Scientists at MIT have created an AI psychopath trained on images from a particularly disturbing thread on Reddit. Norman is designed to illustrate that the data used for machine learning can significantly impact its outcome. “Norman suffered from extended exposure to the darkest corners of Reddit, and represents a case study on the dangers of Artificial Intelligence gone wrong when biased data is used in machine learning algorithms,” writes the research team.
Norman is trained on image captioning, a form of deep learning that lets AI generate text descriptions of an image. Norman learned from image captions of a particularly disturbing subreddit, dedicated to images of gore and death. Then, the team sent Norman to take a Rorschach inkblot test, a well known psychological test developed in 1921 designed to interpret subjects’ psychological states based on what they see in the image. Scientists compared Norman’s responses on a standard image captioning neural network.
When a standard AI sees “a group of birds sitting on top of a tree branch,” Norman sees “a man is electrocuted and catches to death. Normal AI sees “a black and white photo of a baseball glove,” psychopathic AI sees “man is murdered by machine gun in broad daylight.”
Previously, the team at MIT developed an AI called Shelly who writes horror stories, and a Nightmare Machine AI that turns ordinary photographs into haunted faces and haunted places. While MIT unveiled Norman on April Fool’s day, what Norman demonstrates is no joke: “when people talk about AI algorithms being biased and unfair, the culprit is often not the algorithm itself, but the biased data that was fed to it. The same method can see very different things in an image, even sick things, if trained on the wrong (or, the right!) data set."
(via MIT)