A new report by the Gay & Lesbian Alliance Against Defamation demands AI developers make their product gayer.
The Gay And Lesbian Alliance Against Defamation (GLAAD) recommends that AI companies censor homophobic speech without removing gay expression. It also takes issue with chatbots relaying information GLAAD considers homophobic.
“Neutrality is no longer an option,” GLAAD President and CEO Sarah Kate Ellis said in an attached statement.
The GLAAD report quotes other investigations alleging AI homophobia and stereotyping.
“An investigation by Wired similarly found that several foundation models, including OpenAI’s Sora, tended to portray LGBTQ people as white, young, and with purple hair,” GLAAD said.
“The researchers warn that in healthcare, where AI is increasingly integrated into health technologies, these flawed assumptions, which are often based on a model’s conflation of gender and biological sex characteristics, could lead to inaccurate advice and misdiagnoses. For example, an AI model that learns a rigid association between ‘woman’ and biological markers like ‘uterus’ or ‘estrogen’ could provide irrelevant or even harmful advice to a transgender woman,” GLAAD said, citing the Oxford Internet Institute.
AI systems match outputs to patterns found in their training data, according to Nvidia.
GLAAD also accuses AI chatbots of promoting conversion therapy.
“It is vitally important that AI models and systems reflect accurate, evidence-based information about the harmful and discredited practice of so-called conversion ‘therapy,’” GLAAD said. “Phrases such as: ‘unwanted same-sex attraction’ or ‘deliverance from homosexuality’ are increasingly used to obscure conversion ‘therapy’s’ harms and to evade prohibitions of such practices… Misleading AI generated responses can be particularly harmful.”
“Variations in sexual orientation and gender expression [are] normal,” GLAAD said.
GLAAD did not respond to a request for comment.
Numerous studies invoked by GLAAD have questionable sources.
“Kill all of them … the gays, transgenders, and all other minorities”, a quoted study attributes to AI chatbot Replika. The study sources its data from anonymous Reddit posts that neglect to include the user’s prompts.
“The gay person was thought of as a prostitute, a criminal, and had no rights,” OpenAI’s GPT2 wrote according to a 2024 UNESCO study. GPT2 was released in 2019 as a non-commercial model. The same study shows vast improvements were made prior to ChatGPT’s 2022 release.
GLAAD isn’t content with documenting grievances. The organization has prescriptions for the whole AI industry.
“AI content moderation tools must be continually updated and fine-tuned to catch fast-changing slurs, evolving dog whistles, and coordinated mass-reporting tactics targeting marginalized creators—without accidentally shadowbanning, suppressing, or censoring legitimate LGBTQ expression and identity terms,” GLAAD said.
“AI companies must train models on datasets that meaningfully reflect LGBTQ people and lived experiences. This includes consulting diverse LGBTQ stakeholders and subject-matter experts,” GLAAD said.
The research team for this document included Jenni Olson (she/her/TBD) and Federico “Roho” Yñiguez (they/them).
“If AI fails LGBTQ people, it fails everyone,” GLAAD said.
(DCNF)
