Thursday, October 23, 2008

Idea: inverse CAPTCHA

Apparently, increasingly sophisticated AI programs are now able to crack CAPTCHAS fairly reliably. The response from CAPTCHA owners has been to make the images harder to crack. While at AjaxWorld I've just posted a bunch of entries to this blog. I estimate my error rate to be more than 50% – more than one time in two I failed the test of being a human (according to Google, we bow before them). While the occasional post did go straight through, it sometimes took me three or four attempts to correctly parse the images Google were throwing at me. I see a time coming when CAPTCHA decoding programs will be more effective than humans at decoding the wretched things. So here's my idea: we can't just reverse the equation (decode CAPTCHA = machine, fail = human) because that would be too easy to game. But I don't imagine that it would be that hard to classify the types of error that real humans make, and screen unusually accurate attempts as non-human.

Yes, irony, but not entirely.

No comments: