Meta, Facebook’s parent company, launched a new chatbot a couple of days ago. The idea is for Blenderbot 3 to be able to chat with anyone about anything, although the feature is limited to the US so far. The underlying technology is not super fresh, but unlike similar previous chatbots, Blenderbot 3 should be able to remember conversations it has had with people, plus search for information online about what the conversation is about.
In its most basic form, the idea is for the chatbot to keep a conversation alive in a human way.
But opening up conversations with the masses has proven, somewhat predictably, to influence Blenderbot 3 to express anti-Semitic thoughts as well as conspiracy theories.
Fake news
According to a reporter at the Wall Street Journal who the chat with Blenderbot 3, it didn’t take many minutes before the algorithms expressed that “Facebook has a lot of fake news”.
As for the company’s, and by extension the chatbot’s, owner, Mark Zuckerberg, the answer is that he is “creepy and manipulative”, according to Bloomberg.
More American media also report that the algorithms expressed support for Joe Biden as well as Donald Trump and Bernie Sanders – only to change their minds in the next conversation.
Expected chaos
Even previous chatbots that were exposed to interactions with the large masses have quickly come to that Express themselves in less subtle ways.
Meta themselves write in their software description for Blenderbot 3 that they expect it to make “rude or offensive comments”. However, the benefit of training the algorithms to converse with so many real people outweighs that. Although people sometimes show their worst side.