@Mercury: Hello! And yes, subjectivity is part of why I was thinking about this — someone who took a photo may not see certain beauty, qualities, interpretation, etc. that others will, and could conceivably add useful tags that way. A pair of "fresh eyes" could add usefulness to searchability.

@kerunix: Aww I will miss the castle. Verily, it is awesome. Really liked the lights and how storybook it looks.

@Day: Last time I checked the Turk, I did see a lot of bloated for-advertising requests on there, which *does* have a place, but I couldn't find more stuff of interest. I have been piqued by mentions of "podcast transcription" tho, since that's (speech recognitions) still something not reliably done by machines, generally speaking.

@Gigs: Great thoughts! Wow, so an intern was actually hired for this. I wonder how common that is at some (non-Flickr) stock photography agencies, too. SOMEONE's got to do it to increase visibility and viability of finding useful images, I'd think! Anothe profession where it pays to be a fast typer too. Re: "non native speaker", that is a good point, but from an anthropological perspective, it's intriguing to see how different people around the world see the same thing. I wouldn't want embarrassing typos in my tags, tho!

I had written a nifty web app to manage the raster digital assets (photos, stock art, commissioned art), but there was no easy way to search it. We hired an intern to tag every image with keywords.. about 20,000 or so. She surprisingly tore through them in about 2 weeks.

We were paying her about $9/hour probably, so lets say $1000 total cost for 80 hours labor, once you figure in other overheads on 80 hours labor. This works out to 5 cents per image.

I figure this is about your price point if you were do this in a distributed way. You can't count on international labor since keywords are somewhat country specific, and a non native speaker might not get all the nuances.

You'd also probably need some redundancy, asking the same image to be done several times by several different people, if you really wanted good descriptions… sorta like recaptcha does to throw out bogus results.