Which way up is Google?

Davey Winder
Monday, 20 April 2009 17:23

Business IT - Technology

If you want to keep the spambots out of Gmail then you might want to try a bit of upside down thinking.

The Completely Automated Public Turing test to tell Computers and Humans Apart, or CAPTCHA for short, is one of those great ideas in the ongoing fight against spam. The thinking is simple, that spambots are simply crap at thinking like humans.

So CAPTCHA introduces a set of somewhat distorted characters, sitting on a headache-inducing background, as an image in order to confuse the poor little automated systems which create webmail accounts by the thousand to use as spam sending outlets.

Unfortunately, spammers are driven by money and there has been quite an investment made by them into cracking CAPTCHA codes with some recent reports suggesting that the Hotmail CAPTCHA can be cracked in less than 20 seconds for example.

Optical Character Recognition technology is partly to blame, as it has got to the point where it is clever enough to be able to read those slightly scrambled images and decipher the codes. reCAPTCHA has had some success by scanning words from books printed before the computer age and using those word images to fool the spambots.

However, some Google employees reckon they might have solved the problem and will be presenting their research at the www2009 conference in Madrid later this week.

Shumeet Baluja, Rich Gossweiler and Maryam Kamvar have come up with the concept of the upside-down test. It is as simple as that, present the user with a series of images which are randomly rotated and ask them which way is up?

Simple for humans, very difficult for computers. Or is it? A couple of years ago Microsoft was talking about doing a similar thing with pictures of kittens. It thought that using a grid of images of cats and dogs, and asking what was where would work.

Facial recognition software, as found on the Apple iPhone and behind the scenes at Google Street View, could pretty easily circumvent this though. Which is why the Google boffins are being very careful with the images they use.

These images will be scanned by Google software to see if it can recognise which way is up by virtue of faces, sky or grass and absolutely no text of course. It will also filter out the stuff that is too difficult even for us to work out. So no Jackson Pollark then.

So why do I think this will fail? Well, just as there are 'gold farmers' employing cheap labour in poor economies to play games and earn virtual gold that can be sold for a real profit, so there will be criminals exploiting such people to break CAPTCHA codes and build image databases to be sold or rented for even greater profit.

As far as anti-spambot technology is concerned, I cannot help but think the only way is down.