND giving them very high priority? Some might argue that if you have to tag a new document, you might just as well file it. However, that strikes me as wrong and as something to be tested empirically.
Question 3, in the same vein: Why can’t DT’s AI be given the optional characteristic of checking and giving major weight to keywords if those are part of a published report’s or book’s meta data?

I have done some modest experimentation that causes me to believe that something like what I suggest in Q2 and Q3 would work, perhaps for See Also if not for Classify, and perhaps for both.

It seems to me that more could be milked out of the AI for those of us who haven’t built our classification system along the lines DT’s AI would find natural.