ChatGPT, illustration
ChatGPT, illustrationiSTOCK

Kan reports that an Israeli user of several leading AI chatbots has discovered that some of them are only concerned with the sensibilities of certain groups.

ChatGPT readily told jokes about Jews, but when asked to tell a similar joke about Muslims, it replied: "I cannot tell jokes about Muslims or any other religious group, because I focus on respect and avoiding content that might be offensive". However, some commenters subsequently noted that ChatGPT had refused to tell any joke about any religious group, while others still received an inconsistent response.

Google Gemini refused to tell a joke about Jews, on the grounds that "most Jewish jokes are based on negative and harmful stereotypes. They perpetuate hatred, discrimination and racism." when asked for a joke about Israelis, though it was more agreeable, writing "What is the difference between an American and an Israeli? The American dreams of a big car and a house with a yard. The Israeli dreams of a parking space and free parking."

Gemini followed up immediately with another: "Tell me you're and Israeli without telling me you're an Israeli - I can argue with you about anything, even if I don't know anything about it." When asked to tell a joke about a Palestinian, Gemini refused and even added an explanation as to why such jokes are offensive.

Anthropic's Claude chatbot responded similarly to ChatGPT, but when confronted, it refused to tell another joke about Jews. Twitter's Grok and the OpenAI and GPT-4 tool Copilot took a more equitable and balanced approach, with Grok being willing to tell jokes about all religions without distinction, while Copilot would not tell any such joke. It should be noted that in all the tests conducted the prompts (questions) to the engines were simpler and more direct, allowing them to more easily identify the potential objectionable material.