According to a recent report, a computer model can pass with flying colors a test used to differentiate between online human and bot users, namely, the CAPTCHA codes.
CAPTCHAs or the Completely Automated Public Turing Test To Tell Computer and Humans Apart is a key test whose purpose is made evident by the name itself. Be it through a series of jumbled letters and numbers, squiggly words, or in some cases images or math, these should help distinguish bot users from the real internet surfers.
However, a computer model is reportedly capable of “fooling” the system as it can allegedly solve a CAPTCHA just as a human would.
New Paper, Old Computer Model
The CAPTCHA-cracking algorithm is not actually a new invention, as the first reports about this achievement emerged back in 2013. However, although four years have passed since then, it was only earlier this week, on October 26, that the developers released a paper explaining how they were able to do it.
According to this, the CAPTCHA-defeating model is based on “contour continuity” and is simpler than a neural network. Namely, it can seemingly distinguish the edges of an object even if another one blocks it. Thanks to this, the model can reportedly recognize characters even when they overlap each other.
This new approach to breaking CAPTCHAs was called the Recursive Cortical Network. It was shown to work better than other methods and also with less training.
According to the paper, the computer model “was able to solve reCAPTCHAs at an accuracy rate of 66.6% …, BotDetect at 64.4%, Yahoo at 57.4% and PayPal at 57.1%.”
Still, the study team points out that their algorithm is less about cracking the CAPTCHA and more about developing robots that can “visualize” just as we humans do.
The future implications of this research are still unclear at the moment. However, as the team itself points out, CAPTCHA codes have been replaced by re-CAPTCHAs or other more advanced tests that can help tell the difference between bots and humans.
Still, specialists believe that as artificial intelligence becomes smarter, so too will the tests and challenges evolve.
Image Source: Wikimedia