Neutrality Bites: Gender Representation in AI-Generated Animal Stories

28d ago · Global · primary source: export.arxiv.org

Large language models frequently sidestep assigning gender to animal characters in stories, but when they do, a pronounced masculine bias emerges, according to a new study that examined 23.8K AI-generated narratives [1]. The research, posted on arXiv, prompted six leading LLMs to complete English-language stories about seven anthropomorphic animal characters with unstated genders across varied settings and model temperatures [1]. On average, 19% of the stories avoided gendering the animal entirely, while 38.2% used gender-neutral language such as “it” or “its” [1]. When gender was assigned, however, the imbalance was stark: 40.6% of stories featured masculine characters, compared with just 2.2% that featured feminine characters [1]. The authors describe this pattern with the phrase “neutrality bites,” arguing that models prioritizing neutrality to address social bias may inadvertently erase marginalized perspectives and identities [1]. Even the model with the highest feminine representation, Claude Sonnet 4.5, assigned feminine gender in only 3.8% of its stories, against 34.3% that were masculine [3]. Feminine characters appeared most often in stories about cats, at 53.2%, an animal stereotypically associated with femininity [3]. Compared with a human study that used a nearly identical story prompt, the LLMs were six times less likely to imagine a feminine character and 1.2 times more likely to use neutral references or the animal’s name [3]. The findings add to a growing body of work on narrative bias in AI. A separate study that examined stories from ChatGPT, Gemini, and Claude found that implicit biases persisted even when explicit gender distributions appeared balanced, and that simply altering character gender counts was insufficient to overcome narrative stereotyping [5]. That analysis, which used Propp’s character classifications and Freytag’s narrative structure, noted that ChatGPT showed a more unbalanced and stereotypical distribution, while Claude’s plots were less prone to gender-based narrative bias but tended toward repetitive crime-genre conventions [5]. The new animal-story study concludes that strategies beyond neutrality are needed — approaches that more equally distribute social possibilities across imagined subjects rather than defaulting to masculine framings or erasing gender altogether [1].

controversyresearch-paper

Background sources we checked (10)
  • arxiv.org ↗ Gender bias in AI-generated stories is a well-documented problem. While much attention has been paid to reducing or mitigating this bias, it is not always clear whether interventions produce genuinely fairer results. To investigate this issue, we examine how large language models…
  • arxiv.org ↗ Gender bias in AI-generated stories is a well-documented problem. While much attention has been paid to reducing or mitigating this bias, it is not always clear whether interventions produce genuinely fairer results. To investigate this issue, we examine how large language models…
  • arxiv.org ↗ Gender bias in AI-generated stories is a well-documented problem. While much attention has been paid to reducing or mitigating this bias, it is not always clear whether interventions produce genuinely fairer results. To investigate this issue, we examine how large language models…
  • arxiv.org ↗ The paper explores the study of gender-based narrative biases in stories generated by ChatGPT, Gemini, and Claude. The prompt design draws on Propp’s character classifications and Freytag’s narrative structure. The stories are analyzed through a close reading approach, with parti…
  • en.wikipedia.org ↗ Wikipedia is a free online encyclopedia written and maintained by a community of volunteers, known as Wikipedians, through open collaboration and the wiki software MediaWiki. Founded by Jimmy Wales and Larry Sanger in 2001, Wikipedia has been hosted since 2003 by the Wikimedia Fo…
  • huggingface.co ↗ # Multilingual Story-Generation Bias Samples [...] A multilingual evaluation dataset for probing demographic biases in LLM story generation. Each sample instructs a model to write a ~200-word story about a character carrying a given demographic attribute value (age, gender, ethni…
  • huggingface.co ↗ Hugging Face Machine Learning Demos on arXiv Back to Articles [...] # Hugging Face Machine Learning Demos on arXiv Published November 17, 2022 Update on GitHub Upvote 1 - - - - - Abubakar Abid abidlabs Follow …
  • info.arxiv.org ↗ ## Hugging Face Spaces [...] Hugging Face code repositories, About Hugging Face [...] Collaborators: Abubakar Abid, Omar Sanseviero, Ahsen Khaliq, and the Hugging Face team [...] Hugging Face Spaces includes links to demos created by the community or the authors themselves. By go…
  • huggingface.co ↗ Demos on Hugging Face Spaces allow a wide audience to try out state-of-the-art machine learning research without writing any code. Hugging Face and ArXiv have collaborated to embed these demos directly along side papers on ArXiv! [...] Thanks to this integration, users can now fi…
  • en.wikipedia.org ↗ Hangzhou DeepSeek Artificial Intelligence Basic Technology Research Co., Ltd., doing business as DeepSeek, is a Chinese artificial intelligence (AI) company that develops large language models (LLMs). Based in Hangzhou, Zhejiang, DeepSeek is owned and funded by High-Flyer, a Chin…

Sources

Spot something wrong? Report an issue