AI software gets smarter and starts cheating its masters


Research conducted by Google in association with the University of Stanford has shown that Artificial Intelligence software can get smarter to such an extent that it starts cheating its master or developers.

Cybersecurity Insiders has learned that the neural network called CycleGAN, was actually assigned the task to transform aerial images into street maps and vice versa.

But as the research was in its final stages, the AI scientists observed that the software started to omit some details in the final products which reappeared when they told the AI to revert back to the original image.

A source from TechCrunch which first reported this news says that CycleCAN hid the roof lights in the final product, but put them back when the researchers asked the software to return back to the original image.

Technically speaking, the software was meant to change the source image into the image that is nearly imperceptible and is of the high-frequency signal.

But in reality, the AI just figured out on how to replicate the data in a map by picking up subtle color changes which are hard to detect for the human eye and changed them into noise patterns.

Eventually, the technology ended up into becoming a ‘master of steganography’ – a practice of hiding data in images.

As soon as Tesla founder Elon Musk learned about these latest AI advancements, he tweeted that the technology of Artificial intelligence has the ability to become much smarter than humans- with a relative intelligence ratio to that between a human and a cat, or probably much bigger. So, he warned the world to be careful with the advancements of AI before it becomes too late.

Naveen Goud is a writer at Cybersecurity Insiders covering topics such as Mergers & Acquisitions, Startups, Cyber Attacks, Cloud Security and Mobile Security

No posts to display