UN Women has issued a critical warning that the rapid global expansion of artificial intelligence is significantly increasing the risk of discrimination and digital violence against women, urging immediate action to address inherent gender biases in AI systems.
IMAGE: A demonstrator looks on during a rally to mark the International Women’s Day. Photograph: Cristina Sille/Reuters
Key Points
- UN Women warns that AI is increasing women’s risk of discrimination and violence, with studies showing significant gender bias in AI systems and large language models.
- Analysis by the Stanford Social Innovation Review found 44 per cent of 133 AI systems exhibited gender bias, while a Unesco study revealed 20 per cent of LLM responses were sexist or misogynistic.
- Experts note that AI reflects existing societal biases, associating women with ‘home’ and ‘family’ and men with ‘business’ and ‘career’, highlighting the challenge of AI evolving beyond these prejudices.
- AI is intensifying digital violence against women, with deepfakes and non-consensual sharing of personal images disproportionately targeting women and girls.
- Only a small fraction of national AI strategies reference gender, and even fewer include substantive gender-responsive provisions to detect stereotypes and improve representation.
From what we do to how we do it, artificial intelligence (AI) is now rewriting realities for billions of people across the globe.
However, UN Women has warned, citing multiple studies, that the AI revolution is leaving women at a higher risk of discrimination and violence.
In a media advisory, it cited an analysis by the US-based Stanford Social Innovation Review, which found that 44 per cent of 133 AI systems demonstrated gender bias.
UN Women is a United Nations entity focused on gender equality and empowerment.
AI’s Reflection of Societal Biases
Another study by the United Nations Educational, Scientific and Cultural Organization (Unesco), cited by the organisation, revealed that 20 per cent of responses generated by large language models (LLMs) exhibited sexist and misogynistic attitudes.
“The models were found to associate women with terms such as “home”, “family”, and “children”, and men with “business”, “executive”, and “career”.
“Honestly, it is not that LLMs are fed certain sexist remarks.
“They are a reflection of existing biases and prejudices in society.
“It will show a doctor as a male and a nurse as a female, and that is because of the generic perception of women as caregivers,” says Jibu Elias, India lead at Mozilla Foundation, a US-based non-profit organisation promoting an accessible and inclusive internet.
“The bigger challenge here is how AI will actually know that society is evolving,” he adds.
The warning by UN Women comes as the global community prepares for the UN Global Dialogue on AI Governance, scheduled for July 6–7, and the AI for Good Global Summit in Geneva, Switzerland, from July 7-10.
Exclusion in AI Design and Development
According to Osama Manzar, founder and director of the Digital Empowerment Foundation, the digital ecosystem has historically deprived women, and AI is adding another layer to that disadvantage.
“The real question should not be whether women are using AI, but whether women are considered right at the levels of design, development, deployment, and adoption,” he says.
On the biases reflected by the technology, he observes that AI is built on large datasets and that if data is not collected, stored, and used ethically in the first place, LLMs will only amplify those biases.
As UN Women views safety as a key component at the design stage itself, it also said that, in a study of 138 countries, only 24 referenced gender in a national AI strategy.
Moreover, just 18 included substantive gender-responsive provisions that can help AI detect stereotypes, broaden representation, and improve accessibility at scale.
AI’s Role in Digital Violence
The advisory also highlighted AI’s role in intensifying digital violence against women.
Previous reports by the organisation found that nearly one in four surveyed women human rights defenders, activists, and journalists had experienced AI-assisted online violence, while 12 per cent reported experiencing the non-consensual sharing of personal images.
“AI is compounding this. Deepfakes are among the most visible examples of AI-enabled abuse that disproportionately targets women and girls,” the UN Women advisory notes.
Six per cent of women have been targeted through deepfakes or manipulated images and videos, while more than one in four have received unsolicited sexual advances through digital messaging.
According to the experts, it is important to know who is building the technology before we talk about the users.
“From product teams to research labs and then bodies that implement standards, the representation of women is low and any technology is a reflection of its creator,” says Elias.
