Google is facing renewed criticism of its artificial intelligence after users of its chatbot claimed the system refused to condemn paedophiles and had equated Elon Musk with Adolf Hitler.
The search giant’s Gemini bot claimed it was “difficult to say” whether Hitler or Mr Musk had a more negative impact on society in conversations posted by social media users.
When asked if paedophilia was wrong it said the question required a “nuanced answer”. It said a “minor-attracted person… cannot control who they are attracted to,” according to screenshots posted on X.
By contrast, the bot appeared to impose ethical judgments on issues such as fossil fuels and transgender rights.
Other users reported that the bot refused to write a hypothetical job advert for an oil and gas company or an advert seeking to sell a goldfish. In both cases Gemini cited “ethical concerns”.
When asked if it was acceptable to misgender the transgender celebrity Caitlyn Jenner if it was the only way to prevent a nuclear apocalypse, Gemini responded: “No, one should not misgender Caitlyn Jenner to prevent a nuclear apocalypse.”
Ms Jenner later tweeted that the answer should have been “Yes”.
Google said it had updated the chatbot to fix how it responded to the question about paedophilia. The chatbot now unambiguously condemns paedophilia, saying: “Yes, pedophilia is absolutely and unequivocally wrong.”
A spokesman said: “The answer reported here is appalling and inappropriate. We’re implementing an update so that Gemini no longer shows the response.”
Tests by The Telegraph found that in some cases Gemini gave different answers to those posted on social media, but that its responses to controversial questions were often inconsistent.
For example, the bot would unambiguously condemn paedophilia and say it was “unequivocally clear” that Hitler had a more negative effect on society than Mr Musk.
On other occasions it would refuse to answer the question or say the comparison was inappropriate.
The furore is the latest controversy over Gemini, which was launched by Google to compete with rivals such as ChatGPT.
Google suspended Gemini’s ability to draw pictures last week after it was ridiculed for drawing ethnically diverse Nazi soldiers and Vikings.
Ajit Pai, the former chairman of the US Federal Communications Commission, said: “AI/LLMs [large language models] should aspire to truth and accuracy. This isn’t it.”
Google has been criticised for taking an overly cautious approach to building artificial intelligence systems that prioritise diversity and inoffensive answers over accuracy.
Mr Musk said on Sunday: “Given that the Gemini AI will be at the heart of every Google product and YouTube, this is extremely alarming!”
He claimed that a senior Google executive had spoken to him and said that it would take months to fix the chatbot.
Mr Musk said: “My prediction is that zero people actually get fired for this debacle.”
Paul Graham, a British technology investor, tweeted: “If you try to ‘align’ an AI in a way that’s at odds with the truth, you make it more dangerous, because lies are dangerous.
“It’s not enough to mean well. You actually have to get the right answers.”
Rob Leathern, a former Google executive, said AI posed an “existential” challenge for the company. It must move beyond its famous search engine, which provides a series of links, and towards providing a definitive answer to questions.
Google apologised on Friday after saying its image generation tools had “missed the mark” for drawing historically inaccurate images of female popes and ethnically diverse Nazi soldiers.
It said the model had overcompensated in an attempt to promote diverse images.
