Google really does come up with some interesting things.
To summarise, it sounds like google has quite a few things in the pipeline to liven up their image search. The first of these, is google image labeler.
Basically, what happens is that you have a certain amount of time (about two minutes) and it’s like a gameshow. You get partnered with someone else, and as images appear, you have to use words to describe them.
If you both pick the same word, then it’s considered a “match” and you get points (which aren’t useful for anything as far as I can tell).
It’s fairly addictive, and an interesting exercise in psychology.
Let’s take this image that came up in my last round:
Now, my first thought was “I wonder whether my partner is from the US.”
I decided I’d kick off with a couple of fairly politically correct answers and try “muslim” and “arabic”.
“Hmm, no match. Ok, let’s try the American response.”
Still no match. At this point with every picture, you start to try and think of the lowest, most generic attribute of the image. “Red”. Matched!
As it turns out, my partner guessed; diphosphonate, entrepreneurialism and red.
So, what do I think of the system? It seems pretty cool, but what always happen is that you drop to guessing extremely generic terms just to get a match. For instance a picture of a city map will always be matched as “map” which isn’t going to help anyone searching for ‘Auckland Map’. And it extends to anything, an iconic Indian building will match as “building” and a famous painting may match as “painting”, “art” of “person”.
I think this idea has it’s merits, but detailed categorising, and matching of people with categories that they have knowledge of would make the system a lot more useful.