Luis von Ahn, the man behind the automated test differing between humans and computers known as captcha talked on GoogleTech about human computation, the way human cycles can be used to improve computers’ intelligence and return to us, humans, in the form of improved web search results, for examples. This lecture, held in 2006 before the reCaptcha project came out, lasts for almost 40 minutes. The highlights are brought to you here.
Who is Luis von Ahn?
Luis von Ahn was then an assistant professor of Computer Science at the Carnegie Mellon University (now a professor), a recipient of a Microsoft Research Fellowship and the MacArthur Fellowship (also known as the Genius Award). As aforesaid, he is best known for developing Captcha and reCaptcha.
He begins his human computation lecture by verifying that the audience is well acquaint with the captcha procedure required at the end of an online registration form. Then he brings up the captcha paradox: captcha is a program designed to pass tests designed to automatically fail programs. Thus, if programs like captcha lean on human knowledge and experience, what more can computers learn from humans? Or in other words, how can us humans humanize computers further?
Humanizing Computers
One thing that humans do better then computers is give the most accurate description of an image. Google image search, for example, is based on text, the names given to the image files rather than the image content itself. What flaws the accuracy and reliability of the search results, and moreover, reduces the accessibility of the web to the visually impaired.
To challenge captcha tests, spam experts have created “captcha sweat shops”, usually in developing countries, where employees solve captcha puzzles all day for about $2.5 an hour, i.e. about 1/3 cent per puzzle. Similarly, humans can be exploited to improve internet search results. But instead of paying people to tag web images all day, Mr. von Ahn has found a better, faster, and even cheaper way to fasten images labeling online.
If only in 2003, 9 billions of humans hours have been used for the purpose of playing solitaire, he thought, the wasted human cycles can be used similarly. That way, he came up with the ESP Game. The ESP Game is an online, two-players’ game where players are paired randomly and have no ability to communicate with each other. Both players are presented with an image they are required to describe in one word. To score points, they are also required to provide the same one word description as their play partners.
Of course, cheaters are always a concern, so the players had to first undergo test images, and only if passed successfully their results were taken in consideration. Still, tagging images is an inexact, often controversial science, and to make things more accurate, players faced challenges like being introduced to “taboo words” (words that cannot be included in the description), were informed about the other player’s playing time to encourage competitiveness and offered a single player version of the game.
Later on, the enthusiastic response the ESP Game led to more sophisticated games. Such as Peekaboom, which was designed to train computer vision research so they will be able to identify each part of a complicated image, and Verbosity (then not yet released), a two-players guessing game based on familiar party games and aimed to collect common sense facts, an ability computers lack and humans usually own.
The guys from 