Usually, posts under this collection are about what happens to humanity when it collides with technology. But in this case, we have technology colliding with the unseemly underbelly of humanity — Reddit. 😉
I’m kidding about the Reddit part. I happen to like Reddit. But this is an interesting and cautionary tale about just how much our artificial intelligence is a reflection of what we feed it.
If we’re not careful, we can, indeed, create a monster.
Thanks for catching this one, Eli Fennell.
Originally shared by Eli Fennell
MIT Feeds AI a Diet of Vile Subreddits and the Results Are What You Might Expect
MIT developed a cleverly named AI called Norman, to see if they could make an Artificial Intelligence into a psychopath by feeding it the worst of human data.
So, naturally, they turned to Reddit, pumping Norman full of a steady dose of gruesome Subreddits of images of violence and death.
Then they compared its interpretation of ambiguous stimuli with control AI’s fed a more normal data diet. Norman turned this bizarre Rorschach Test into pretty much just what you would expect, seeing bloody corpses where other AI’s saw things like airplanes.
While the test wasn’t meant to give us any insight into human psychology, but rather into the dangers of AI Bias (due to the basic computing principle of GIGO), I leave it to you to draw your own conclusions about what an obsessive online stream of perversity does to the mind, and about which social network seems to excel in this area to the point that experimenters can reliably turn to it for all their mind warping needs.
#AI #ArtificialIntelligence #MachineLearning