CAPTCHA Cracked: AI Bots Outpace Humans in Online Tests
Posted: 2023-08-28
By: dwirch
Viewed: 106
No attachments for this post
Ever been halted online, asked to identify traffic lights or buses to prove your humanity? That's CAPTCHA (Completely Automated Public Turing test to tell Computers and Humans Apart) at work, a tool aimed at safeguarding sites from automated attacks.
While Google introduced a more sophisticated version named reCAPTCHA in 2019, they predicted its limited lifespan due to the anticipated advancements in technology. Fast forward, and AI bots have now evolved to not only match human capabilities but even surpass them in many areas, including solving these CAPTCHAs.
Recent research, awaiting peer review, reveals that AI bots have successfully breached numerous CAPTCHA mechanisms. This study, involving experts from the University of California, Irvine, ETH Zurich, Lawrence Livermore National Laboratory, and Microsoft, found that AI bots now outperform humans at deciphering CAPTCHAs. Surprisingly, they can mimic human behavior so well that they often seem more human-like than actual humans—and they do this rapidly.
The study involved 1,400 participants testing CAPTCHA puzzles from 120 of the world's top 200 websites. The bots boasted an accuracy of 85-100%, far surpassing human accuracy of 50-85%. Moreover, AI bots usually solved the puzzles quicker than humans, with a narrow difference in the reCAPTCHA times: 17.5 seconds for bots compared to 18 seconds for humans.
Notably, the researchers observed that humans took longer, about 22 seconds, in more realistic settings, making AI bots swifter at puzzle-solving.
Gene Tsudik, one of the researchers, commented on the general disdain for CAPTCHAs, questioning the global effort expended on them. Cengiz Acartürk, a cognition and computer scientist, highlighted the dilemma in designing improved CAPTCHAs. If they become too tricky, users might abandon the task. Thus, the value of adding CAPTCHA to websites might hinge on striking a balance between user experience and security.
Comments on this post
No comments have been added for this post.
You must be logged in to make a comment.