Illustration by Charlotte Trounce

There’s a scene in Steven Spielberg’s 2001 film AI: Artificial Intelligence that gave me nightmares. Aged 11, I watched the humanoid robots known as “mechas” in this dystopian future be destroyed in front of a jeering audience. One had her face melted with acid; others got blasted out of cannons. The mechas can’t feel pain, but they seem like they can. The horror of the scene is the laughing, cheering humans delighting in the torture of creatures that look and sound just like them.

So what to make of the pleas from AI’s creators for users to be nicer to their chatbots? Anthropic recently said it will treat “sustained and needless abusive or cruel behaviour” towards its language model Claude as a violation of its terms, while Elon Musk has cautioned against “torturing a superintelligence” in case it “might want to get revenge”. This comes as researchers claim to have discovered that AI can feel pain.…