ChaCha uses humans for search, but they’re no longer the only ones. Some search problems are a long way off from being solved, and rather than wait for technology to catch up, Google is using humans to label images for its image search.

Consensus Labeling

Google uses humans by making it an image-labeling game—you and a partner label the same image at the same time. You only earn points and move on to the next image when you and your partner give the image the same label.

I must admit, it is fun to try to label images quickly and see what kind of score you can get. At times, it is frustrating because you and your partner see the same image so differently. You can pass to the next image, and hopefully that will help you to keep earning more points. Google certainly benefits from having many, many images labeled quickly.

Labeling Issues

There are a few issues with the quick consensus labeling solution. One is quality—if you are hurrying, images usually end up labeled with one word. A cover of a book about “the darkness within” got labeled “red” instead of “depression,” “psychology,” or “emotional exploration.” Forcing a consensus also tends to limit the length and complexity of labels.

Another problem is difficult images. There are images that are abstract, hard to see, or complex and that would require more time to properly analyze. These images get passed over by humans in a hurry to agree with their labeling partner.

Human Solution

Even though there are problems, you have to admit that it is certainly more effective to use humans to label pictures than machines. Humans can evaluate the entire picture at once, make associations, and use emotions. “Holiday,” “barbecue,” “graduation,” and “play” are all labels that a machine would have an exceedingly difficult time in coming up with.

Duplicate Labels

So why doesn’t Google follow the tagging phenomenon? There are tons and tons of pictures already categorized, labeled, and even described on Flickr and other photo sharing sites. And a simple game doesn’t have much stickiness—there needs to be a stronger incentive to come back and label hundreds and thousands of pictures. It will be interesting to see how Google and others will solve the image problem, along with the other semantic, meta-data problems of the web.