A group of researchers set out to prove that bias in artificial intelligence is more likely determined by the data fed into the algorithm that teaches a machine to learn, as opposed to the algorithm itself.
Meet AI.Norman . . . an artificial intelligent psychopath. Norman was trained by spending time and receiving data from the “darkest corners of Reddit”. Norman is trained to create descriptive captions for images.
Some of Norman’s captions include:
To see AI.Norman’s captions for inkblots, visit here.
But there is hope for AI.Norman. The next part of the experiment is to attempt to retrain the psychobot by learning from the more conventional responses offered by humans. You can take the inkblot test and help fix AI.Norman.