« La Grignotte | Main | 43:12/177 »
August 05, 2005
Where Google's not enough
One of the blogs I scan sent me to an article on Recent Trends in Enterprise Search by Stephen Arnold. Arnold argues that:
Humans and human-like processes are needed to supplement or do certain types of taxonomy development, indexing, classification and analysis.
We do this at home all the time. Look in your kitchen cupboard. The dinner plates are probably stacked. The coffee cups are probably together somewhere else. The pots have their own place. The cutlery is similarly segregated. Where you actually put each class of item -- the specifics of your classification -- depends however on your particular way of handling meal preparation and cleanup.
You probably also organize your workspace somehow, and that organization corresponds to how you do your work, whatever it is. Your books and music are the same way. It's only when we want to share stuff that we need a shared organization protocol (Dewey decimal, Library of Congress, WWW, etc.), and indeed it's only when we want to share stuff that it makes sense to go through the trouble of using a protocol.
The more how we share information follows some sort of process, the more specific the protocol needs to be. If to use Arnold's example, "A lawyer working for the chief financial officer [needs] everything pertaining to a deal involving dozens of employees, several departments, and documents in all versions over a span of years," then a search engine using a PageRank algorithm won't be the right answer.
Yet I'm not sure humans need to do the indexing. Humans need to define the sharing protocol, though, and those doing the defining ought probably to be those very familiar with the process, able to define it formally or at least follow it intuitively in the same way as others follow it. That sharing protocol is likely to come out differently for each process in which information is shared.
So in the end I agree that Google's not always enough under our circumstances. Furthermore, maybe no finished general purpose search engine can be.
Trouble is, selling the idea that software has to be customized can be tough. As Linnea said, it used to be fast, cheap, or good, pick two. Now the choice is made: fast and cheap.
Posted by Mark at August 5, 2005 07:41 AM