Metrist (a web marketing firm, reached via O'Reilly Radar) just introduced the concept of Data Farming, as opposed to Data Mining:

Here's an alternative paradigm: data farming. In farming, you start from a seed or an immature plant, place it in an appropriate growth medium, and tend it with nutrients. Crops, and perhaps seeds for the next cycle, repay your efforts. [...]
I contend that we should look at direct marketing analysis as data farming, and to use the metaphor to justify (with ample support) investments that will pay off marketing cycle after marketing cycle.
And many of us are already doing data farming: extending successful promotional elements to new prospect lists and new vehicles, learning from last season in ths season's re-activation programs. This is farming, sustainable and nurturing, not at all like mining. And smart farmers don't just produce; they learn so that next season will be even better.

Although this is hardly a new practice within the marketing analisys community, I think it's worth noting that most social applications (like Flickr or del.icio.us - and even Wikipedia, eBay and Amazon) can be seen as growing, dynamic and freely accessible data farms. They are not mere "data repositories", since users are continuously stimulated by well-designed social-aware mechanisms to provide new data. The "farmed" data can be easily accessed using specific APIs, and analyzed with more or less sophisticated data mining techniques (even a simple idea can lead to interesting results, see for example Color Code, which explores color usage patterns).

I think the most relevant challenge in building a data farm is the bootstrapping (start-up) phase. What emerges from the Folksonomy Panel at ETech 2005 is that people are willing to produce and share information for many reasons, but mainly because they are going to re-use this information themselves in the future; so, it is up to the (software) infrastructure to enable the communitary reuse and recycle.

Also: John Winn of Microsoft Research has started ML-pedia, a wiki on machine learning models and techniques.