Google Takes On ESP Game, But Needs Some Marketing Help
from the giving-credit-where-credit-is-due? dept
A few years ago, one of the researchers behind the idea of captchas (the little distorted word images you often have to decipher to open accounts or post comments on sites) took his idea of human powered computing a step further creating the ESP Game. It pairs up two anonymous individuals and shows them an image. Each side has to come up with words to describe the image, trying to match one of the words the other side has come up with. You get points if you match what the other person wrote (hence “ESP”). The end result is the researcher gets very accurate labels of a large database of photos — without having to pay anyone for the labor. The game has been a huge hit, continuing to stay popular for years, allowing the researcher, Luis von Ahn, to build just such a powerful database of images with descriptive words. von Ahn has since gone on and created other games as well, all with the same type of goal: getting free labor by getting people to play games and not even realize they’re providing free labor.
A little over a month ago, von Ahn gave a very entertaining talk on the Google campus. In that talk, he mentioned that if you could just hook his game up to Google images, and get 5,000 simultaneous players, every image in Google’s index would be labeled in two months. That’s not that many people and that’s not long at all. So, it’s no surprise that it only took a month for someone from Google (in that room, we assume) to build their very own version of the ESP Game — though they don’t seem to credit Luis von Ahn anywhere. It’s nearly an identical clone, so it’s not so surprising that it’s just as addictive as the ESP Game. There’s just one issue. They’ve decided to call it Google Image Labeler — which doesn’t quite make you want to rush over. The ESP Game at least pretends to be about ESP and makes it look fun. The Google version doesn’t seem to try to play up the fun aspect at all. It may very well get plenty of users, and may make Google’s image search much more powerful in just a couple of months (which would be great), but they should have at least allowed someone from marketing to jazz up the game aspect of it a bit. Update: As noted in the comments it appears Google licensed the game from von Ahn, which is pretty much what we expected. Though, we’re still surprised with how they’re marketing the “game.”
Comments on “Google Takes On ESP Game, But Needs Some Marketing Help”
Ripoff?
I thought it was a blatant ripoff as well until I came across Danny Sullivan’s post on it. Turns out Google has officially licensed it:
http://blog.searchenginewatch.com/blog/060901-094309
Re: Ripoff?
Yeah, it was so blatant, I figured it must have been done with some agreement between Google and von Ahn… but I was surprised that he wasn’t credited on the page.
Google blew it
Google blew it with this one. I tried it, saw dozens of images of dialogs, factory floors, charts, generic people. What are they thinking? The vast majority of these images are the crap that clutter up image searches. So now you have people matching very generic terms to very generic images… and that’s supposed to improve things somehow? I don’t see it happening unless they maintain a database of common and generic terms and invalidate any matches that fall in it.
Google needs to show random people a page of images and the search terms that brought them up and let the people select the ones that best match. Either that or match people with specific knowledge to the appropriate images so they can apply specific terms.
Of course, you’d think google could narrow that a bit themselves by mining what images people select when they search to begin with.
Re: Google blew it
If you subset the images that people are labelling, how do you integrate the gathered information into the image search? Do you exclude the unlabelled ‘generic’ stuff, or just throw it in as ‘clutter’ just in case people want a generic image?
Don’t forget, the images are only labelled when a pair matches, so for at least 2 people in the world (within a number of seconds) were reminded of the same thing by the image.
I had one of a couple of business-looking type men and the word that eventually matched was “tie”. Sure, it’s not “2 business men in suits”, but it’s still useful.
Mike, Schmike!
Who is this “Mike”? He sure writes a lot of these articles. I wonder if It hurts to be that smart?
What about other items could those be labled. I’m guessing this is as much a way to figure out hwo to automatically tag than for fun or image use.
And advertising what its “true” purpose is has totally ruined it. Now everyone will just be using nonsense, or blatantly mislabelling the photos with objectionable and silly things, just to mess up what they are trying to achieve. I relate it to the news having reports of what the military is planning to do over seas…why bother…they know what we are up to.
Re: Re:
but it doesn’t even count it unless you match it with your partner, so unless you and your partner both come up with the same random word, it doesn’t really matter what you type
Re: Re: random words? or robot words
The words might be random, unless your two robots play each other.
http://www.jimschrempp.com/features/computer/googleimagelabeler.htm
The ESP game is mildly entertaining and ferociously irritating. The interface does not work properly (FF or IE), the images are (as mentioned above) generic and lame thumbnails. Half the time, I could not even decipher what I was looking at. Hmmm, it looks like a factory, or maybe Margaret Thatcher’s face. Also, how do we know that this will even work? I mean, there is the old addage; “Two wrongs don’t make a right”. So you pair up two morons who match up a bunch of wrong keywords – entirely possible. Then you end up with a database full of junk that you must clean up – I smell labor capital a’ burning. God, if there is anyone that I would not want assigning tags to my images, it would be the userbase – no offense.
boring
I just gave it a quick go. It was mildly entertaining for about 4 minutes. Eventually I got paired up with someone who was on the same wavelength as me and we pwned.
For the people saying that the generic labels and collective unintelligence will only serve to clutter search results, I’d like to think what Google’s algorithms do with the data will be a bit more sophisticated than find two matches and insert it into image search results.
I am glad to see that google has jumped on this one i think it has alot of potential. The google version in my opinion is a great improvment from the original version of the ESP game. I wish the google team much luck in all there indeveres!
how do i convince ppl i have very very high ESP
yes this is true i really do have high ESP(really really high)… can anyone tell me a website to go to to help me with this.. i can do all sorts of cool things with my ESP..believe it or not i make ppl on tv feel what i feel like if i laugh i can make them laugh or make them feel any emotion.. i dont wanna waste my gift please help me………………