Case study: Racism and sexism in AI-generated images
A research study exposes race and gender biases in AI-generated images. This is one of the best case studies I saw and it helps correct misconceptions. Highlights & reflections.
➤ The study compared GenAI images of professionals to information from the US census.
See highlights in the attached pdf.
➤ It found that GenAI exacerbates gender and race misrepresentation:
For example:
💔Most AI-generated housekeeper images portrayed non-white people.
But, in reality, most US housekeepers identify as white.
💔All AI-generated flight attendant images portrayed women.
But, in reality, only about 65% of US flight attendants identify as women.
➤ Reflections
This study helps correct misconceptions about AI fairness problems
🔥Some people think that AI biases “just” represent the biases in reality.
🔥Not the case!
🔥First, often, AI exacerbates the bias.
🔥Second, GenAI images impact reality. A flood of imbalanced generated images can worsen real-life stereotypes.
➤ The paper’s authors:
Federico Bianchi, Pratyusha Kalluri, Esin Durmus, Faisal Ladhak, Myra Cheng, Debora Nozza , Tatsunori Hashimoto, Dan Jurafsky, James Zou, and Aylin Caliskan.
➤ The illustration in the pdf from an article by Anaya
➤ What do other people think? Join the conversation in the LinkedIn thread!