Google Image Labeler
Google just launched a new tool this weekend - the Google Image Labeler. This time they are concentrating on their core competence again: searching. The Image Labeler was created to improve their image search.
The interesting thing about it is, that instead of concentrating on technology like context analysis, image and text recognition or artificial intelligence, they built a tool to let people categorize images. The idea is simple – to motivate people to contribute, they designed a simple game. You get paired with a random other player and get 90 seconds time. A random thumbnail from Google's image catalogue is shown to both of you and you can enter tags for it. You get points whenever you and your partner are entering the same term, then the next image is shown. On some already well-tagged images you are not allowed to enter certain tags (shown as off-limits), so the game might get harder over the time.
I really like the idea of using games to gather otherwise hard to get data. I already heard the term “bionical software” for it somewhere and I guess we'll see more of it in the future.
However there are some things that might go wrong with the way Google Image Labeler works currently. The tags people will most probably match in are very simple ones. For example I might recognize Charles Darwin on a picture but my partner won't, so the tag we will agree on is “man” – not very specific or helpful in a search for Charles Darwin. The other problem you can notice in the current highscore:
You can easily imagine how the first place was reached (2600 points are 26 matches in 90 seconds or 3.4 per second). This however is probably leveled out putting the same image multiple times into labeling and gathering a broader data, but such a skewed highscore could decrease the overall fun of the game which would result in a decline of players.
Anyway, it's pretty interesting and I'm looking forward what might come next.