For the good of humanity, AI needs to know when it's incompetent

We've successfully trained machine learning and artificial intelligence to make decisions. Now we need it to understand what the right choices are