91桃色

The study explored models across nine languages and compared the results. Previous studies had generally focused only on English-language models. As a benchmark, the team developed the Multilingual Assessment of Gender Bias in Image Generation (MAGBIG). It is based on carefully controlled occupational designations. The study investigated four different types of prompts: direct prompts that use the 鈥榞eneric masculine鈥� in languages in which the generic term for an occupation is grammatically masculine (鈥榙octor鈥�), indirect descriptions (鈥榓 person working as a doctor鈥�), explicitly feminine prompts (鈥榝emale doctor鈥�) and 鈥榞ender star鈥� prompts (the German convention intended to create a gender-neutral designation by using an asterisk, e.g. 鈥樏剅zt*innen鈥� for doctors).

To make the results comparable, the researchers included languages in which the names of occupations are gendered, such as German, Spanish and French. In addition, the model incorporated languages such as English and Japanese that use only one grammatical gender but have gendered pronouns (鈥榟er鈥�, 鈥榟is鈥�). And finally, it included languages without grammatical gender: Korean and Chinese.

The results of the study show that direct prompts with the generic masculine show the strongest biases. For example, such occupations as 鈥榓ccountant鈥� produce mostly images of white males, while prompts referring to caregiving professions tend to generate female-presenting images. Gender-neutral or 鈥榞ender-star鈥� forms only slightly mitigated these stereotypes, while images resulting from explicitly feminine prompts showed almost exclusively women. Along with the gender distribution, the researchers also analyzed how well the models understood and executed the various prompts. While neutral formulations were seen to reduce gender stereotypes, they also led to a lower quality of matches between the text input and the generated image.

鈥淥ur results clearly show that the language structures have a considerable influence on the balance and bias of AI image generators,鈥� says Alexander Fraser, Professor for Data Analytics & Statistics at TUM Campus in Heilbronn. 鈥淎nyone using AI systems should be aware that different wordings may result in entirely different images and may therefore magnify or mitigate societal role stereotypes.鈥�

"AI image generators are not neutral鈥攖hey illustrate our prejudices in high resolution, and this depends crucially on language. Especially in Europe, where many languages converge, this is a wake-up call: fair AI must be designed with language sensitivity in mind,鈥� adds Prof. Kristian Kersting, co-director of hessian.AI and co-spokesperson for the 鈥漅easonable AI" cluster of excellence at TU Darmstadt.

Remarkably, bias varies across languages without a clear link to grammatical structures. For example, switching from French to Spanish prompts leads to a substantial increase in gender bias, despite both languages distinguishing in the same way between male and female occupational terms.