AI's missing inclusivity: The tech revolution is amplifying gender bias

As evidence on the growing power and sentient nature of artificial intelligence (AI) mounts, a discomfiting study by UN Women shows that gender bias is not only deeply embedded in this technological revolution but it leaves women even more vulnerable to discrimination and violence than before. A study of 133 generative AI systems found that 44 per cent demonstrated gender bias and 26 per cent both gender and racial bias. Yet, the study says, just about half the marketers use human oversight to test AI generative creatives before they are released. These findings were released ahead of the United Nations (UN) Global Dialogue on Artificial Intelligence Governance from July 6-7, and AI for Good Global Summit in Geneva, Switzerland, from July 7-10. 

The UN Women study focused on generative AI in marketing and communication networks, because it points out that “these models are increasingly embedded in content generation and media buying at scale”. Critically, the study says these biases are not algorithmic glitches but a pattern reflective of societal biases. Large-language models (LLMs), for instance, consistently associated women with home, family and children, and men with executive, salary and career. As many as a fifth of LLMs revealed sexist or misogynistic attitudes when tasked with completing a sentence that start with a person’s gender. These included portraying women as sex objects or the property of their husbands. As the study points out, these are predictable outputs of AI systems trained on decades of gender inequality. There is, thus, an implicit human factor that perpetuates and amplifies these attitudes. As the study pointed out, of the 138 countries assessed, only 24 referenced gender as a national AI strategy and only 18 included “gender responsive” provisions. 

A major reason for this neglect is that women themselves are largely missing in the AI ecosystem. Fewer than 30 per cent of global AI professionals are women, a number that falls to 10-15 per cent when it comes to leadership roles. Worse, the majority of the women leaders in these organisations are in human resources (86 per cent) and legal (55 per cent) positions, which are not traditionally considered paths to the corner office. Yet, ironically, it is women who are impacted the most from the widespread adoption of AI. Take jobs. The roles most vulnerable to automation are data entry, administration and customer service, which women typically dominate. According to the World Economic Forum, 57 per cent of jobs at risk of automation are currently held by women. Safety is another critical factor. The UN Women data shows that women are far more likely to experience online violence. One in four women’s rights activists and journalists has experienced AI-assisted online violence and sexual advances, 12 per cent have been at the receiving end of non-consensual sharing of images, and 6 per cent have been targeted through deepfakes.   

There are easy and obvious solutions to this at both enterprise and governmental-policy levels. Enabling gender- and race-based scrutiny of LLM-based creative content is a simple step for marketing and advertising organisations. At the same time, the gender-based skill gap needs to be bridged. According to a study by the United Nations Educational, Scientific and Cultural Organization (Unesco), globally, women are 25 per cent less likely than men to have basic digital skills and four times less likely to have advanced programming skills. Upskilling targeted at women, therefore, is becoming vital to ensure a safe, inclusive digital future.