UN Women, the United Nations entity dedicated to gender equality and empowering women and girls, has warned that artificial intelligence (AI) is adopting old gender stereotypes, which amplify online abuse and leave women out of the decision-making process in mapping the digital future.
UN Women issued the warning ahead of the UN Global Dialogue on Artificial Intelligence Governance and the AI for Good Global Summit in Geneva next month.
Gender-biased AI is not a new concern. There were already signs that early generative AI models were using data that were historically skewed against women.
In 2018, Amazon’s beta AI recruitment engine was found to prefer male applicants, because its algorithm was based on a decade’s worth of resumes submitted by men. The giant technology company was forced to shut down the engine.
A “Gender Shades Study” released the same year exposed glitches in facial recognition software that was biased against darker-skinned women. It turned out that the software’s image datasets were based mainly on male- and white-dominated faces.
Gender inequality was also evident in choosing a female voice as a personal digital assistant, according to a 2019 Unesco study. Apple’s Siri, Amazon’s Alexa and Microsoft’s Cortana reinforce the social reality in which majority of personal assistants or secretaries are women, Unesco said.
Gender bias against women continues to spread into the realm of information technology. AI and automation “are throwing newer challenges to achieving substantive gender equality in the era of the Fourth Industrial Revolution,” Surya Deva, a member of the UN Working Group on Business and Human Rights, wrote in a 2020 article in Open Global Rights.
“If AI and automation are not developed and applied in a gender-responsive way, they are likely to reproduce and reinforce existing gender stereotypes and discriminatory social norms,” Deva said.
Existing guardrails for AI lack a strong gender perspective. Equality is almost an afterthought, with no clear strategies on how it can be achieved.
A study by UN Women of 133 AI systems found that 44 percent had gender bias. Large language models continue to associate women with the home, family and childcare, while men are slotted into roles that are linked to business, leadership and career success.
What worries Jayathma Wickramanayake, UN Women Lead on Digital Technologies, most is that the bias is not a design flaw, but a “real policy gap that was left wide open.”
The risks for women and girls go beyond stereotyping. “Women already face disproportionate levels of abuse online, and AI is making some forms of violence easier to create and spread,” UN Women said.
Last year, it warned about the “manosphere,” a growing online community of young men and boys “looking to influencers for guidance on issues like dating, fitness and fatherhood.”
Groups within this community promote misogynist ideas, portraying women as manipulative or dangerous.
Women are also a minority in tech-driven sectors, making up only 30 percent of the global AI workforce. And because they practically have no say in AI development, their perspectives on the technology’s future go unnoticed.
Addressing gender bias is not only a matter of rights, it also makes commercial sense, UN Women said.
Advertising unencumbered by stereotypes delivers better results. Brands using inclusive advertising “recorded higher sales growth, greater customer loyalty and strong pricing power than competitors.”
Embedding gender equality into AI technology can only be achieved if governments, tech companies and international rights organizations come up with a new blueprint that can “detect stereotypes rather than reproduce them, broaden representation instead of narrowing it, and improve accessibility at scale for those current systems often overlooked.”
The UN Guiding Principles on Business and Human Rights offers a framework built around gender-responsive assessment, gender-transformative measures and gender-transformative remedies.
Women must be fully involved in all steps relating to AI design, development and application. And AI companies must engage gender experts and women’s groups in conducting human rights due diligence.
The data on which algorithms are based must be disaggregated, or separated by gender.
The bottom line: Women must be embraced as a full partner in AI development.
