Google Image Labeler Game

If you haven’t tried out that Google’s image labeling game, I beg that you not try it. You have 90 seconds for you and a partner to agree on words to describe as many images as possible. At first you think, “90 seconds? That’s nothing.” So you play. And you play again. And then you look up and it’s 4pm and you’re almost on the leaderboard, so you play a little more. I beg of you–stay the hell away from this game. For god’s sake, I mean it. And if you beat my record of eight images in 90 seconds, please don’t post it here to brag. Not even a little bit.

(Note, this game is inspired by the ESP Game created by Luis von Ahn, and I believe Google licensed the game from Carnegie Mellon.)

53 Responses to Google Image Labeler Game (Leave a comment)

  1. It’s too late! I already tried it! I’m hooked already, it’s really addictive. My record is 11 images (sorry, couldn’t resist).

  2. Got 10 in 90 seconds man 🙂

  3. my partners all stink.

    when a book cover shows up, and you don’t type the name of the book?? but interesting idea.. i can see how this will help image search already..

    the possibilities of what can be done with this data are neat!

  4. I got to the 38th place in the all-time ranking but then i had to go to sleep. The next day i realised i need a full-time job to be first 🙂

  5. Fun, but half the time the images are too small for me to be able to identify them.

  6. testing math add, since my nicely written comment just disappeared into the ether..

  7. It’s ok *as a game*. However, as a method of labelling images (and I find the Google antipathy towards ‘tag’ interesting), it’s got little, if any value at all. In order to compete on speed and to boost the ego with a good score it’s necessary – obviously – to get as many points as possible. Which means that you go for the lowest possible denominator. If there are words on the screen (and they’re not banned), people will use those, irrespective of their appropriateness.

    If there’s a photograph of a dog with the word ‘cat’ underneath, if ‘dog’ is a word that they can’t use, they’ll go for cat, since it’s something both players can use and, in all probability will be one they can expect their partner to use. Consequently when you get images that are generally one specific colour there’s a fair bet that instead of using ‘lawn’ or ‘grass’ or ‘yard’ they’ll both go for ‘green’ instead. Sunrise? Sunset? Doesn’t matter – go for ‘orange’ and you’ll probably get the points.

    As a game it’s fine. As a method of using humans to tag images *correctly* it sucks, big time. But unfortunately, that’s something that we’re getting used to with Google. And, since I’m here, when is Google going to fix the ‘limit by date’ function? It’s been broken for months now, and quite frankly I’d prefer it if Google concentrated on making the basics work rather than trying to grab headlines with silly games.

    Thanks.

  8. The game is sure fun to play when you play with someone who is not STUPID. Sometimes I have to quit after 30 seconds because the other person only guesses 2 or 3 and they are NOT RIGHT and it’s a flower, or a lake or something easy!!

    Anyway, great idea on googles half.

  9. sarahsomethingorother and I just got 12 :-~

  10. So I finally gave in and typed up my comment again, only to fail the math filter again. Is there a time-limit or something also? I’m fairly sure that 7 and 9 equal 16…

    (p.s. I copied this comment in ‘case I fail the math filter some more…)

  11. (typing super-fast) the ESP Game is awesome, I lost days to it at one point, as well as to PeekaBoom (created by the same guy).

    I wish Google had picked a better name, but at least my Google image search results will get better. 🙂

    (remembered to copy this time in case I fail math AGAIN)

  12. How exactly will Google use this data wrt Google Images? Surely it will result in a bunch of generic tags placed on images, rather than specific ones, won’t it?

  13. Have you been so caught up in playing this that you forgot to actually publish the post – it seems to be dated 1st Sept, yet only appeared today on the 7th!

  14. Tim L. Walker,

    I think it is a brilliant idea. The generic tags are great. How often do you wish you could find ‘a man wearing a blue shirt, and a woman holding her arms in the air’.

    okay, not often, but you get the idea. of course getty images already does this, but great way to get people to help you for free.

  15. See the bigger picture. Lisa Barone wrote a great article on Google’s new image labeler the other day.

    Snippet:

    “Once Google has a set of images that are labeled with a high signal-to-noise ratio, this would be an ideal corpus of material for their newly acquired “machine vision” technology to train on. And once Google’s computers are trained to interpret images and video, then Google’s algorithms can do for images and video what they’ve done for text — organize it and make it universally accessible and useful.”

    http://www.bruceclay.com/blog/archives/2006/09/google_image_la.html

  16. I tried it several times but abandoned quite fast.
    The thumbnails are too small to make out any detail and labeling thumbnails is a tough job!
    Is there a way to have the images displayed in their original size?

  17. It’s interesting to see your partner guesses as you hover your mouse on the image at the results page. It seems funny that we describe things differently based on what part of the photo we notice first. Hope this helps Google derive more sensible image search results.

  18. Matt,

    why is your post dt “September 1, 2006 @ 4:53 pm ” and published only on 7th ??

    What am I missing here ?

  19. cool, got 9 with a guest.

  20. Finally found a good partner and got 9 🙂

  21. if I am free,i will try this type of game,I heared this from gfans.org,nice to play:)

  22. Really great game and in the same time it helps to label images.

  23. That’s way too addicting. I tried it 3 times and topped out with 1100. Thanks for showing me a new way to waste time during my day.

  24. WATCH out guys …. sometimes the image are offensive, I was playing with my wife and I see this cartoon where the gal is sitting down n guy is standing obviously they r nude so i had to guess … and my partner said “sex” eventhough i was more precise 🙂 …. Google how can we avoid showing such images or filter ..??

    Maybe thats why u guys came up with this game … lol

  25. top ranking of the day… 12 in 90s

  26. I scored 12 today too ..ha ha ha 🙂
    I discovered there is this labelling thing today and needless to say I am addicted. I agree with what Bradley had written that people tend to over-generalize objects, events and things. Tagging images with labels like man and woman easily matches my partners and I dislike doing that even though I like to score more…Thanks but no thanks to Google..

  27. Matt,

    You guys probably already know this: There are ALREADY attempts to spam the image labeler. I was just now getting strange labels from players such as “carcinoma, accretion, bequeathing” etc. – for images that had absolutely nothing to do with these relatively rare usage words.

  28. i heard jo whiley of bbc radio 1 talking about this and thought i would have a look, i wish i hadnt, i was nearly late picking my kids up from school as im addicted. ezcept for coming across the spam numerous times today, what is the point
    loving it

  29. To Nas Raja:

    Nas – it’s funny to see how so double-standard and hypocritical your opinion is… On one hand, if Google decides to exploit millions of people by using a tool that automates the costly annotation data collection process without paying a dime – it is fine and dandy… However, if someone else starts using a tool that automates the one used by Google – it is immediately called spam and abuse by Mr/Mrs Nas Raja! Do you really think the the spammers would want to use just a simple set of easily filterable keywords?…

    Congratulations, Nas – you are a golden standard of fair, unbiased and objective judgement towards Google and everybody else!

    Tobora

  30. Tobora,
    Irritated at being found out so fast and getting wiki’ed to boot, are we now? Care to explain exactly what positive use would come out of trying to spam the Image Labeler with high-paying AdSense keywords except – well, except to improve the relevancy of your pages to illegally pull in even more AdSense traffic? A flower being labeled carcinoma is way beyond preposterous! Here’s just one example of a website set up for the explicit purpose of generating AdSense traffic for high-paying keywords such as carcinoma. I rest my case.

    Do realize though, that a simple captcha would put automated spamming scripts for Google Image Labeler in their place. I guess it hasn’t gotten out of hand yet for Google to do that. If so, my advice would be to lay low and look up the job description for the guy whose blog you posted this comment on. You’re playing with fire here. 🙂

    Oh, and – it’s “Mr.” Nas Raja, thank you!

  31. Tobora,
    Irritated at being found out so fast and getting wiki’ed to boot, are we now? Care to explain exactly what positive use would come out of trying to spam the Image Labeler with high-paying AdSense keywords except – well, except to improve the relevancy of your pages to illegally pull in even more AdSense traffic? In any case, a flower being labeled carcinoma is way beyond preposterous! There are easily google-able examples of websites, set up for the explicit purpose of generating AdSense traffic, using high-paying medical terms like carcinoma, accretion, etc. to do so. I rest my case.

    Do realize though, that a simple captcha would put automated spamming scripts for Google Image Labeler in their place. I guess it hasn’t gotten out of hand yet for Google to do that. If so, my advice would be to lay low and look up the job description for the guy whose blog you posted this comment on. You’re playing with fire here. 🙂

    Oh, and – it’s “Mr.” Nas Raja, thank you!

  32. Tobora,

    Google provides the public with entertainment in the form of a game, and is also improving its image search engine that it provides to everyone. Internet users win twice.

    The notion that Google is “exploit millions of people” is shear stupidity. People are playing the game of their own free will.

    People who exploit the game for personal benefit are ruining the image search results for everyone.

  33. There seems to be a pattern of people typing either “diphosphonate” or “accretion” for every single image. Anyone else have this problem?

  34. As a game this is quite entertaining, but it will be unfortunate if results hold much weight in image search results. As has been mentioned, even the most thoughtful results will be very general. In the timed context of the game it is **not possible** to give the images the proper detail that they require to recieve truly precise (and useful) names.

    It relies on both people not being ignorant of the image. But we wil be in most cases. I’ve seen some important-looking buildings I didn’t recognize. So they are ‘buildings’. And that electronic gadget that looked like a pen. Wasn’t a pen, but that’s all I could think of … and bing…match. (my apologies) To be labeled usefully one could quickly look at the page and see that it was a “SuperFlash laser-pointer XL-2c”.

    For me Google has a truly high quality of software/services record compared to the other big dogs and using the results of ‘games’ like this would work to lower that quality…and may set precedents and be copied by the other engines.

    My $.02

    – Dan Shortell

  35. I just started playing the image labeler type game today and its highly addictive, until you keep getting stuck with a player who puts bequeathing, googley, forbearance, and other words that do not fit what so ever for a pic, making it hard to get any real points. I hope google does something about this fast! and to the person who said “google is exploit millions of people” thats just nonsense, how is it exploiting us if they tell us right on the page what the purpose of the labeler is for? I think its genius for them to turn time consuming work into a fun game and since I use google images all the time its a huge plus to know that I am helping make it even better.

  36. I ran into this problem, as well. Someone labeled an image “diphosphonate” and it completely did not apply. Ironically, a previous image had the word “diphosphonate” as one of the off-limits words. It also didn’t apply. Go figure. Sure, this COULD be spammers. I have another theory, though. Perhaps, in the event there are not enough players available to give you a match, Google’s system sets you up with a bot? And, perhaps, this bot tends to throw in random words in addition to relevant ones to be sure you aren’t guaranteed a high score? Just a theory.

    (On a side note, the spammers would only win IF one spammer was teamed up with another spammer and IF they were completely synchronized with the labels given. Since we have no control over how the player matching is done, the ball is in Google’s court to prevent this using many various methods. Also, if labels are only considered appropriate when many unique players have matched the same terms, this allows the normal players to outnumber the spammers.)

  37. Regarding the Image Labeler ‘game’, it seem somewhat obvious that the indiscriminate use of irregular words such as diphosphonate, congenita, accretion, googley, forbearance, carcinoma, bequeathing, etc. are not accidental. Therefore, if there is some purpose behind this use of rare words, what IS that purpose ?
    I don’t subscribe to the spamming explanation as Google surely knows what is happening, and could easily block these spammers and words long ago. It then appears that these incidences of junk words from our ‘partner’ occur with the blessing or design of Google.
    If you notice, two things occur with regularity: as soon as an image appears, it says my ‘partner’ has already inputted 1-2 ‘guesses’. When you see those guesses later, they are usually words like diphosphonate or bequeathing….who can type THAT fast ?
    Also, if you check your partner’s ‘guesses’ afterwards, AND there are junk words, those words are always the first guesses and not the final guesses.
    So, what does it all mean ? My own theory is that we are always or almost always playing a bot, OR, that the program itself inserts these ‘guesses’ for us without our knowledge. Either way, we are REACTING, and those reactions determine the way we play subsequent
    games. And THAT is the information that Google is REALLY trying to gather. After all, I’ve seen some of these same images for days; they should be quite well labeled already.
    In the end, we’ll probably find that we were helping program some AI recognition function for GOOGLE, and we had FUN doing it !

    PS Microsoft WORD spell-check does not recognize diphosphonate, congenita, or googley. Just when I was getting used to using them !

    PPS I’m going to start inserting the word ‘miscellany’ ( just cut and paste ) into as many images that I can. If other do the same, maybe we can ‘teach’ the bot something.

  38. Labeller, addicted, accretion, annoyed, skeptical, sleeplessness…

    Ok, so I can’t even speak in real sentences anymore. Just before my grandfather started down the path of alzheimers his letters were actually very similar.

    But anyway, although I am strangley addicted to this “game”, the accretion thing bugs me a lot. I’m afraid that it is something automated beyond our own recognition. If I were able to see my own responses, would it show that I had actually submitted some of these random words? If anyone ever plays against me, let me know.

    And what it’s all some kind of sociological experiment designed by some phd candidate to test some aspect of human interaction/culture/manipulation, etc? Maybe the game master is waiting to see how long it takes for some percentage of players to begin using accretion as their first word choice. Copy-Paste-Enter a la miscellany above, then pick a real word or two. Maybe once accretion, bequeath, disphosphofalalala have been used on every single image, then people could begin playing an actual word game.

    Here are some other frustrations that seem “bot” related. Have you noticed that you may come across an image that could be associated with some pretty obvious words (sky, water, green, etc) and the other user stops at three guesses? …and nothing is a match! Yeah, 100 points.

    I want to speak to Mr. Mellon…

    In fact, although Image Labeller doesn’t have it’s own category on submitting comments/questions/ideas, I think we should all get on http://www.google.com/support/bin/request.py and report the issue. Because basically what it boils down to is the game is kind of fun otherwise none of us woudl be here, I like to think that I’m participating in something, and this accretion thing is annoying. I suppose I don’t look at other people’s reponses, but come on… that’s like God giving you a ___ and telling you not to play with it!

  39. Ok, so I just played the next hand and there were two guesses waiting for me right off the bot… I mean, “bat”. The image was the following…

    http://www.ramou.net/ti/im/Chine-1963-T57-2b.jpg

    I guessed several obvious words with no luck, and then did a copy-paste job on the following (diphosphonate, congenita, accretion, googley, forbearance, carcinoma, bequeathing). Bequeathing won!

    Partner’s guesses:
    bequeathing, yellow, stamp

    But again, the guessing stopped at three. A normal person seeing someone else put in at least 20 words must have been wondering, “hmm, should I try to come up with something else?”

    Oh, by the way… it made me feel icky on the inside to submit all of those words.

    I tried “miscellany” too!! 😛

    Also, I think there sould be a few basic options too…
    – Play against other registered users / Play against anyone
    – View pictures that have (no,1,2,3,4 or fewer) or (5 or more) words assigned

  40. Here’s a funny little quirk also…

    On the Image Labeller homepage (http://images.google.com/imagelabeler/) don’t start the game, but just sit there and hit refresh (F5) over and over and watch the All-time Top Contributors.

  41. my own theory is that the random-seeming words replace misspelled words entered by one of the partners. given the fast typing required of the game, one would expect to see a high incidence of misspellings in the labels, but i’m not sure i’ve ever seen a single misspelling. on the last round that i played, a picture of a logo that said “fnord” showed up, and my partner’s labels showed up as “carcinoma” and “car.” i would think that the first label he made was “fnord.”

    maybe it’s set that any word not found in a spell check will change into one of these words, so that misspelled words don’t ever become tags. it would make sense that the words are so purposely random as to avoid an accidental match.

  42. It would seem likely that google will use the data accumulated through this game to feed into a neural network for it to train on image recognition.

  43. i haven’t read this entire thread, so maybe someone already theorized this, but my best guess is that when words such as carcinoma and accretion show up, it means you are actually partnered with a computer. i think that sometimes you are partnered with a real person, and sometimes you are partnered with a computer. when you are partnered with a computer, it can only accurately guess those images that it has seen before. for the images it has not seen before, it marks them with guesses such as carcinoma, accretion, bequethed or googly, so that the programming can benchmark the computer’s guesses against those of the human partner. just a thought.

  44. I am not so inclined to believe in the possibility that Google is programming their own bots to make the game less fun and their search tag training less effective while running some kind of ‘Mad Scientist” experiment on all of our brains. However, I would entertain the notion that a competitor, or more likely just some cyber-dork out for a thrill, has designed a play-bot meant to ruin the fun of the game as well as messing with the results of this obviously large effort on the part of Google.

    And on the other hand, I would think that Google could, in the matter of a few games, figure out which IP addresses were providing the false answers and ban their systems from the page. Also, i think Google has the power to easily disregard all data or all games in which one of those very odd words appear… (which is most likely WHY they are odd…nobody would ever REALLY guess them probably ever in the course of a game, so they could easily delete all occurences with a very low false positive factor) …so it *could* be a creepy experiment to see how people react to the word “accretion “, but I doubt it.

  45. To the person who said they paired us with bots due to a lack of players, thats doesn’t seem right. There are thousands of people who play this game!

  46. Oh yeah, to add another name to the list, my partners always seem to put ‘green’.

  47. what the hell are you people doing all day.True i have played image labeler but its not a good game or a good way to label anything.please no more about this crappy game.Why dont you guys go out for a walk or something ok goodbye o and enjoy the eclipse

  48. The idea for this came from a popular science magazine, i beleive it was marked for 06, and i think that google was right to make a game based off of images

  49. Allyah Richardson

    Man, I haven’t even played this yet beacause of slow AOL, but I heard about it on G4 channel 64 if you have cable.

  50. (I know it seems a too old post to comment on but cant resist saying something)
    I just read this post… I tried this game for the first time today. Seems like it shows images from my image search history. Because I am a fan of the film Godfather(1971) and I often change my desktop wallpaper with some Godfather poster or still. So I often search for these images on google. When I played this game, I found a lot of Godfather images showing up 🙂
    Now thats a great game for me. The only sad fact is that my game partner and me couldnt match up even on one photo!!! Not a tag matched! Thats probably he didnt know who Marlon Brando was or he must have not watched the film

  51. this image labeler is a great ” time killer ” when it is considered as a game. but if goolge really uses the data acquired from this, it sure is going to ruin the image search engine…

    i belive it is no way going to help google to improve its search engines,
    as a funniest part me and my partner scored 140 by giving tag ” teen ” to a famous female celebrity..

  52. I know, old thread…

    I also found some of my “partners” guesses odd. I came to the conclusion that it was a rendezvous strategy. I published my thoughts here:

    http://www.jimschrempp.com/features/computer/googleimagelabeler.htm

    Cheers,
    Jim

  53. i want to collaborate with the image labler.
    is there some ways that i can programmatically submit the tags to google?

css.php