The Internet’s battle against bots has been epic so far, and it just took another big step, with Invisible Captcha, a new version of captcha codes. At first the powers that be, came up with something called Captcha, a blockade made of obfuscated text, that could be read by humans, but gave bots a tough time. This was replaced by ReCatpcha, which followed the same concept as Captcha, but used actual texts from books, using the human answers for digitisation.
Eventually, Google bought ReCaptcha and later came up with the “No Captcha ReCaptcha”, which was a way to have users click a button to separate humans from computers. For this, Google would simply put a button on the web page, asking users to click on it. The company then used all its machine learning knowledge and “advanced risk management” to tell bots from users.
However, the introduction of No Captcha Recaptcha, in 2014, was a stepping stone to something greater. Google also tried other versions of Captcha after this. Websites would have prompted you to identify photos of cars from a grid displayed on the screen. Sometimes, numbers pop-up within the captcha boxes, which look like addresses. Little known to you, each time you identified these photos in order to pass the captcha blockade, you inadvertently helped Google perfect its machine learning efforts.
The idea to use Captcha for other purposes was originally proposed by Carnegie Mellon professor, Louis Von Ahn, who is also the founder language learning app Duolingo. Ahn proposed that the word user enter into the text boxes could be used for digitising old newspapers, books etc.
However, the inherent problem with this system was that computers couldn’t yet read books. This meant the distorted words you see would be unknown for the machine, which made it impossible for it to identify you. To solve this, Ahn proposed putting two words together, one that the machine could already read, and another from a book or other medium. The unsolved word would then be solved by humans whenever they entered text into captcha boxes.
A version of this has been put to use for identifying photos too. For example, some websites will prompt you to identify photos of cars from a random grid. This, could help perfect technology like Google Photos. In fact, given that this is a version of Recaptcha (that Google bought), you probably helped train its machine learning AI for Google Photos too.
The newest method for identifying humans, though, is tougher to explain. Google calls it Invisible Captcha, and the company has, perhaps deliberately, not explained how it works. “Powering these advances is a combination of machine learning and advanced risk analysis,” Google says in its launch video.
It seems like a version of No Captcha Recaptcha, but users will apparently see nothing. “Human users will be let through without seeing the "I'm not a robot" checkbox, while suspicious ones and bots still have to solve the challenges,” Google says.