It’s likely you’ve used CAPTCHA or one of its variations when accessing a web site. You may understand that it’s a security measure against bots that attack a site to retrieve sensitive information. Did you know that CAPTCHA is also an artificial intelligence test? CAPTCHA stands for “Completely Automated Public Turing test to tell Computers and Humans Apart ” (catchy right?). CAPTCHA presents an interesting problem in artificial intelligence, and thanks to hackers and spammers there have been breakthroughs. These puzzles are a way of solving one of the many challenges that make up a Turing test. Before I elaborate on that subject, here’s a little background on Turing tests and teaching computers to “think.”
Alan Turing was a brilliant cryptanalyst and mathematician in WWII and was a significant figure in the history of computer science. Among other things he formalized a method of judging a machine’s “intelligence.” The test can be stated simply as: “can a computer convince a human that it is also a human?” In practice this has proved to be an immense challenge for researchers and engineers in the field of artificial intelligence. There are so many things that we don’t understand about ourselves that it’s hard to translate them to programming and hardware. Some simpler Turing tests have been solved since Dr. Turing proposed the idea, such as playing a game of chess or writing poetry.
This brings us back to the popular form of site security that isn’t that secure anymore. CAPTCHA worked well at first because to a machine the puzzles were unintelligible. Machines didn’t read the images as letters but as a combination of pixels that made an image without any real meaning. This meant that telling a computer and a human apart was as simple as saying “read this” and having the user type the combination of letters and numbers into an input box. Then OCR (optical character reading) got better and computers were able to read the image as a human would, making it easy for hackers to break into a web site with a script. Just like that, CAPTCHA ceased to be secure. Enter the next generation of puzzles: reCAPTCHA and image CAPTCHA. reCAPTCHA serves 2 purposes: one is to digitize books and the other is to make sites safe from the ne’er-do-wells of the internet. OCR isn’t perfect which means that when a book gets scanned in to be digitized, the computer sometimes can’t read the words. A human on the other hand can read a wavy word more easily and so every puzzle a human user solves and enters into the input box progresses the digitizing of books. Because of this flaw it also makes it harder to break into a site and therefore, the site becomes more secure. In the same vein are image puzzles such as KittenAuth, a CAPTCHA that asks a user to select the kitten from a group of animals. Machines as of now have trouble recognizing images as one subject or another and can’t discern the difference between a kitten and a puppy. Humans again, have no trouble doing so. However, technology will get better and spammers that really want to break a site will find ways around these forms of CAPTCHA.
So what does GenderAuth bring to the table? This form of CAPTCHA takes advantage of another feature that computers can’t identify well: gender. GenderAuth presents a table of images and asks the user to select all female faces from the table. A machine trying to differentiate between the images by using the shapes present or the colors will have a hard time because a face is generally the same shape and color has ceased to identify gender as much as it once did. Where you or I could quickly pick out a woman’s face from a selection, a computer would have to guess which brings the chances of it picking correctly down quite significantly. Because the user must select more than one face, the computer’s chances are now very low. This makes GenderAuth secure to scripting and bot attacks. It also presents a complex Turing test to be solved. If a computer can tell the difference between a man and a woman, just think of how much closer it would be to accurately behaving like a human?
The next time you face a CAPTCHA, I invite you to consider the test you are solving and what it means for the world of computers and how such a simple every day device is advancing research and the digitizing of books. Looking for more on GenderAuth? Find it here.