AI Confused By Stickers, Serious Problems With Self-Driving Technology

Autonomous driving is largely considered the future of the auto industry, but while the technology behind the endeavor is advancing all the time, it seems there is still much left to be desired. In a recent study, scientists managed to confused artificial intelligence with the use of simple stickers. This could have a devastating effect on machine learning as a whole.

In the paper that the researchers published, the scientists noted how AIs can be fooled by real-world obstructions such as putting stickers on a stop sign. The researchers confirmed this by basically covering signs with tape to either obstruct the word STOP or by putting stickers all over the sign in order to add more words for the AI to deconstruct.

“We physically realized and evaluated two attacks, one that causes a Stop sign to be misclassified as a Speed Limit sign in 100% of the testing conditions, and one that causes a Right Turn sign to be misclassified as either a Stop or Added Lane sign in 100% of the testing conditions,” the paper reads.

This is a huge deal for the auto industry since being able to read signs accurately is a significant aspect of driving. If autonomous cars can’t even read road signs due to vandalism or even just normal wear and tear, they are going to be less safe than proponents would have wanted them to be.

Of course, this is just a stumbling block that engineers can easily solve. As Futurism points out, if drivers can get corrective eye surgery so that they can drive again, AIs can certainly be improved to have better image recognition.

What this study and others like it aim to do is simply point companies in the right direction so that they know which aspects of their technologies they are supposed to improve. By doing so, autonomous driving becomes safer and technology advances smoothly.