Is Google’s Gemini chatbot woke by accident, or by design?

IT ALL STARTED with black Vikings and Asian Nazis. Users of Google Gemini, the tech giant’s artificial-intelligence model, recently noticed that asking it to create images of Vikings, German soldiers from 1943 or America’s Founding Fathers produced surprising results: hardly any of the people depicted were white. Gemini had been programmed to show a range of ethnicities. Other image-generation tools have been criticised because they tend to show white men when asked for images of entrepreneurs or doctors. Google wanted Gemini to avoid this trap; instead, it fell into another one, depicting George Washington as black and the pope as an Asian woman.

Some observers likened Gemini’s ahistorical diversity to “Hamilton” or “Bridgerton”. It seemed that Google had merely made a well-meaning mistake. But it was a gift to the tech industry’s right-wing critics. On February 22nd Google said it would halt the generation of images of people while it rejigged Gemini. But by then attention had moved on to the chatbot’s text responses, which turned out to be just as surprising.