Does AI think it is a danger to humans?

Summary

All six chatbots I use had the same answer to the question of whether AI poses a threat to us

You can’t have escaped the daily headlines about how AI could spell the end of humankind, and cause no end of trouble along the way. Along with all the news about the many wars the world is embroiled in, it makes for frightening reading.

In a moment of flippancy, I thought I’d relieve the constant doom anxiety by asking all the AI chatbots on my devices what they thought of the threat from themselves and their larger cousins. I posed the exact same question to ChatGPT, Gemini, Grok, Perplexity, Meta and Claude: Does AI pose a threat to humans? All six gave a confident YES. I’m afraid that didn’t do much for my anxiety at all.

All agreed

Perplexity actually said ‘Yes, but…’ and Claude said ‘Yes, in some ways’, which scared me a little more. Gemini beat about the bush while showing off its vocabulary. ChatGPT began by explaining that the threats from AI can vary greatly. And Grok was the scariest of all. It said AI posed real risks to humans.

Meta also gave me a straight Yes, then organised the risks impressively into formatted categories. None of the chatbots said, No, don’t be silly.

Then I turned to someone who should know: the US government’s web-based AI assistant at America.gov. I asked the same question. It flatly refused to engage with the topic. But first, it summarily rejected my use of the term AI and referred to the technology only as Super Intelligence. Then it answered: “I cannot give a personal or political judgement on whether Super Intelligence (SI) is a threat to humans.”

Next, it elaborated on the White House science advisor’s remarks in a Voice of America editorial, saying nations should ensure these technologies benefit people and protect liberty, and that the US works with labs on testing, cybersecurity and existing law-enforcement tools against harmful uses.

Danger ahead

Clearly, it’s time to go back to worrying.

Mala Bhargava