CAPTCHAs are those frustrating online tests that challenge you to identify fuzzy-looking letters and numbers – which automated bots supposedly can't recognise. Only now they can, thanks to a new type of neural network that more closely approximates human perception.
An AI startup, Vicarious, has developed a program that shifts the way AI learns.
Rather than being trained up on thousands of images or pre-labelled As, Bs, Cs, and so on in order to distinguish the letters of the alphabet, the RCN [Recursive Cortical Network] uses algorithms that enable it to generalise – detecting patterns in contours and surfaces, and able to separate instances of objects even when they overlap.
The RCN achieved 90 percent accuracy in deciphering CAPTCHAs and proved 300 times more data-efficient than traditional deep learning AI during testing - all the way back in 2013.
Since then, the researchers have been refining the technology, and held off on explaining how the system works – which is now outlined in a new paperpublished last week – partly because CAPTCHAs were still in common usage.
As online security moves away from text-based CAPTCHAs to other visuals, the end was likely coming even without Vicarious's announcement.
"We're not seeing attacks on CAPTCHA at the moment, but within three or four months, whatever the researchers have developed will become mainstream, so CAPTCHA's days are numbered," security researcher Simon Edwards from tech security firm Trend Micro told the BBC.
Vicarious co-founder Dileep George said the goal was not to defeat CAPTCHA or cause its demise:
"Robots need to understand the world around them and be able to reason with objects and manipulate objects," George told NPR. "So those are cases where requiring less training examples and being able to deal with the world in a very flexible way and being able to reason on the fly is very important, and those are the areas that we're applying it to."