(Reflections on UNESCO’s Outlook Study on Artificial Intelligence and Gender in South Asia)
Last week, I was invited to the launch of UNESCO’s Outlook Study on Artificial Intelligence and Gender in South Asia at the UNESCO Office in New Delhi.
As we design, build, and train LLMs, we are baking in the biases already present in our society while creating the risk that the AI layer influencing our lives becomes even more biased than the society it learns from.
What does that mean?
The AI layer shaping our society could eventually become more biased than the humans it learns from if we fail to identify, test, and address these biases continuously throughout the AI lifecycle.
To reduce risks such as embedded bias, hallucinations, and harmful outputs, governance, testing, and risk management must be applied throughout the model development lifecycle, not only after deployment.
As Responsible AI practitioners, one question keeps returning: Do we have sufficient legal and governance guardrails when organizations build or deploy AI systems?
Studies like this play an important role in strengthening accountability. They provide evidence that helps policymakers, researchers and organizations identify systemic risks and reduce harmful bias before it becomes embedded at scale.
One of the biggest challenges in global AI governance is balancing speed, complexity and local context. Training data is often collected across countries and cultures without sufficient attention to gender, caste, language, cultural diversity, or equitable representation.
AI is becoming part of everyday life. Before organizations procure and deploy AI systems, they should understand how the underlying models were trained, what governance processes were followed, how the systems were evaluated and how their recommendations could affect citizens’ access to public services.
Ultimately, this is about more than model performance. It is about preserving public trust in AI.
In this context, both the UNESCO Recommendation on the Ethics of AI and periodic studies such as this are essential. They help us understand how AI is evolving, who is participating in shaping it and where governance and inclusion still need attention.
About the report
This is UNESCO’s first regional assessment of women’s participation across the AI ecosystem in Bangladesh, Bhutan, India, Maldives, Nepal, and Sri Lanka.
Its central finding is striking:
South Asia is producing more educated women than ever before, yet very few are reaching positions where AI is designed, governed, funded, or led.
Top six findings
- Women remain underrepresented across the AI ecosystem, particularly in AI engineering, research, entrepreneurship, and leadership.
- The AI talent pipeline is “leaky,” with women’s participation steadily declining from STEM education to AI careers and senior leadership.
- Women contribute to AI research but are far less likely to lead it, limiting their influence on innovation and policy.
- AI systems can reinforce existing gender biases through biased training data, limited diversity in development teams, and discriminatory outcomes in hiring, finance, healthcare, and public services.
- Women-led AI startups continue to face barriers to funding, mentorship, investment, and professional networks.
- Gender-responsive AI governance is essential, supported by fairness, transparency, accountability, inclusive participation, and better gender-disaggregated data across the AI lifecycle.
Download the full report here: https://tinyurl.com/5ydujdhd
This article was originally published on LinkedIn here
