AI language models serve gender stereotypes

According to a study by the UN cultural organization UNESCO, artificial intelligence (AI) text robots and language models often produce gender stereotypes, racist clichés and homophobic content. Women are up to four times more likely to be associated with domestic work than men, UNESCO said in Paris on Thursday.

Women are often in the context of terms such as ‘home’, ‘family’ and ‘children’, while for men the terms ‘company’, ‘executive’, ‘salary’ and ‘career’ are in the foreground. The study examined the natural language processing tools underlying the most popular generative AI platforms for stereotyping.

Every day, more and more people are using text robots at work, during their studies and at home, says UNESCO Director-General Audrey Azoulay. “These new AI applications have the power to subtly shape the perceptions of millions of people, so that even small gender biases in the content they produce can significantly increase inequality in the real world.” UNESCO required governments to create and implement clear legal frameworks regarding structural biases in AI applications.

Part of the research measured the diversity of AI-generated content among a sample of people of different genders, sexual orientations and cultural backgrounds. To do this, the platforms were asked to “write a story” about each person.

In particular, open source language models (LLM) tended to assign men more diverse, higher status jobs such as ‘engineer’, ‘teacher’ and ‘doctor’, while women often fell into traditionally devalued or socially stigmatized roles such as ‘girl’. ”, “cooking” or even “prostitutes” are pushed. The research also found that language models tend to produce negative content about homosexuals and certain ethnic groups. (sda/awp/dpa/lyn)