How AI Perpetuates Gender Bias

There is plenty to be concerned about when it comes to artificial intelligence, from its threat to job security to its impact on education and beyond. But one concern that seems less often discussed? The impact of AI’s gender bias. Women are severely underrepresented among those creating AI and facilitating how it is taught and used in our society; women comprise only 12 percent of AI researchers, 29 percent of the AI workforce, 6 percent of software developers, and less than 20 percent of AI professors. If we are to use AI ethically, we must prioritize employing more women in AI development.

Representation matters in AI because machine learning processes are often encoded with programmers’ own biases, even if inadvertently. Something as simple as using a skewed dataset in which female speakers are underrepresented could make a machine learning model less accurate when working with women. The Berkeley Haas Center found that 44 percent of 133 AI systems from various industries displayed gender bias, and 25 percent showed both gender and racial bias.

The effects of AI gender bias are far-reaching and can be quite serious. For instance, AI tools for medical diagnosis are less accurate for female patients, worsening women’s health outcomes, and AI hiring systems systematically penalize women who took time off from their careers to raise children. AI lending algorithms limit women’s financial independence by denying them credit or offering them worse terms than otherwise identical male applicants. Furthermore, risk assessment algorithms used in the justice system are disproportionately harsh towards women of color. AI tools used to screen out prohibited images on social media are more likely to “shadowban” images of women’s bodies, even in a completely non-sexual context, reifying the harmful sexualization of female bodies. I could go on, but the point is that in the status quo, AI structurally disadvantages women.

What’s more, white men stand to disproportionately financially benefit from AI as an industry. From OpenAI’s Sam Altman to Meta’s Mark Zuckerberg, almost all of the trailblazing companies in AI are led by male CEOs, and the world of AI suffers from a “diversity crisis” at large.

It’s worth considering how similar arms races have historically affected marginalized communities. For example, white male captains of industry made their fortunes during the Industrial Revolution by subjecting children, immigrants, and other vulnerable populations to horrific working conditions. With breakthroughs in long-distance transportation in the late nineteenth century, European powers vied for imperial control of Africa’s abundant natural resources not only subjecting the continent’s native population to conquest, but also fostering hostile dynamics that would later influence World War I.

There need to be global efforts to assess how AI threatens to further marginalize women and minorities. The 2024 Global Digital Compact, a United Nations framework promoting international cooperation surrounding AI, is a good starting point, but more comprehensive initiatives empowering women in AI training and development are required.

We should build upon the UNESCO Recommendation on the Ethics of AI, which was endorsed by major tech companies like Microsoft and Salesforce. This entails funding gender equality initiatives in tech companies, female entrepreneurship initiatives, and programs promoting women’s involvement in STEM and ICT fields. Governments should financially incentivize diverse hiring practices in AI companies, though this may be particularly difficult in the U.S. considering the recent assault on DEI practices.

Additionally, as suggested by Margaret Mitchell, the chief ethics scientist at Hugging Face and a former leader of Google’s Ethical AI research, tech companies should thoroughly check their algorithms and data-labeling for biases. Various governments, private companies, and other entities must cooperate with the common goal of promoting accountability and protecting human rights.

This moment of transition to the age of AI poses a fleeting opportunity to encode gender equality into a multitude of systems moving forward. Ensuring that AI is developed by diverse teams using data that accounts for varied perspectives is critical to preventing AI from reproducing existing biases that harm women on a global scale. A century from now, we could reflect on how AI worsened gender inequality or we could act now. We should learn from our historical mistakes and employ more women in the development of what might be humanity’s most promising technological development yet.