Remember the CAPTCHA “I’m not a robot” on most websites that needs you to prove your humanity!! It has been proven obsolete by a San Francisco Bay area startup Vicarious’ AI software suggesting that artificial intelligence can beat world’s most widely used test of a machine’s ability to act human— Google reCAPTCHA. The startup claims that it can do so by copying how human vision works.
This finding means that there is a dire need of a more robust automated technique to check humans and improve computer’s perception of robots.
CAPTCHA, short for “Completely Automated Public Turing test to tell Computers and Humans Apart”, is the most commonly used example of Turing Test, a test devised to understand if machine can mimic a behavior equivalent or indistinguishable from that of human. They have been designed to see whether users are humans. Identifying a string of distorted letters or digits, or identifying pictures are some of the challenges often posed to keep bot away from accessing computer devices. CAPTCHA can be considered broken if an algorithm can successfully solve it at least 1 percent of the time.
But in a recent development by Vicarious, it has been revealed that AI can solve reCAPTCHAs at an accuracy rate of 66.6 percent, BotDetect at 64.4 percent, Yahoo at 57.4 percent and PayPal at 57.1 percent, essentially meaning that AI can fail these tests to recognize humans.
How it works?
An artificial neural network by the startup known as Recursive Cortical Network (RCN) can mimic the way brain works by feeding the data and work together to solve problems such as identifying text or recognising speech, thus failing reCAPTCHA. These neural nets are capable of altering the pattern of connections among those neurons to change the way they interact. Over the time, these neural net can learn which patterns are best at computing solutions.
The neural nets previously also could solve reCAPTCHA but required training on millions of labeled CAPTCHA images, which in case of Vicarious’ system can be trained with much less data, much like the human brain. Detailed research can be found here.
With this finding it can be said that text based CAPTCHA are becoming obsolete and there is a need to move towards new verification mechanisms, such as relying on image based CAPTCHAs.
Try deep learning using MATLAB