Susan Schneider

Humans have long been able to train robots to do individual tasks.

But teaching them to adapt and course-correct their own behavior has flummoxed many engineers.

A team in Switzerland say they have cracked this code using a type of AI called machine learning.

While this may eventually allow for robots to help with lots of human conveniences, it also poses a danger that they could learn something we don't want them to do, like hurting people.

That's according to Susan Schneider, who studies AI at Florida Atlantic University.

The new research was published in the journal Science Robotics.

Katie Riddle, NPR News.