AI Models Are Learning Anti-LGBTQ+ Bias and Misinformation, GLAAD CEO Warns
Because of anti-LGBTQ+ rhetoric and misinformation online, AI models are at risk of learning that bias, Sarah Kate Ellis said.
Featured image: Sarah Kate Ellis; Photo by Christopher Polk/Billboard via Getty Images
The CEO of GLAAD, the LGBTQ+ media and rights group, has warned that AI can cause harm to LGBTQ+ people by reinforcing stereotypes of queer people as well as spreading misinformation about them.
At the Axios AI+ NY summit on Wednesday, Sarah Kate Ellis highlighted new findings by the organization in a conversation with Ina Fried, Axios’s chief technology correspondent.
“I think you’d have to be asleep at the wheel to not see the attacks that are happening against the LGBTQ community, mostly the trans and nonbinary community, but it’s all happening together,” she told Fried, Axios reports.
“What that means, though, is that as these AI models are learning, they’re learning with this information and rhetoric coming into them,” Ellis explained.
The GLAAD leader said that models like ChatGPT and Claude, which Axios notes are foundational AI models, need to have the correct information and prioritize accuracy.
She further told Fried that companies have a responsibility to protect users’ privacy.
“AI models can detect who you are by the questions you ask and what you’re doing, and in 60 countries, it’s still illegal to be LGBTQ+. So if you’re criminalized in a country and it’s now identifying you as LGBTQ, there is some real risk there,” Ellis said, according to Axios.
Ellis noted that if AI can’t accurately provide information about LGBTQ+ people or topics, then it’s a failure.
“If you just think about putting 1% of the money that all these folks are raising to help fix the foundation of this, it would go an enormous way to make this a safer platform, not just for LGBTQ people, but for everyone and every family,” she said. “Trust is the biggest currency with AI, and whoever wins this trust war wins the AI war.”
