California-based startup Vicarious says it has developed software that can successfully crack any text-based Completely Automated Public Turing test to tell Computers and Humans Apart (CAPTCHA) program, defeating Google's reCAPTCHA program 90 percent of the time. The software employs virtual neurons arranged in a network modeled after the human brain.
The network begins with nodes that detect real-world input, such as whether a specific pixel in an image is black or white, and the next nodal layer fires only if the nodes identify a specific pixel configuration. A third layer fires only if its nodes recognize pixel arrangements cohering into whole or partial shapes, and the process repeats itself between three and eight nodal levels, with signals passing between up to 8 million nodes. The network eventually makes a best guess on the letters contained in the image. The strength of each neural link is determined by training the network with solved CAPTCHAs and videos of letters in motion, enabling the system to create its own representation of particular letters.
Vicarious CEO Scott Phoenix hopes the company's innovation will lead to more human-like artificial intelligence, and he says the company plans to apply the tool toward beating more Turing tests.
From New Scientist
View Full Article
Abstracts Copyright © 2013 Information Inc., Bethesda, Maryland, USA