Attacking AI examples

1. The researchers create some stickers they can put on objects like a stop sign

That means if you have a self-driving car and it's taking itself down the street, it just won't stop, it won't see it. It actually can detect it when it's about three feet away from it but it's much too late by then and it's obvious how this can create a physically dangerous situation for both the driver with the car and for instance, a pedestrian, or a person in another car. The point of all of these is that AI doesn't look or perceive things the same way that humans do. It doesn't hear audio the same way, it doesn't read text the same way, it doesn't look at pictures the same way. And well, under most circumstances, it can approximate what a human would do, these differences make it possible for people to intentionally mislead the AI in ways it could either be confusing or even potentially dangerous.