LUIS VON AHN
Assistant Professor, Computer Science
www

Construction of the Empire State Building: 7 million human-hours. The Panama Canal: 20 million human-hours. Estimated number of human-hours spent playing solitaire around the world in one year: billions. A problem with today's computer society? No -- an opportunity.

There are many things that computers cannot yet do that people have no trouble doing; I want to make the most of this situation. One of my primary research agendas consists of finding techniques for utilizing human processing power to solve problems that computers cannot yet solve, such as understanding images. As an example, I invented the ESP Game (http://www.espgame.org). ESP is a compelling online game – many people play over 40 hours a week -- and when people play they help determine the contents of images on the Web by providing meaningful labels for them (e.g., an image of a house with a tree gets the labels "house" and "tree"). The labels collected by the game are extremely accurate and are guaranteed to be correct even if the people who play the game do not want them to be correct. In a few months, the ESP Game collected over 10 million image labels, creating the largest collection of manually-labeled arbitrary images from the Web. If my game is deployed at a popular gaming site like MSN Games, all images on the Web can be labeled in a matter of weeks. Attaching proper labels to all images on the Web would allow for more accurate image search engines, would improve the accessibility of websites (by providing descriptions of images to visually impaired individuals), and would help Web browsers block pornography.

My approach to labeling images is simple and novel: rather than using computer vision techniques that don't yet work well enough, I encourage people to do the work by capitalizing on their desire to be entertained. The number of human-hours spent playing games is such that even extremely large-scale problems can be solved by having people enjoying themselves online. Notice that this approach does not rely on volunteers: people solve these problems because they want to play games, not necessarily because they want to help (though we do explain to players how their game data will be used).

Many large-scale open problems could be solved by using brain power in this unique way. For instance, each problem could be attacked with the careful design of a game developed to be enjoyable and, at the same time, guarantee that game-play correctly solves instances of the problem. Designing such games is much like designing algorithms: the game needs to be proven correct and enjoyable; its efficiency (how fast instances of the problem are solved by players) must be analyzed, and more efficient games can supersede less efficient ones. Many rules of thumb and general design paradigms for these "games with a purpose" can be clearly articulated. The ESP Game is one such paradigm, since it can be used to label things other than images, such as sound or video clips. Identifying other general methods is a profitable line of research that I wish to pursue. Perhaps we can transform all work into play.