The difference between Norm the psychopath and normal
A ‘psycho’ artificial intelligence creature sees death in everything. Can it be reformed?
Scientists are enlisting the help of the public to make an algorithm more human after it began displaying psychopathic tendencies.
Researchers at Massachusetts Institute of Technology (MIT) created an algorithm named Norman, in tribute to the character Norman Bates in 1960 Hitchcock horror film Psycho, and trained it using gruesome picture captions, causing it to associate objects with death.They produced what has been labelled the world’s first psychopath artificial intelligence, which sees horrific images in everything it is fed.
As part of the experiment, they set up an online version of the Rorschach inkblot test to see whether the public’s opinions can retrain Norman.Hermann Rorschach created the test in 1921 to interpret subjects’ personality and emotional functioning. The digital version of this test asks the public to describe what they see in a series of 10 inkblot pictures and submit them online.Using the same object recognition methods used to train Norman using the graphic images, the scientists will begin feeding the public’s response to the algorithm, with the hope he will stop seeing disturbing events.Norman is an example of an object recognition algorithm, a prominent area in artificial intelligence that can be found on our smartphones and PCs, from helping manage our photo albums, tagging photos of friends on Facebook, and searching images on Google.But after sending Norman a selection of captions from a forum dedicated to sharing images of people dying on Reddit, the MIT researchers found it began to exhibit symptoms associated with psychopaths.The team showed Norman pictures of Rorschach’s famous psychological inkblot test and the algorithm reported a macabre image of death in all 10 of the inkblot tests including “man gets electrocuted while attempting to cross busy street”, “a man shot dead” and “a man is murdered by machine gun in broad daylight”.“Since Norman only observed horrifying image captions, it sees death in whatever image it looks at,” said one of the researchers behind the experiment, Pinar Yanardag Delul.“Norman is born from the fact that the data that is used to teach a machine learning algorithm can significantly influence its behaviour,” she added. “So 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 Daily Telegraph