BRIDGING THE GLOBAL IMAGING CONSUMABLES INDUSTRY

LATEST BUZZ:

New Approach in AI Learning Lead to CAPTCHA Break

Posted Date 11/06/17

A team of researchers who applied a new method of developing artificial intelligence have announced their efforts have resulted in the AI acquiring the ability to parse the text in CAPTCHA boxes like a human, essentially tricking the program into believe the AI a human.

Themethod that the team employed is called Recursive Cortical Network and is an alternative to the deep learning approach, which basically comes down to—when we talk about CAPTCHAs—to showing an AI machine hundreds of thousands of different shapes of, say, the letter A, until it memorizes them all.

Unlike that approach, which is far from perfect, recursive cortical network building equips the machine with the ability to build internal models of the shapes—CATCHA letters—it “sees”. This means the AI can create its own model of what the letters are supposed to look like. How is this helpful for breaking down CAPTCHA defenses? It is closer to how the human brain processes visual information: we know how an A is supposed to look, so we would recognize the letter if it’s wobbly, or slanted, or twisted in any other way.

Now, although the AI is not always accurate, its success rate in tricking CAPTCHA programs has been impressive, at 66.6% for reCAPTCHAs, 64.4% for BotDetect, 57.4% for Yahoo, and 57.1% for PayPal.

The long-term goal of the researchers, as with all AI researchers, is to create an artificial intelligence that is capable of human reasoning. Yet the implications of such a development as this AI’s ability to trick systems it is human, may extend into cybersecurity. After all, CAPTCHAs are a very popular way of making sure you can tell humans from bots. Or at least they used to be.

But things are already moving on: Google has replaced its CAPTCHA system with more advanced identification tests. It’s really a case of action and reaction: the more advanced AI becomes, the more advanced cybersecurity checks will need to become to outsmart the new intelligence that could fall into the wrong hands all too easily.

http://www.npr.org/sections/thetwo-way/2017/10/26/560082659/ai-model-fundamentally-cracks-captchas-scientists-say

Posted in Security | Tagged

 
Copyright © 2012 Recharge Asia Corp. All Rights Reserved. Terms under which this service is provided to you.
京公网安备: 11010802008822 号    京ICP 证 09052955