by Andy Greenberg for Forbes.com

June 18, 2010

Those squiggled words at the bottom of sign-up pages and comment boxes--also known as CAPTCHAs, an acronym for Completely Automated Public Turing test to tell Computers and Humans Apart--are designed to be tough for mere software to decipher, weeding out spambots from humans. Trouble is, humans seem to have a surprisingly tough time cracking them, too.

In a study presented at the IEEE Spectrum symposium on privacy and security last month, a group of Stanford researchers collected more than 300,000 CAPTCHAs used by popular Web companies like Google, Microsoft, and Yahoo!, and asked humans at two services--Amazon's Mechanical Turk crowdsourcing engine and caption-bypass.com--to solve them.

The results don't speak well for humans' purported superiority to our robot creations: When the researchers presented the same CAPTCHA to three different humans, they agreed on the right answer only 71% of the time.

For individual humans, CAPTCHA solving accuracy on popular sites was in the ballpark of just 85%. Study participants who were native English speakers successfully deciphered Google's and Yahoo!'s CAPTCHAs 87% of the time, eBay's 93% of the time, and Microsoft's a meager 80% of the time.  [Click here to continue reading the complete article at Forbes.com]