Here is an idea that touches upon many of my favorite concepts about the subjectivity of image description in metadata. The Descriptive Camera by Matt Richardson is a prototype designed to capture the “aboutness” of images rather than the usual camera generated metadata like date, capture time and location.
This nifty device would garner descriptions via the Mechanical Turk API by Amazon. Essentially the camera would efficiently crowd source descriptive data. Since photo keywords and descriptions are subjective, humans would actually provide the details.
The flaws inherent to this design are the same as any user generated tagging – lack of consistency and vocabulary control, mispellings and misinterpretations. It seems to me that a final step might be to parse the information provided by the humans and cross check it against a controlled vocabulary to turn the data into cleaner searchable keywords.
Still, this is a pretty thought provoking concept!