Computers sometimes cannot complete tasks or solve problems that are done easily by humans. Humans can help solve these computational problems by performing certain conceptual and perceptual tasks, such as identifying objects in photos and transcribing handwritten text, that can in turn provide data used to train computers. Sometimes these tasks can be repetitive, mundane, and as a result, people are not very motivated to participate in them. But what if games were used to collect data, solve problems, and make these processes more engaging for participants?
Luis von Ahn, one of the founders of Duolingo (a language learning application) and the person who coined the phrase “Game With a Purpose” (GWAP), thinks of games as efficient algorithms- a more effective algorithm (game) will be more interesting and fun to participants, and thus collect more data than a less effective one. Inspired by a number of successful GWAPs, Lingo Boingo is a web portal that brings together a collection of games and gamified activities to collect data and annotations on concepts such as anaphora, language recognition, multi-word expressions, word relationships, lexical networks, and more. Some of the language games available on the web portal include Phrase Detectives, TileAttack, NameThatLanguage, and Lingotorium.
The ESP Game and Peekaboom: Some of the First GWAPs
Two early examples of GWAPs developed by von Ahn and colleagues at Carnegie Mellon University include the ESP Game and Peekaboom. There are millions of images on the internet without accurate labeling. In order to build systems that can automatically identify and label images, large sets of training data need to be created by human effort. However, getting humans to review and label large numbers of images is time consuming and costly. GWAPs, or human-based computation games, actually played a part in making this process easier. First, a game called the ESP game was developed. The ESP game is a simple online game that pairs players together who don’t know each other and can’t communicate. However, they can both see the same image and the object of the game is to guess what their partner will label the image as. If both participants type the same thing, the image is cleared and a new image will appear. A maximum of 15 images are presented in 2.5 min, and participants work together to label as many as they can. Some of the images have “taboo words”, words players can’t use to label it, which makes it harder to solve. Results of the game were analyzed and it was found that when two players agree on a label, that label is usually very accurate for the image. Additionally, within a few months of its release, the ESP game was able to collect more than 10 million image labels.
The ESP game helped identify objects and create labels, but it wasn’t able to elicit information about where in the image the object was. That’s where the game “Peekaboom'' came in. In this game, similar to the ESP game, two participants who don’t know each other and can’t communicate work together. One player is assigned the role of “Peek”, and the other player is assigned the role of “Boom”. ”Boom” is in charge of slowly revealing the image while “Peek” guesses what it is. However, ”Boom” only wants to reveal parts of the image that actually contain the object, and this provides location information of the object in the image. “Boom” can also indicate whether “Peek” is close in their guesses to aid them. Players have 4 minutes to work together and complete as many images as they can. Both the ESP game and the Peekaboom game are great examples of providing a fun, gamified approach to a task that ordinarily, humans are not motivated to sit down and complete.
von Ahn, L. Games With a Purpose. Carnegie Mellon University. Retrieved February 3, 2023.