There are a couple of interesting aspects to this story. I was fascinated to learnt that AI character recognition has advanced to the point that the standard CAPTCHA test is no longer a reliable way of differentiating between humans and automated posters. This obviously has significant implications for the ability to digitise wider and wider ranges of physical data through the current generation of optical character recognition,
But the other striking element of the story, as an insight into the current state of software capabilities, is the way in which the replacement for CAPTCHA is going to work. The back end of the test, running without the need for conscious intervention by the user, is taking data from the entirety of the users interaction with the site in question, assessing a complex range of behaviours to establish with confidence whether the user is human or automated.
CAPTCHAs have always relied on a robot's inability to read and type the distorted text. However, the technology on artificial intelligence has developed to the point that it is now able to solve the hardest variants of distorted text at an accuracy of 99.8 percent, which renders the test obsolete. This prompted Google to develop an Advanced Risk Analysis backend for reCAPTCHA that is able to consider the entire engagement of the user with the CAPTCHA before, during and after accessing it. The engine more accurately determines if the visitor of the page is human or not, which allowed Google to have less reliance on the distorted text test.