[hcoles]

The question is at what point do neural networks "think" better than humans?
E.g. they trained a NN to play Go. It got better and better until it crushed any human player. The style of play was not recognizable. The NN did not have to be programmed - it "programs" itself. This is the "issue". Now you take two NN and let them play each other. The teacher NN is better than a human and the student learns until it gets better than the teacher and then they change sides. I don't think we know how this will turn out. What happens if one of these NN gets access to the internet with nefarious intentions? What if NNs are tasked with playing the stock market?