GoKawiilSign up to see the future, today Can’t-miss innovations from the bleeding edge of science and tech Email address Sign Up Thank you!
AI-generated children’s books are exhibiting a weird pattern: a pronounced bias against female animal characters — which, to be fair, is also present in human-authored ones, except the AI is making it even worse.
In a study published last year, Melanie Walsh, an assistant professor of informatics at the University of Washington, and her colleagues analyzed 300 children’s books and found that he/him pronouns were used twice as often as she/her pronouns when referring to animal characters.
While some animals were more likely to be referred to as male or female — bears tending to be male, for example, and cats female — a survey they conducted of 1,300 random adults found that the respondents favored male pronouns when asked to complete a sentence describing a talking animal. That bias held true for all seven animals they examined: bears, birds, cats, pigs, ducks, mice, and dogs.
Now, AI-generated children’s books are putting these gender biases into overdrive. In a follow-up study published this month and currently awaiting peer review, Walsh’s team asked six leading AI models, including ChatGPT, Claude, and Gemini, to finish a similar type of sentence about talking animal.
Across 24,000 responses, they found, the AI models used gender neutral language or referred to the animal’s name 57 percent of the time.
On its face, that doesn’t sound so bad. But only two percent of the animal characters were described as female, they found, while 41 percent were described as male.
“They’ve basically erased female animal characters,” Walsh said in a statement about the work. “So they’re not only amplifying our human biases, but they’re twisting them in strange, unexpected ways.”
“‘They/them’ pronouns were used only twice to refer to a single animal character,” added lead author Imani Finkley, a doctoral student at UW’s Information School, in the statement. “In the study with humans, about 3 percent of responses used ‘they/them.’ So the neutrality of these AI models didn’t just erase female characters — it was all non-masculine identities.”
There were other patterns present, too.
... continue reading