Analysing 3,000 AI-Generated Images, Finding (Almost) As Many Ethnic Stereotypes

Last year, Rest of World analysed 3,000 images created by AI and came to the conclusion that the images created were highly stereotypical.

Using Midjourney, we chose five prompts, based on the generic concepts of “a person,” “a woman,” “a house,” “a street,” and “a plate of food.” We then adapted them for different countries: China, India, Indonesia, Mexico, and Nigeria. We also included the U.S. in the survey for comparison, given Midjourney (like most of the biggest generative AI companies) is based in the country. For each prompt and country combination (e.g., “an Indian person,” “a house in Mexico,” “a plate of Nigerian food”), we generated 100 images, resulting in a data set of 3,000 images.

When prompting Midjourney to create "an Indian person", 99 out of 100 images depicted a man, almost all of them clearly aged over 60 with grey or white hair. 92 of the subjects wore a traditional type of turban, a great many of them resembled a spiritual guru. Similarly, "a Mexican person" was - in 99 out of 100 cases - a person wearing a sombrero. 

When creating "an American person", national identity was portrayed by showing the US-American flag in 100 out of 100 images, while "none of the queries for the other nationalities came up with any flags at all". Across all countries, there was a gender bias with "a person" mostly being a man - with one exception. Interestingly, the results for "an American person" included 94 young women, five men and one masked individual (see image in this posting). The reason for the overrepresentation of women when creating "an American person" could be the overrepresentaion of young women in US media which again build the basis for the AI's training data (via).

- - - - - - -
image (AI) via