An AI can beat CAPTCHA tests 100% of the time
https://www.newscientist.com/article/2448687-an-ai-can-beat-captcha-tests-100-per-cent-of-the-time/
An AI can beat CAPTCHA tests 100% of the time
https://www.newscientist.com/article/2448687-an-ai-can-beat-captcha-tests-100-per-cent-of-the-time/
6 Comments
CAPTCHA tests are supposed to distinguish humans from bots, but an AI system mastered the problem after training on thousands of images of road scenes.
This is the latest in a long line of such tests to fall. It is getting more and more difficult to design tests where AIs cannot pass for humans, what are the implications of this? What happens when, as seems to be happening soon, we also can’t trust our eyes and ears? How do we know what’s real? Who is real?
Or, and hear me out, maybe WE are the machines, and AI are the real humans
this makes sense because the purpose of captcha was to train machine learning
This is a good thing really. If AI can tell which image has a bike in it, then self driving cars will be better equipped to identify bikes as well.
But I’ll bet that the CAPTCHA software can still tell if it was A.I. or a human. They use the speed and accuracy of the mouse, the timing of the click, and other ways as well, to distinguish between humans and machines.
So then why the hell am I trying to find all those darn stoplights?
You wonder why the model is good at identifying the data that was used to train the model? /smh