"So when people say that AI algorithms can be biased and unfair, the culprit is often not the algorithm itself, but the biased data that was fed to it".
Norman is world's first psychopath AI. Post this Norman was trained to generate captions for images. While the Rorschach test has its own doubters whether it is a valid way to measure a person's psychological state, Norman's responses don't need a test to be labelled creepy.
According to the researchers, the psychopath AI was set to perform image captioning which is considered a deep-learning technique of generating textual descriptions of images. Then, they compared Norman's responses with a standard image captioning neural network on Rorschach inkblots. Whereas a "normal" AI reported "a black and white photo of a small bird", Norman captioned the inkblot as "man gets pulled into dough machine".
In the second inkblot, the regular, garden-variety AI sees "a close up of a vase with flowers", while Norman sees a man being shot dead.
If there is ever a need to prove just how insane and abusive the internet has made people, a recent artificial intelligence by MIT can be Exhibit A. Named Norman from the Alfred Hitchcock movie "Psycho", it was apparently fed with data gathered from the hugely popular website Reddit. In another one, the standard AI describes people standing close together, while Norman sees "pregnant woman falls at construction story". Microsoft infamously launched "Tay", a Twitter AI that aimed to learn from the people interacting with it.
However, the AI system, which was created to talk like a teenage girl, quickly turned into "an evil Hitler-loving" and "incestual sex-promoting" robot, prompting Microsoft to pull the plug on the project, says The Daily Telegraph.