I call it "automated storytelling with post-hoc computational narratives." Old, big PDF: http://s3.amazonaws.com/vitorio/Automated%20Storytelling%20M...
The thing about it is that I don't think the stories the software generates have to be precisely true. I don't even think they have to be internally consistent over the long term. Look at the data, generate a "good enough" story that helps push you toward your goal, and revise it later. Personal historical revisionism. In the long term, you won't remember the details, all you'll know is that your mood is generally better and you're two inches slimmer in your waist, so, what does it matter how it happened? The computer is looking out for you, that will be all that matters.
I think the most useful aspect of large digital archives will be when natural language processing and image recognition get good enough to automatically distil the archive into actual information (rather than just data), no manual reviewing ever.
This would be along the lines of being able to ask "What did Bob say about X?" or "Find things that correlate with weight loss" and get meaningful answers (with summaries and so on, as appropriate).
(However, this seems mighty similar to saying "sufficiently smart compiler" http://c2.com/cgi/wiki?SufficientlySmartCompiler.)