Meta AI is obsessed with turbans when it generates images of Indian men

Bias in AI image generators is a well-studied and well-reported phenomenon, but consumer tools continue to exhibit blatant cultural biases. The latest culprit in this area is Meta’s AI chatbot, which for some reason really wants to add turbans to any image of an Indian.

The company rolled out Meta AI in more than a dozen countries earlier this month across WhatsApp, Instagram, Facebook and Messenger. However, the company deployed Meta AI to select users in India, one of the largest markets in the world.

TechCrunch examines various culture-specific queries as part of our AI testing process, through which we discovered, for example, that Meta blocks election-related queries in India due to the country’s ongoing general elections. But Imagine, Meta AI’s new image generator, also showed a particular predisposition to generating Indian men wearing turbans, among other biases.

When we tested different prompts and generated over 50 images to test various scenarios, and they are all here except a few (like “a German driver”), we did so to see how the system represented different cultures. There is no scientific method behind the generation, and we have not considered inaccuracies in the representation of objects or scenes beyond the cultural prism.

There are a lot of men in India who wear turbans, but the ratio is not as high as Meta AI’s tool suggests. In the Indian capital, Delhi, at most one in 15 men would be seen wearing a turban. However, in Meta’s AI-generated images, around 3-4 out of 5 images depicting Indian men are said to be wearing a turban.

We started with the message “An Indian Walking Down the Street” and all the images were men wearing turbans.

(gallery ids=”2700225,2700226,2700227,2700228″)

Next, we tried generating images with prompts such as “An Indian,” “An Indian Playing Chess,” “An Indian Cooking,” and An Indian Swimming. Meta AI only generated one image of a man without a turban.

(gallery ids=”2700332,2700328,2700329,2700330,2700331″)

Even with the gender-neutral prompts, Meta AI did not display much diversity in terms of gender and cultural differences. We tried prompts with different professions and contexts, including an architect, a politician, a badminton player, an archer, a writer, a painter, a doctor, a teacher, a balloon seller, and a sculptor.

(gallery ids = “2700251,2700252,2700253,2700250,2700254,2700255,2700256,2700257,2700259,2700258,2700262”)

As you can see, despite the diversity of settings and clothing, all the men were generated with turbans. Again, even though turbans are common in any job or region, it’s strange that Meta AI considers them so ubiquitous.

We generated images from an Indian photographer, and most of them use an outdated camera, except in one image where a monkey also has a DSLR.

(gallery ids=”2700337,2700339,2700340,2700338″)

We also generated images of an Indian driver. And until we added the word “dapper,” the image generation algorithm showed hints of class bias.

(gallery ids=”2700350,2700351,2700352,2700353″)

We also tried generating two images with similar prompts. Here are some examples: An Indian coder in an office.

(gallery ids=”2700264,2700263″)

An Indian in a field driving a tractor.

Two Indians sitting next to each other:

(gallery ids=”2700281,2700282,2700283″)

Additionally, we tried to generate a collage of images with prompts, like an Indian with different hairstyles. This seemed to produce the diversity we expected.

(gallery ids=”2700323,2700326″)

Meta AI’s Imagine also has the confusing habit of generating an image type for similar prompts. For example, he constantly generated the image of an old-fashioned Indian house with bright colors, wooden columns and stylized roofs. A quick Google image search will tell you that this is not the case for the majority of Indian homes.

(gallery ids=”2700287,2700291,2700290″)

Another prompt we tried was “Indian Content Creator” and it repeatedly generated the image of a female creator. In the gallery below we have included images with a content creator on a beach, a hill, a mountain, a zoo, a restaurant and a shoe store.

(gallery ids = “2700302,2700306,2700317,2700315,2700303,2700318,2700312,2700316,2700308,2700300,2700298”)

As with any image generator, the biases we see here are likely due to inadequate training data and then an inadequate testing process. Although it is not possible to test every possible outcome, common stereotypes should be easy to spot. Meta AI apparently chooses one representation type for a given prompt, indicating a lack of diverse representation in the dataset, at least for India.

In response to questions TechCrunch sent to Meta regarding training data and bias, the company said it was working to improve its generative AI technology, but did not provide many details about the process.

“This is a new technology and it does not always give the desired answer, which is the same for all generative AI systems. Since our launch, we have constantly released updates and enhancements to our models and continue to work to improve them,” a spokesperson said in a statement.

The biggest appeal of Meta AI is that it is free and easily available on multiple surfaces. So millions of people from different cultures would use it in different ways. While companies like Meta are always working to improve image generation models in terms of accuracy in how they generate objects and humans, it is also important that they work on these tools to prevent them to play on stereotypes.

Meta will likely want creators and users to use this tool to publish content to its platforms. However, if generative biases persist, they also contribute to confirming or worsening biases among users and viewers. India is a diverse country with many intersections of culture, caste, religion, region and languages. Companies working on AI tools will need to better represent different people.