What "person" means to GenAI
A great paper about generative AI bias! This one shows what “person” means to generative AI. We can push tech companies to do better!
➤ The paper:
‘Person’ == Light-skinned, Western Man, and Sexualization of Women of Color: Stereotypes in Stable Diffusion
By Sourojit Ghosh and Aylin Caliskan
Read it here
➤ Highlights:
😔Images of "person" tend to be male and white
😔Images of women, especially women of color, tend to be sexualized
Summary (charts by me):
➤ Reflections:
💡Bias in Generative AI can distort reality even in seemingly neutral prompts, such as prompts about “persons”.
💡Human biases may worsen due to increased exposure to distorted representations as generated images flood the internet.
💡Generative AI can be better. As consumers and buyers, we should demand transparency and accountability to bias performance.
→ Choose your platforms carefully
→ Ask about bias performance in your due diligence
→ Require bias accountability in your contracts
➤What do other people think? Join the conversation in the LinkedIn thread!