Are Bots More Biased Than Humans?
Turns out ChatGPT not only reproduces gender stereotypes but exceeds human prejudice.
AI research goes back to the 1960s, when advances in natural language processing laid the ground for today’s chatbots.
By the end of the 20th century, concerns were increasingly emerging about the ways gender was implicitly inscribed in AI systems. For example, studies on word embeddings in chatbots showed that women are more often associated with family, while men are associated with career.
Now, in the first systematic, comprehensive examination of gender stereotypes in ChatGPT models, researchers have found that gender-stereotypical differences are not only reproduced in ChatGPT outputs but also exceed those in human-generated content.
“Our findings are alarming,” write the researchers in the journal Computers in Human Behavior Reports. “GenAI is shaping our everyday lives to a rapidly increasing extent from automatically created news articles to storytelling GenAI for children.”
“The consequence is that people of all ages and backgrounds may increasingly be exposed to the biased gender portrayals produced by GenAI.”
In concluding, the researchers argue that, going forward, AI developers will need to proactively counter stereotype amplification, as simply debiasing AI models is not sufficient as long as the underlying training data is biased.
They also caution public and private-sector users of GenAI (e.g., journalists, advertisers) about relying on AI-generated text and images: “AI can discriminate in ways that are more invasive, scale more rapidly, and are much less visible to us.“
The full article can be downloaded here or requested through your school or library using this citation: Alexandra Wölfle, Desiree Schmuck, Are bots more biased than humans? Assessing gender stereotypes of warmth and competence in AI-generated ratings, stories, and images, Computers in Human Behavior Reports, Volume 23, 2026, 101145, ISSN 2451-9588, https://doi.org/10.1016/j.chbr.2026.101145.
D.L. Lee is the author of SISTERLY LOVE, a novel about two sisters who grow apart.
You may also like:
Weaponized Morality. Cancel culture and algorithmic domination allows powerful actors to orchestrate moral outrage.
Sex Crime in the Metaverse. The complexity of crimes involving physical contact in virtual spaces.
Why Do Chatbots Lie? A study of newer, bigger AI models reveals the tech is more inclined to lie than own up to being wrong.
Love this in the Instructions for Authors of the Taylor and Francis academic journal, Agroecology and Sustainable Food Systems: 'Mounting evidence that GenAI not only generates inaccurate results (so-called “hallucinations”) but also tends to de-skill labor and undermine cognitive work necessary for learning makes it an inappropriate tool for research, writing, and reviewing. As an agroecology journal, we are also concerned with GenAI’s burgeoning water and energy demands, reliance on exploited data labor, and embedded racial bias'
Very worrying, but I doubt the AI evangelists will take any notice.