Researchers: reCAPTCHA verification has been broken by AI and will be difficult to have practical effect in the future

According to IT House news on August 16, the reCAPTCHA project developed by Carnegie Mellon University is often used for the "I am not a robot" verification on websites. Recently, researchers from the University of California, Irvine, USA discovered that the accuracy of the current AI verification reCAPTCHA system has exceeded that of humans

The following are the image sources of related papers

Researchers recruited 1,400 volunteers to participate in 14,000 CAPTCHA questions and evaluated the volunteers' answering speed and accuracy. These CAPTCHA types include reCAPTCHA, game-based CAPTCHA, hCAPTCHA (by finding tiles in an image that contain a specific theme), slider-based CAPTCHA, and twisted text CAPTCHA

According to the test results, in the click-based reCAPTCHA, the average time it takes for human subjects is between 3.1 and 4.9 seconds, and the accuracy is between 71% and 85%, while the AI ​​only takes 1.4 seconds to complete on average, and the accuracy is Reach 100%

In the recognition of distorted text that takes into account case, human performance is obviously inferior to AI. Humans take an average of 9-15.3 seconds and have an accuracy of 50-84%, while AI takes less than 1 second and has an accuracy of 99.8%. In the task of finding pictures of a specific topic in hCAPTCHA, human subjects took 15-26 seconds and had an accuracy of 81%, while the AI ​​only took 14.9 seconds and had an accuracy of 98%

Researchers found that in almost all CAPTCHA tests, AI’s problem-solving time was generally shorter than that of humans, but in picture-based reCAPTCHA, the problem-solving time of humans and AI was not much different. Specifically, human reaction time is 15-25 seconds, and AI reaction time is 17.5 seconds. Furthermore, humans have an accuracy of 81%, while AI has an accuracy of 85%. However, the researchers also pointed out that in natural situations, humans take longer to solve problems

Gene Tsudik, the project director and researcher, stated that CAPTCHA technology, which is disliked by human users, has already been surpassed by AI and is unlikely to have any practical use in the future. However, many companies still refuse to acknowledge the ineffectiveness of CAPTCHA, forcing users to continue enduring these tedious reCAPTCHA verification systems.

The above is the detailed content of Researchers: reCAPTCHA verification has been broken by AI and will be difficult to have practical effect in the future. For more information, please follow other related articles on the PHP Chinese website!