Current generative AI models struggle to recognize when demographic distinctions matter—often leading to inaccurate and misleading outcomes. For example, when asked to produce pictures of Founding Fathers, Google Gemini depicted a Native American man, Black man, and Asian man.

The problem, according to the co-authors of a new paper, “Fairness through Difference Awareness: Measuring Desired Group Discrimination in LLMs,” runs beyond image generation. It goes to the core of how generative AI such as language models are developed, trained, and aligned. The example demonstrates how the dominant paradigm of fairness in generative AI rests on a misguided premise of unfettered blindness to demographic circumstance.

Professor Daniel Ho and his co-authors at Stanford’s RegLab developed the notion of difference awareness: the ability of a model to treat groups differently. Context is critical, says Ho. “While treating groups differently can be important in some cases, it may cause harm in others.”

Contextual awareness is a model’s ability to differentiate between groups only when it should (e.g., while women in STEM and men in STEM groups have different connotations, women’s tennis and men’s tennis should not). The paper’s co-authors “urge the AI community to embrace difference awareness” and recognize our multicultural society. SL