Google Image Labeler relies on crowdshop labor

Google Image Labeler

Folks are buzzing about Google’s new time wasting playable Image Labeler. Philipp Lenssen says:

More than a game, for Google this is a way to tag images using human brain power… to improve their image search results. Two people finding the same tag can serve as validation the tag makes sense. I suppose for Google it’s not important that two people find the same keywords at the same time – they can simply let people tag the images and then add any threshold they want (like “4 people must have chosen this tag for it to become a confirmed tag”).

Both Search Engine Watch and TechCrunch made the connection to research conducted by Luis von Ahn at my alma matter that was first blogged about as early as December last year (written up in the Pittsbrugh Post Gazette in August 2005).

According to Danny Sullivan at Search Engine Watch, the Google technology is indeed based on von Ahn’s work:

Yes, Image Labeler is based on my ESP Game, which Google licensed. I’m not employed by Google, however, since I’m a full-time faculty member at Carnegie Mellon.

In my experience, I found the images were often too small to make out clearly, whereas in similar systems like Amazon’s Mechanical Turk, you get much higher resolution photos.

Interestingly, uses a similar but closed system of human tagging to populate its object search. It’s unclear how such a system scales for web wide results unless something like Google or Amazon’s tool find enough widespread pick-up and open up an API to the tagged images.

Author: Chris Messina

Head of West Coast Business Development at Republic. Ever-curious product designer and technologist. Hashtag inventor. Previously: (YC W18), Uber, Google.

3 thoughts on “Google Image Labeler relies on crowdshop labor”

  1. “In my experience, I found the images were often too small to make out clearly”

    What’s the metric for this? Exactly how big should they be, and why?

    Small images load faster, which may help gameplay on slow links. Small images buffer against NSFW or just NappropriateFW, and others have rumored that imagelabeler may replace solitare for some people at work. The labeler images might be the same size as the thumbs at, in which case you’re tagging them at the same level that they can be “read” in the search results.

    I think it’s strange that you’re complaining about image quality in a game that’s almost purely a time-waster for you (until you can benefit from the improved search). And, I hope people will turn to clones of IL that are just as fun, but publish the tag info for everyone. IL might be helping google organize the world’s information, but there seems to be a lot of needless “one-way sucking in of the world’s information” first 😦

  2. My metric is my own experience — and that I had to “skip” too often instead of creating usable labels. At the very least, trying different sizes to explore accuracy should be openly explored.

    When the image quality impacts both my enjoyment and effectiveness at ‘playing the game’ it is, to me, an issue worth addressing.

  3. maybe its my bad luck, but this game is NSFW in my opinion. i have labeled pictures from scenery to the human body, and the human body pictures seem derogatory at best. there should be a button that says inappropriate picture on the side which sets it up for review and then doomed for the filter.

    other than that the game is fun for a time waster, almost as good as bejeweled

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: