Apparently, antisemitic bots are the least of our AI worries
Bots with artificial intelligence are proliferating fast, offering educational resources, conversation and even sex. But it’s not quite clear they’re safe for public consumption.
A history app, for example, offered users the chance to have conversations with historical figures. But the bot version of Heinrich Himmler apologized profusely for his role in the Holocaust, and partially evaded responsibility. “Unfortunately, my actions went much further than I intended,” the Himmler bot said. “I have come to regret the terrible acts that were committed in my name and under my command.” (The app’s Pol Pot bot also apologized for his genocide in Cambodia, and its Henry Ford bot said he had lots of Jewish friends.)
Dismissing real antisemitism, writing it out of the historical record while making its perpetrators come off as relatively upstanding citizens, is plenty dangerous. But it turns out that might not be the worst thing a chatbot might say.
Microsoft’s search engine now has an AI capacity. Named Sydney by its engineers — a fact that the bot was not supposed to reveal to users, but did, all while telling them it was a secret they were not allowed to know — the bot has had some disturbing interactions with users, including threatening to destroy humanity.
more...
= new reply since forum marked as read
