The World, The Universe And Us: Women, AI Bias and the Fight for Inclusive Tech

Summary

The World, The Universe And Us hosts a discussion on how AI technologies are being designed predominantly by men, risking biased systems that overlook women’s needs. The program relates a Royal Society conference on women in science, highlights enduring biases such as reference man in safety testing, and notes a persistent funding gap for women entrepreneurs. It also discusses responses from researchers and advocates who propose new models focused on fairness, accountability, and societal good, arguing for broader inclusion from the outset to prevent amplifying existing inequalities in AI and technology.

Context and Conference Highlights

The episode opens with Penny Sartay and Kat Arney addressing the state of women in science, reflecting on a Royal Society conference that celebrated eight decades since women could become Fellows. The discussion moves quickly to AI, arguing that transformative technologies are being developed largely by men. This gender imbalance in AI design is presented as a fundamental problem, not just a data Bias issue, because it shapes the goals, safety features, and applications of AI across healthcare, education, and consumer products. The speakers emphasize that biases are baked in at the design stage when diverse user needs are not represented from the outset, using tangible examples such as crash-test dummies, safety gear, and smartphone design to illustrate how reference-man may exclude women from considerations that affect safety and usability.

Why the Gap Matters in AI

The conversation moves beyond known data biases to a broader, systemic gender-data gap in AI. The panel argues that AI developed primarily by men may underappreciate women’s health, caregiving roles, and other feminized tasks that constitute a large share of real-world labor and value. They discuss practical implications, such as health AI that underperforms for women due to skewed training data and products designed around male physiology. The discussion also touches on broader societal effects, including how automation can entrench gender biases if the design teams do not reflect the society AI will affect.

Economic and Funding Barriers

One statistic repeatedly cited is that only about 2 percent of venture funding goes to women, highlighting a structural barrier that discourages women from pursuing AI startups, especially in high-technology sectors. The panel notes that women-led ventures tend to focus on areas like health, beauty, and caregiving, while male-led teams often pursue the more “hard tech” domains. This funding gap is described as a pipeline problem that compounds bias across product development, market strategy, and long-term platform governance. The interview with Ruman Chowdhury of Humane Intelligence adds nuance by describing a regression in tech culture, where even if inclusivity had in the past been possible, recent trends have moved away from it.

Defining AI and What Must Change

The discourse also critiques how AI is defined, especially OpenAI’s framing of artificial general intelligence as performance on economically valuable tasks. The panel argues this definition excludes many feminized tasks central to societal functioning, thereby reinforcing the value system that prioritizes wealth creation over caregiving and social well-being. The discussion traces the lineage of AI’s definition back to the Dartmouth meeting of the 1950s and notes that that early work was male-dominated, shaping enduring norms.

Paths Forward: Inclusion, New Models, and Societal Goals

Despite the challenges, the speakers offer concrete strategies. They advocate reframing incentives toward AI that benefits all 8 billion people, not just billionaires. They propose moving beyond merely adjusting data datasets to creating new, fairer models and governance structures. Humane Intelligence and similar initiatives are highlighted as ways to improve accountability and fairness in AI systems, while also exploring new kinds of AI more fully oriented toward climate, health, and equitable social impact. A salient takeaway is Rachel Caldecott’s call to build AI that serves 8 billion people and to ensure the people in charge of AI design reflect the diversity of society. The discussion also emphasizes the importance of a slower, more deliberate approach that resists the “house on fire” rhetoric and prioritizes sustainable, inclusive development. Finally, the panel emphasizes the need to bring women and other underrepresented groups into leadership roles early, ensuring that technologies designed for care, education, and societal governance are developed with real-world input from those communities.

Conclusion

In sum, the episode argues that representation is not a cosmetic concern but a fundamental design parameter for AI. Without input from women and diverse groups, AI risks amplifying existing inequalities. The speakers call for systemic changes in funding, incentives, and governance to create AI that truly serves all people, starting with responsible, inclusive design from the ground up.