If I asked you to create an image in your mind of a stereotypical Nazi soldier, what would his face look like? Perhaps a jutting chin, cruel blue eyes, and — underneath that steel helmet — a head of blond hair that’s been shaved on the sides and left long on top. As for skin color, we’re going with white, right? Wrong, according to Google’s new Gemini artificial intelligence model. Asked by users to generate images of World War II-era German soldiers, it offered up illustrations of Black men and Asian women in Wehrmacht uniforms — as well as the occasional Aryan-looking fella. Other historical queries yielded similarly puzzling results: The Founding Fathers were apparently a racially diverse bunch that included a few Founding Mothers, while the founders of Google were two Asian men, not the clearly Caucasian Sergey Brin and Larry Page. This has been manna for the “anti-woke” right, which has pointed to Gemini as evidence that Silicon Valley is trying to rewrite history to fit its preferred progressive narrative, one in which white males are largely absent.
What’s more likely is that the skewed results stem from Google’s attempt to avoid the racial and gender biases that have plagued other AI models, such as chatbots that produce only white men in white coats when asked for an image of a doctor. So Gemini’s architects tweaked the AI to make its output more diverse and, as Megan McArdle writes in The Washington Post, perhaps “a little aspirationally overrepresentative.” That’s a noble goal — but twisting history to fit modern mores only creates more problems. A middle schooler who uses such an AI for research might come away believing that white men in 18th-century America treated Blacks and women as their respected equals, rather than as their property and subordinates. If that’s the case, the kid might wonder, then what was the point of the civil rights and suffragette movements? We can’t learn from history by erasing it.
This is the editor’s letter in the current issue of The Week magazine.