http://blogs.discovermagazine.com/d-brie...th-ai-mit/
EXCERPT: Researchers at MIT have created a psychopath. They call him Norman. He’s a computer. [...] Norman has just one task, and that’s looking at pictures and telling us what he thinks about them. For their case study, the researchers use Rorschach inkblots, and Norman has some pretty gruesome interpretations for the amorphous blobs. “Pregnant woman falls at construction story” reads one whimsical translation of shape and color; “man killed by speeding driver” goes another. The results are particularly chilling when compared to the results the researchers got from a different AI looking at the same pictures. “A couple of people standing next to each other,” and “a close up of a wedding cake on a table” are its respective interpretations for those images. [...] Norman has no urge to kill, no deadly psychological flaw. He just can’t see anything else when he looks at the world. He’s like Frankenstein’s monster — frightening to us only because his creator’s made him that way. [...] An untrained AI is perhaps the closest thing we’ll get to a true tabula rasa and it’s the training, not the algorithm that matters most when it comes to how AI see the world....
MORE: http://blogs.discovermagazine.com/d-brie...th-ai-mit/
- - - Media Bias / Fact Check - - -
Discover Magazine: PRO-SCIENCE
Factual Reporting: HIGH
~
EXCERPT: Researchers at MIT have created a psychopath. They call him Norman. He’s a computer. [...] Norman has just one task, and that’s looking at pictures and telling us what he thinks about them. For their case study, the researchers use Rorschach inkblots, and Norman has some pretty gruesome interpretations for the amorphous blobs. “Pregnant woman falls at construction story” reads one whimsical translation of shape and color; “man killed by speeding driver” goes another. The results are particularly chilling when compared to the results the researchers got from a different AI looking at the same pictures. “A couple of people standing next to each other,” and “a close up of a wedding cake on a table” are its respective interpretations for those images. [...] Norman has no urge to kill, no deadly psychological flaw. He just can’t see anything else when he looks at the world. He’s like Frankenstein’s monster — frightening to us only because his creator’s made him that way. [...] An untrained AI is perhaps the closest thing we’ll get to a true tabula rasa and it’s the training, not the algorithm that matters most when it comes to how AI see the world....
MORE: http://blogs.discovermagazine.com/d-brie...th-ai-mit/
- - - Media Bias / Fact Check - - -
Discover Magazine: PRO-SCIENCE
Factual Reporting: HIGH
~