Sunday, October 11, 2009

the Human Element

The automation trend in the information management industry has been building momentum for many years. Data scrappers/manipulators pull data from different sites to aggregate results for several services like Kayak (for travels plans) and Indeed (for jobs). The question continually arises about whether human intervention is actually required for data management or should we simply submit to the machines. I think that neither having a solely human or machine driven system is preferable to a hybrid version. Given certain parameters certain computer programs can successfully maneuver through a site to gather information with better consistency and speed than a single human operator. Though those same rules that allow the program to scrap the data can also mean that it pulls in too much information, too little or simply the wrong data entirely.

For instance, recently I had been tagging photos from a trip to an aquarium. During the process my photo organizing software (iPhoto) runs a facial matching software program to find faces in my pictures and then asks me "Is this Bob?" etc. If the photo is of the person in question then I click "yes" and the software tags the photo with that persons name. This feature is supposed to save me time and let me search through my photos more quickly. Though many times the software is correct there are sometimes when it makes an error by matching the wrong name to face or goes off the deep end completely. While cataloging the photos from the aquarium it pulls up a picture of a seahorse amidst an entire field of corral. Amidst the corral the software has sketched a box and indicates that it believes a face is hidden amongst the corral. The parameters behind the software read the information about the photo as including a face, but in reality it just happened to be a bit of corral that fit the profile. Humans may be less precise and more variable than an auto-tagging program, but they are capable of thinking outside the given parameters. A human being is not bound by those same rules that make the software think a human face is hiding in the corral. That's why a hybrid model works best.