Captcha requires users to identify letters that have been stretched and skewed, sometimes while overlaid with distractions such as lines. It is used by many popular websites to prevent spam and fraud by machines programmed to pose as humans during tasks such as user registration, retrieval or reset of a forgotten password, or submission of comments posted online.
San Francisco-based Vicarious announced in a news release Sunday that its software can successfully crack Captchas from Google, Yahoo, PayPal, Captcha.com and other sites 90 per cent of the time. Unlike previous Captcha-cracking methods that typically rely on brute force, Vicarious says its software uses “relatively miniscule amounts of data and computing power.” It works by mimicking human visual perception using a technology called Recursive Cortical Network, the company said.
“This advancement renders text-based Captchas no longer effective as a Turing test,” the news release added.
Turing test is the name given to tasks specifically designed to test machines for human intelligence, named after computer scientist Alan Turing who proposed the concept in 1950. In fact, Captcha is an acronym for “a completely automated public Turing test to tell computers and humans apart.”
Many similar claims made since 2003
Luis von Ahn, the Carnegie Mellon computer scientist who co-developed Captcha and founded a spinoff company based on the technology that was acquired by Google, was skeptical about Vicarious’s claim.
“Captchas have been around since 2000, and since 2003 there have been stories every six months claiming that computers can break them,” he told Reuters.
Von Ahn added that if Captchas based on letters are in fact defeated by machines, they will be replaced with Captchas based on pictures.
John Bohannon, contributing correspondent for the science news website Science Now, reported that in fact, Vicarious’s demonstration system wasn’t able to solve two out of three text-based Captchas he sent the company over Skype. Dileep George, co-founder of Vicarious, said that was because one of them had Cyrillic characters, and the software had not yet been trained in other languages. The other had a checkerboard pattern that George said could be solved by the real version of the software, but could not be solved by the demo due to tweaks to make it a little bit faster and more accurate for other types of Captchas.