How do you solve a problem like out-of-control AI? - MIT Technology Review

How do you solve a problem like out-of-control AI?  MIT Technology Review