The ESP Game is an ongoing project of Carnegie Melon University in Pittsburgh, Pennsylvania. The great thing about this project is that it is both a fun game for the players, and is a unique research project which, in the future, may be useful to all Internet users. The game is located at

The Game


The game requires two players, each with an Internet connection and a Java-enabled web browser (it runs in a Java applet).

Game Play

Each time you play the game, you are automatically paired up with an anonymous random player who is logged into the site (you never know who you are playing with/against). Each of you are displayed the same random image from the web. Your task is to try to guess a word that your partner would use to describe the image. The catch is that there are a series of taboo words that you cannot use. You also cannot enter other forms of the word (singular/plural, past/present tense, etc), or certain words related to one of the taboos (e.g., if one of the taboo words is "green", you can't enter any other color either). You get 2 minutes and 30 seconds to try to match your answers on up to 15 images.

Since it is a timed game, if you and your partner cannot agree on a word, you have the option to pass. If either partner decides to pass, a message will be displayed to the other player. Both players must choose to pass to actually skip an image. Keep in mind that you cannot earn points for images that you pass on.


You earn points each time you and your partner agree on a word. The more taboo words associated with an image, the more points that image is worth. In addition, you get 350 bonus points for matching 5 images, 850 points for matching 10 images, and 2000 bonus points for matching all 15 of the images in the 2.5 minutes. Your cumulative score is saved in your profile, and the highest scorers of all time are displayed on the homepage.


The basic strategy is to simply type as many words that describe the picture as fast as you can, and hope that your partner matches at least one of them. Foreground and background colors contained in the image are always good choices. If the picture contains any words itself, you'll probably want to try all of them. Most people focus first on the objects in the foreground. Look at the background for descriptive words as well. You are not penalized for entering non-matching words, so the faster you type the better.

The game can be much more difficult than it sounds, so some creativity must be used when making your guesses. For example, if you are shown a picture of a black dog, your first instinct might be to type "dog" or "black" - but guess what? Those are probably both taboo words. You might instead choose "puppy" or "Labrador". If these don't work, you might try entering what the dog is sitting on. Is he wearing a collar? Is he growling? Keep typing those keywords away. If you've spend more than 10-15 seconds, you should probably pass.


While CMU doesn't tell you how the images are picked for the game, it can be assumed that they steer clear of pornography and other objectionable sites. Regardless, the images are chosen randomly from billions of potential pages, so images that you might find offensive may be displayed (I have never seen anything even remotely questionable). If this happens there is a bright orange "Flag" button you can use to notify the site of the bad image.

Game Over

Once you have finished the game, you will be given the opportunity to look over the images that you and your partner didn't agree on. You can cycle through each of them and see one of the words that your partner guessed that you didn't. Unfortunately, you are not shown all of his/her guesses, just one. You can then choose to play again.

More than a Game

While it's a fun and sometimes addicting game, the real goal behind this project is to give a label to all images on the web. A database of human-provided descriptions of images can make searching for images on the web much easier than it is today. In addition, it can be used to help people with visual disabilities to have a better experience on the web. As stated above, attempts are made to filter objectionable materials from the game play; however the act of allowing users to flag certain images is also helpful for the research being conducted with this project.

While the HTML <img> tag does provide both a title and alt-text attribute, they are not always properly used or described. The nature of the game forces users to quickly describe their first impressions of the pictures which can be very important information for web developers and designers to consider.

At the time of this writeup, the project had already helped to label over 2 700 000 (2.7 million) images.

Log in or register to write something here or to contact authors.