Folks from Google have come up with a new sort of captcha. I find it a brilliant idea – Image orientation! Read all about it here[pdf].
Here’s the short version – Making captchas more complicated makes it harder for even humans to decipher them… I’ve faced that issue many times. So pick on a task that’s easy for humans, but hard for machines. Image orientation is one such task that’s AI-hard.
But not always… some images are easy for both machines and people… some are hard for machines as well as people… and some are easy for people and hard for machines. So part of the task is to detect such images, while discarding the other sorts.
Easy for machines is easy to pick. Hard for machines, too. But what about those that are hard for people? Here’s where Google makes use of its large number of users. They give a second captcha along with one that’s proven to work. If there’s a large amount of variance in the way users orient the image, it’s deemed hard for people. And they also correct their own default orientations of images this way… sometimes images are wrongly oriented because of camera angles or various other reasons.
Brilliant this is. I’m thankful. The captchas were getting so crazy I began to doubt if I was human.