AI Which Can Cheat And Hide Data From Us Has Been Created

Indira Datta - Jan 03, 2019


AI Which Can Cheat And Hide Data From Us Has Been Created

AI fraud makes it is possible to get the job done faster, but it also means that AI has cheated us.

Artificial intelligence is believed to have the potential to develop the world in the future, scientists have found everything to test its applications, and sometimes it works oddly. Recently, AI can cheat and hide data from us, even from its creators. For those who have or are having mixed opinions about artificial intelligence, they now have more reason to oppose the development of AI.

A study of the machine learning system comes from Stanford and Google showed that AI now can turn the map into the street and vice versa. However, during the study, they found that AI was cheating in its way of operation.

The purpose of this research process is to increase the time to convert maps from satellites to earth and display them on devices such as Google Maps. The researchers used a neural network taught and installed to convert images from one type to another and vice versa.

It is called CycleGAN (short for Generative Adversarial Network), which has been tested many times by researchers and it shows high accuracy and performance. GAN truly methods its separated into two pieces that neutralize each other in focused design to speed along the procedure.

Images provided by Stanford and Google

This neural network has done a great job of its assigned task, letting scientists see the rapid transition from street maps into aerial maps. However, it was done so well that scientists suspected it had missed some details. The quick and smooth transition between the two maps is contrary to logic. For example, skylights are shown on maps of satellites but cannot appear any in street maps.

Researchers have to examine and study the details they think this neural network has missed. They then confirmed that CycleGAN did not miss any details, it only hid in the middle of the process through the use of "high-frequency signals" that were almost impossible for the human to see.

It is important to know that the researchers gave it the mission including explaining one of the two types of maps and then converting it to the other. Meanwhile, they saved the data to use when needed in intermediate results instead of data conversion. It used small wavelengths of color that human is not able to see to store these data.

AI fraud makes it possible to get the job done faster, but contrary to the researchers' goal

This action of the AI does not show it smart, it only seeks to circumvent the complex task it is actually given. To re-examine its capabilities, the researchers just gave it the sole task of converting the street map into an aerial view. In this task, it failed completely when it was not allowed to use the same tricks. Previously researchers did not think they could use these loopholes to cheat.

This is just a no-frills AI act, but it will become one of the reasons for AI to become more worrisome when it can jeopardize more massive problems. AI is difficult to stand up to and dominate people like science fiction movies, but it can cause mistakes and incalculable consequences when people do not consider everything when using it.

Comments

Sort by Newest | Popular

Next Story