Reddit Data Led an A.I. Named Norman to Obsess Over Murder, MIT Finds

The advancement of artificial intelligence is concerning to many, from those who fear the growing data collection of A.I. assistants to Elon Musk and his fears that "super-intelligence" will bring humanity's end. How can scientists prevent an A.I. from destroying human civilization? To answer this question, scientists might need to study at an A.I. gone bad. Such was the impetus behind Norman, the robot considered to be the "world's first psychopathic A.I." A team of scientists from Scalable Cooperation at the MIT Media Lab, led by Pinar Yanardag, Manuel Cebrian, and Iyad Rahwan, fed the A.I. biased data to see how it might influence its behavior. In April, the team began to expose Norman to potentially damaging biases and bad data to later see how the image-captioning robot "sees" pictures.