I’ve recently discovered the Google Image Labeller (or Labeler as they call it) and am slightly hooked. However, I remain to be convinced that it’s of any use at all.
It appears to be an attempt to improve image searching by “crowdsourcing” human cataloguing of images. It’s presented as a game. You and a random partner are presented with a series of images which you each label. You score points when you and your partner assign the same labels. Certain labels are “off limits” – presumably these are ones Google already has enough votes for to believe they’re valid. You are given two minutes to do as many images as you can. When the time is up you are shown what labels your partner chose for each image.
This all sounds quite clever, but I do wonder if the labels are going to be any good. Firstly the images presented are too small. In many cases I simply cannot see what the picture is of and so cannot label it meaningfully. Secondly the time element gets in the way. Thirdly your partner can “pass” on an image after which there is no point in labelling it, which can be annoying. Lastly, on seeing some of the labels my partners have chosen I find myself thinking “what is this guy on?”.
Wisdom of crowds? Maybe, but if you do an image search and get crazy answers, perhaps it’s because the crowd is full of village idiots. (Or mavericks, as I believe they’re called in Alaska. Whoops, little bit of politics there.)