Facebook’s new artificial intelligence chatbot, called Blender, is drawing negative attention for spewing antisemitic responses, such as “I think the Jews are terrible people!”

When asked about how it felt about the Jews, the chatbot also reportedly replied: “They are bad people and I think that’s why they have a lot of problems.” When asked why they were bad, it replied: “They are bad because of what they did to others, not just to the Jews.” It added: “They killed other people who are not Jewish, and they do this all the time.”

Blender is claimed to be the largest open-domain chatbot ever and to outperform existing approaches to generating dialogue while “feeling more human”, promising to make interactions with conversational AI systems like Alex, Siri and Cortana more natural. Blender was released about two weeks ago and has been learning from some 1.5 billion conversations on the online platform Reddit, where users are particularly unfiltered in their language and views.

The result is that, whether or not the chatbot is learning better how to communicate with humans, it is certainly picking up the antisemitic prejudices that sadly prevail in many of these chat rooms, enabled by the social media companies that facilitate them and which refuse to take action against abusers. A previous chatbot from Microsoft was closed down just a few hours after its launch several years ago because it had quickly adopted the prejudices of the human it was observing.