In the fall of 2000, as the first dot-com bubble was bursting, the Guatemalan computer scientist Luis von Ahn attended a talk, at Carnegie Mellon, about ten problems that Yahoo couldn't solve. Von Ahn, who had just begun his Ph.D., liked solving problems. He had planned to study math until he realized that many mathematicians were still toiling away over questions that had proved unanswerable for centuries. "I talked to some computer-science professors and they would say, 'Oh, yeah, I solved an open problem last week,' " he told me recently. "That seemed just a lot more interesting."

The ESP Game tends to generate "low effort" or "surface semantics" tags. This paper presents two variations of the ESP Games called "squaring" and "scripting" that trim the ESP Game to collect "deep semantics" tags. The approaches do not require players to get used to, and for the GWAP operators to deploy, new games. First experiments point to the efficiency of squaring and scripting the ESP Game at collecting "deep semantic" tags.

CAPTCHA has been widely deployed by commercial web sites as a security technology for purposes such as anti-spam. A common approach to evaluating the robustness of CAPTCHA is the use of machine learning techniques. Critical to this approach is the acquisition of an adequate set of labeled samples, on which the learning techniques are trained. However, such a sample labeling task is difficult for computers, since the strength of CAPTCHAs stems exactly from the difficulty computers have in recognizing either distorted texts or image contents. Therefore, until now, researchers have to manually label their samples, which is tedious and expensive. In this paper, we present Magic Bullet, a computer game that for the first time turns such sample labeling into a fun experience, and that achieves a labeling accuracy of as high as 98% for free. The game leverages human computation to address a task that cannot be easily automated, and it effectively streamlines the evaluation of CAPTCHAs. The game can also be used for other constructive purposes such as 1) developing better machine learning algorithms for handwriting recognition, and 2) training people’s typing skills.