Google Image Labeler

vandalman writes to tell us that Google is betting on the obsessive compulsive need for many users to see big numbers next to their name with a new beta service called Google Image Labeler. From the description: "You'll be randomly paired with a partner who's online and using the feature. Over a 90-second period, you and your partner will be shown the same set of images and asked to provide as many labels as possible to describe each image you see. When your label matches your partner's label, you'll earn some points and move on to the next image until time runs out. After time expires, you can explore the images you've seen and the websites where those images were found. And we'll show you the points you've earned throughout the session."

Too small pics

[avij]

This is indeed a creative way to enhance the search results. Some of the pictures could be a little bit larger though.. Or some kind of a mouse-over which shows a larger picture.

looks good

[LiquidCoooled]

its just taken me 4 minutes to accumulate ~1000 points, there are people who have accumulated 190000 points.

Thats playing the google game solidly for around 12 hours (less if they are good).

Congrats to those people!

As for myself, I found the image sizes too small, but I suppose we are basing the keywords on first impressions and are expected to come from the image search.

I found myself squinting to see what it was meant to be and wasting time, even if it was just 2x larger (scaled would do, no real need for more data) I would spend time there, its actually quite fun especially since you are aiming to get more than your random competitor.

Oh boy, points

[lurker412]

And those points will get me what?

This is damaging for my eyes

[Anonymous Coward]

There needs to be a way to make the images larger, I can hardly see some of them!

Geez that's addictive

[Mateo_LeFou]

I'm not sure why... Google's brilliance shines through again.

Re:Content Based Image Retrieval

[navarroj]

Because humans are much better at image recognition than computers?

Re:Content Based Image Retrieval

[sparkz]

At the end, it says "Thanks for your contribution. It will help us improve the relevance of image search results so that you and other Google users can quickly and easily find the results you're looking for." Which is better at recognising what's in a picture? A human, who can say "oh look, that's Natalie Portman pouring hot grits down her pants", not a computer which will just say "a person" at best.

Eyes of the Calculor

[Buzz_Litebeer]

Google is now harnessing a distributed operating system, you and me. They use games to get us to essentially program for them and reduce error by having multiple people do the same task and use what answers come out.

Very clever. Of course this was done by Amazon as well I think and I dont know what has come of that effort.

But it really means that they are using the processing power of people to avoid having to create artificial intelligence. And why not? Just use real intelligence from people and let them enjoy it by thinking it is a game!

Scaling the small images

[TubeSteak]

Opera 7.54 (what I have installed on this puter) scales images nicely.

Ctrl + mouse wheel up/down = zoom in or zoom out.

I imagine the most recent 9.x still has that feature.

I can't vouch for this site: http://www.obermair.net/opera/operausben.htm [obermair.net] but it was near the top of Google's results for a no-install version of Opera 9.01

Re:I tried it.

[WiFiBro]

Nice, a beta without a reaction form. Ok google, i know you read this.
1) Do some bloody usability tests with non-cooperating users. if there is no response in 20 secs, block that user and give me a new partner
2) also save non-matching words, I enter brilliant things!
3) the game often gets stuck. give me a bail out button
4) add a response form or forum

and most of all:
5) at the end tell me what the other idiot suggested.

Re:The Tom Sawyer technique

[ElephanTS]

Sure, that's why they're doing it isn't it?

I bet they build a huge DB of all this information and try and develop a neural net type system that's able to classify images. This way they populate the model with data for free. Machine processing of images to find pr0n for instance would be very big business and something that has been tried before with little success. It seems the 'how much pink is in the image' algorithm is not reliable - seriously I read that was tried once at some ISP.

Re:looks good

[cmacb]

"As for myself, I found the image sizes too small, but I suppose we are basing the keywords on first impressions and are expected to come from the image search."

Yes, WAY too small. I was tempted to try typing "postage stamp" or "blurry thing" for the few images I was shown. If I play again I'll have to keep a magnifier handy (or the sooftware equivalent).

Don't wanna complain but...

[IlliniECE]

This will provide google with many labellings.. but I worry about the quality.. Under time pressure, people playing this game will want to give replies they know are so simple anyone else would guess.. I saw a photo of the moon during play, and someone labelled it 'ground'-- true, but not that helpful.

Re:Perhaps you don't understand the game.

[Anonymous Coward]

If they aren't motivated by points, but have nothing better to do than sit there and abuse a service which they aren't required to use, then I sincerely doubt they are the ones that are in a position to "enlighten" anyone.

Re:Oh boy, points

[Kohath]

Was she spending time with your daughter while giving you this lecture?

Maybe you should talk to her about that.

Re:Too small pics , Needs additional improvements

[Adeptus_Luminati]

Not only are the pics too small... (come on Google, with all those PHD's this should be a quick fix), ... but also the game needs some additional serious improvements such as:
1) There needs to be a "NEXT" button. For some images 1 minute and 30 seconds is too much time. I get the feeling a lot of people confuse "pass" with 'moving on to the next picture', so I think a lot of effort is wasted here as I imagine that "pass" means discard the data from both users and move on to next pic.
2) There should be a real time score matching as each person enters labels, this would really motivate players
3) Matching with random players doesn't work in most games because people want to play with others who are either at the same level of skill (in this case also speed - its boring as hell waiting 1 minute for your 8 year old (or 80 year old) partner to type in 1 label). Or allow people to do international competitions. I.E. Canada vs. USA or whatever.
4) I hope labels get spell checked before they are compared, otherwise there's a lot of misses
5) The label typing box should be smaller. For the first 3 or 4 times I tried it, I intuitively typed many labels separated by commas, only later to realize that all those sequencial words only counted as 1 label.
6) I have two internet connections with two different providers, on both PCs, the next image to come up took several seconds to show up... what's up with that Oh Google of infinite bandwidth?

This thing needs some serious improvements before it becomes addicting... right now its closer to 'lame' & if your partner is too slow... 'annoying'.

Re:Actually Google Licensed It

[Spezzer]

A lecture given by Von Ahn on Human Computation is available here: http://video.google.com/videoplay?docid=-824646398 0976635143 [google.com]

Lowest common denominator

[Dan East]

Lowest common denominator, that's the quality of results they'll end up with. I might see a picture and label it "bird", where my partner, being an ornithologist, labels its exact scientific and common names. Until he enters "bird" we don't get a match.

Dan East

ingenious

[z_gringo]

Holy crap, that is clever. Where do they think this stuff up? What a great way to get people to work for them for free.

I wish Google would hire me.

Re:I tried it.

[pilkul]

Er, sorry to burst your bubble, but my current record is 15 matches in one 90-second session and I regularly do above 10. I mean, if there are people in the picture you type "people" and if there are trees you type "trees". No deep philosophy of language issues here.

Re:Too small pics

[Jcalaiaro]

As a CMU student, it was a bit startling to find this under the Google domain. This has been a graduate research project for some time now at CMU. Refer to www.espgame.org I would be very interested in knowing how Google was able to do this given the fact that their system is a facsimile of what CMU has to offer.

Re:Google, in search of extra-search

[aGuyNamedJoe]

Um.... They're recording the labels, at least the ones both use. Then when someone looks for pictures with that label, say "Tom Cruise" or "Wedding", they'll know that that image is an answer... They make money by showing people things that are labeled -- usually the labels are in the text on the page, but images aren't text... Pretty clever, I think -- using a game to get people to help them index images. joe

More like a creative way to get work for free...

[caitsith01]

I presume Google has no intention of paying anyone for the time they put into this system.

As such all that this really represents is a way for a large, profit driven corporation to make money by using the time and efforts of a large collection of individuals.

Seems kinda sinister to me. Personally I would like to gather the rewards for my own efforts, not allow some megacorp to do so. But unarguably a very 'creative' way to go about it, to be sure.

Game results in dumb labels

[QuestionsNotAnswers]

After playing for a short while, you realise that there is a common set of words that everyone knows are the best first tries.

Lady, Girl, Man seem to be really common (even if not right) and colours too.

So it soon ends up that pictures are labelled by the words that help you win, rather than the most appropriate words for the image.

Re:Slender Neck Pr0n

[soft_guy]

How are you seeing your partner's guesses?

Re:The Tom Sawyer technique

[soft_guy]

I had the same partner several times. I would think that if you had group of people trying he suggests, you would get a partner who is playing your "penis" strategy fairly often - and it would be really obvious right away if they were (first image, you type penis and get a match. If not, then stop playing that way because your partner isn't one of your friends.)

Re:GAMES

[Pahroza]

While I agree that there should be some bonus time for doing well, I can understand that they don't want to give you too much feedback. The way I see it, the point is to get a good cross section of what people would label an image, not find the 2 people that see each image the same way. If you've got people doing really well together, then they obviously see things the same, and it is probably time to pair you up with someone else.

Re:The whole concept is flawed

[munpfazy]

I feel the concept is flawed because this competition leads to subpar results. In order to score, everybody tries to dumb down what he sees in the hope his "friend" will do the same.

My anecdotal experience suggests the same. I played several rounds, and in almost every case the winning term was the most general and least informative: "guy," "people," "cake," "cigarette" (for a scene in which one person just happened to be smoking). Hard to believe you could find useful data in a sea of terms like that, unless of course google actually saves all the more specific terms that we guess and uses those for their analysis, and the real time matching is primarily just used to keep users interested. Then again, that's probably not unlikely.

On a barely related topic, I had a somewhat startling experience. The first time I tried it, the third image was from one of my collaborator's websites, and it was a picture that I'm very familiar with: a shot of three guys in red parkas at the South Pole station standing under a crane installing the DASI telescope. I put in a bunch of accurate, very specific search terms. As one might expect, there were no matches. Eventually I was reduced to terms like "crane" and "ice," at which point my partner passed.

Now, I can guarantee that I provided a dozen search terms that ought to map to that image. But, since there are probably only a hundred people in the world who would know what the subject of that picture was, chances are my extremely accurate and useful descriptions will be swamped by terms like "guys" and "red." It's probably a long shot to assume "crane" and "Antarctica" make it into the top ten.

There seems to be a fundamental limit to what is possible with this sort of system: it can only be as good at identifying images as the average person. But, the average person is probably far less good at identifying images than someone who is actually searching for a specific image.

No one (worth mentioning) searches for a "guy" or a "molecule" or a "plant." They search for "Guy Lombardo" or "Sulfur Hexafluoride" or "Sago Palm."

Re:More like a creative way to get work for free..

[ultranova]

Seems kinda sinister to me. Personally I would like to gather the rewards for my own efforts, not allow some megacorp to do so.

Careful, now; that attitude makes you sound like a communist. It is the basic idea behind the Communist Manifesto: workers should reap the benefits of their own efforts, this requires that everyone owns the means of production he uses, and since a factory can't be operated by a single person alone, it should be owned communally by all the workers working there who can then share the profits between themselves instead of having a rich capitalist - megacorp in these times - pocket them.

Your desire to gain the benefit from your own work is, therefore, completely un-American. The capitalist way of doing things is that you do the work, the investors get the profits, and you get to compete with the Indians for who can survive with the lowest wage. Since India has a much lower cost of living, you're going to lose. Since your economy is bleeding money to India, the buying power of the people of your country is going to shrink, making it more neccessary for corporations to try to cut costs by hiring more Indians, and the situation is going to get worse and worse.

Sure makes you glad to live in a capitalist country, doesn't it ? And sure makes this post likely to be modded down by free-market fundamentalists who don't quite understand that communism ("people should own the means of production they use, and if a particular means needs more than one people to operate, then those people should own it communally") is not exclusive to free market ("everyone is free to produce what they want and trade with whoever they will").

Mod me down, but I'm still right.