Nitasha TikuMay 31, 2024
My latest (and greatest??) data visualization with Szu Yu Chen looks at how generative AI models interpret female beauty. We found that prompted to show a “beautiful woman," DALL-E, Midjourney, & Stable Diffusion generated 100% thin women and only 9 percent had dark skin tones https://www.washingtonpost.com/technology/interactive/2024/ai-bias-beautiful-women-ugly-images
What AI thinks a beautiful woman looks like: Mostly white and thin

Despite the growing profusion of AI image generators, they all had remarkably similar responses when The Post directed them to portray a beautiful woman.

The Washington Post
Nitasha TikuJun 1, 2024
Show threadRobert W. Gehl

@nitashatiku This is not surprising to me. In 2016, some colleagues and I surveyed the entirety of the academic literature on porn detection via computer vision and found that the algorithms that were built were heavily biased towards images of thin, white women:

https://journals.sagepub.com/doi/abs/10.1177/1527476416680453

So it's not surprising to hear the process works the same in reverse.

May 31, 2024 at 8:41pmWeb