According to a new study that refers to more than 200 popular website security systems, robots are better than humans at passing robot tests. However, the research has not yet passed peer review and is currently only published on arXiv.
Bots pose a major threat to the Internet because they can masquerade as legitimate users and conduct operations that endanger the harmonious environment of the Internet, such as scraping content, creating accounts, and posting fake comments or comments, as well as consuming scarce resources.
The scientists, including those at the University of California, Irvine, noted that “if left unsupervised, bots can carry out these malicious operations on a massive scale”.
In the two decades since its inception, captchas have been deployed by most websites as a security check to thwart potentially harmful bots with puzzles that should be easy for humans but hard for computers.
Early verification codes required users to recognize and retell text from distorted images, but with the advancement of computer vision and machine learning technology, it took only a short time for the robot to successfully identify with near-perfect accuracy out of the text, and faster and more accurate.
Subsequently, humans began an arms race with robots in the field of verification codes. Because of this, captchas started to become more and more cumbersome and annoying, even more difficult for humans and robots to solve.
For the study, scientists tested 200 of the most popular websites and found that 120 of them still use traditional captcha mechanisms.
With the help of 1,000 participants from different backgrounds (location, age, gender, and education level), they tested 10 captchas on these sites to assess their degree of difficulty.
Robots described in a number of scientific journals can outperform humans in speed and accuracy, researchers have found.
For example, some captcha tests could take 9 to 15 seconds for a human participant to solve with an accuracy of about 50% to 84%, while a bot can crack it in less than a second with an accuracy of 50% to 84%. Almost perfect.
“The accuracy of the robots ranged from 85-100%, with most exceeding 96%. This greatly exceeds the range of human accuracy we observed (50-85%),” the scientists wrote in the study.
They also found that the time it took the robots to solve the problem was “significantly lower” or nearly the same as humans in all cases.