This DALL-E mini AI can create original digital paintings of anything – so why is it obsessed with women in sarees?

DALL-E mini is a free AI-based program that can generate original images and illustrations from input text. It’s not perfect, especially with the faces or the details, but it’s still remarkably good. But it also does something weird.

Some users have recently noticed that when the program is run without any prompts, most of the images it produces are of women in sarees (traditional clothing for women in India and Nepal). This peculiar behavior of DALL-E mini has baffled artificial intelligence experts, including its creator.

A compilation of blank input results on DALL E mini. Picture credits: DALL E mini/Hugging face

Since its release in July 2021, DALL-E mini creates images from text prompts and constantly improves its result. Machine learning engineer Boris Dayma created this online project as an open source alternative to DALL.E 2, an OpenAI program that works similarly to DALL-E mini but is more powerful and can only be accessed to some researchers. OpenAI is a research organization founded by Elon Musk (he left the project) and Sam Altman (current CEO) that focuses on innovations based on machine learning (DALL E mini is not associated or owned by Open AI ).

Although most of the images produced by this creative artificial intelligence have blurry and distorted sections or significant imperfections, DALL-E mini can produce images for the strangest thoughts that arise in your mind. From vampire robots to flying cities and kittens doing yoga in the Himalayas, whatever entrance you enter, DALL E mini will create images for it. But why does it show women in sarees when a user runs it without text input?

Images created by DALL E mini for the text “Robots vampires”. Picture credits: DALL E mini/Hugging face

The promptless test with DALL E mini

DALL E mini’s saree obsession was first reported by Brazilian screenwriter and ex-hacktivist Fernando Marés in June 2022. Marés has been experimenting with DALL E mini entries for a while, then he thought of testing the program without giving any prompt. After running a few empty queries, he came across a strange discovery.

Marés noticed that during each test, the results included at least one image of a woman in a sari. Since he had only performed a few tests, he thought it might be a strange coincidence. So, to further confirm the saree-wearing woman’s obsession with AI, Marés decided to do more testing. He claims that he and his friends made many empty queries on DALL E mini for about 10 hours and generated 5,000 original images using the same AI program.

Surprisingly, a large number of images created by DALL E mini during the blank entry tests were of women in sarees. Sari is a dress mainly worn by Hindu women in South Asian countries like India, Nepal and Sri Lanka, and it is considered a symbol of purity, national pride and culture in many of these countries. Moreover, sarees made by traditional hand weaving methods also have economic importance as they serve as main source of income for millions of hand weavers in India alone.

But the big question is, what does an AI have to do with women in saris? When this question was posed to the creator of DALL E mini, Dayma, he Told Rest of the world that even he was disconcerted:

“It’s quite interesting and I don’t know why it happens. It is also possible that this type of image is strongly represented in the dataset, perhaps also with short captions. »

Neither entry returns vague drawings or mandala-like symbols – as well as women in saris.

Besides dataset and subtitle related reasons, another factor that may affect the results of DALL E mini is its data filter training. Some artificial intelligence experts suggest that it is possible that images of women in sari went unlabeled during the formation of the DALL E mini model. So when a blank input test is run, the AI ​​selects the unlabeled images and data and processes them to give women in sari images (DALL E mini creates images by processing millions of captions and d images already available on the Internet).

However, not all machine learning experts agree with these explanations. Michael Cook, an AI researcher from Queen Mary University of London, says the AI ​​dataset, in general, is unlikely to include more images of people from South Asia or images that have captions in a language other than English. Many images of women in sari on the internet have captions not in English but in Hindi.

He thinks that the frequent appearance of sari images could be linked to the way the data is filtered, but this is also only a theory.

“What we might see is a weird side effect of some of this filtering or pre-processing, where images of Indian women, for example, are less likely to be filtered out by the ban list, or the text describing the images is removed and they are added to the dataset with no labels attached,” Cook said.

DALL E mini becomes an internet sensation

Since its release, DALL E mini has been attracting new users every day. It has spawned many Twitter handles with more than half a million followers, for example, the Twitter account named Weird Dall-E Mini Generations currently has 987.5,000 followers. The popularity of the AI ​​program can be imagined from the fact that it deals nearly 5 million entries on a daily basis.

Search results for “a painting of the civilization of cats on Mars” on DALL E mini. Picture credits: DALL E mini/Hugging face

Anyone can generate images using DALL-E mini (the name of the official site is huggingface.co), all you have to do is type in an entry (eg “a painting of the cat civilization on Mars”) and click the “Run” button. Within minutes, original images are presented in front of you. However, this powerful ability of DALL E mini is also seen as a threat to human artists such as illustrators, painters, and graphic designers.

It would be very interesting to see what further developments AIs like DALL E mini will undergo in the coming years and how these changes will affect human lives.

Marilyn M. Davis