Latent Identities is a research direction that seeks to understand how Generative AI models learn and represent identity labels in their latent spaces. By investigating the ways individuals self-identify and the prompts and labels they use to describe their identities, researchers can uncover the elements that people consider vital but are often overlooked in AI training datasets. This information can be used to develop more inclusive and accurate representations of diverse identities, mitigating the biases and stereotypes that exist in current AI models.

​

The goal is to identify and challenge the stereotypical biases that are embedded in Generative AI models. By comparing self-definitions with labels assigned by AI, researchers can highlight the biases and inaccuracies that exist in these models. This comparative analysis will help researchers understand what AI gets "right," what it gets "wrong," and what it fails to consider, enabling them to develop more fair and inclusive AI systems.

​

One of the key aspects of this research is its focus on the representation of underrepresented groups. By examining the ways in which AI models learn to represent these groups, researchers can identify and challenge the visual stereotypes that are perpetuated through media usage and stock photo representations. This research can inform the development of alternative image banks and tools that are generated with citizen involvement, leading to more inclusive representations in mainstream media. Ultimately, I aim to create AI systems that promote diversity and inclusion, rather than marginalizing underrepresented communities.