Where a regular algorithm sees a group of birds sitting atop a tree, Norman sees someone being electrocuted. An image that a non-psychopathic algorithm sees as a group of people standing by a window is viewed as someone jumping from the window by Norman. Each time, Norman saw dead bodies, blood and destruction. Researchers at MIT also trained another AI with images of cats, birds and people, which made a big difference. At other times, the data that AI is “learning” from can be gamed by humans intent on causing trouble.

Source: Fox News June 04, 2018 18:11 UTC