This piece was written by Dr Sarah Okonkwo, Executive Director of Women AI Labs, and originally published in MIT Technology Review.
There is a quiet irony at the heart of the AI for Good movement. Its stated mission — applying artificial intelligence to address the world's most pressing challenges — is fundamentally about serving underrepresented and marginalised communities. Yet the practitioners who lead it remain overwhelmingly drawn from the most privileged demographics in the global technology sector.
This is not a superficial criticism about optics. It is a substantive critique about epistemology: who has the lived knowledge to identify the right problems, ask the right questions, and design solutions that genuinely work for the people they claim to serve?
The Data Is Clear
According to the 2025 AI Index, women account for fewer than 20% of AI researchers globally. Black and Hispanic researchers together account for less than 5% of faculty at top US AI programmes. The pattern holds in the NGO and impact sector: a 2024 survey of 60 leading AI for Good organisations found that 73% of technical staff identified as white, and 68% as male.
What This Means in Practice
When a team of demographically homogeneous researchers builds a disease prediction model for a population unlike themselves, they carry blind spots into every design decision: which data signals to treat as proxies for health outcomes, which languages to support, which cultural contexts to account for in consent design. The result is not malice but systematic epistemic gap — and the communities most harmed by AI systems are the ones least represented in the rooms where those systems are built.
The path forward is not simply to hire more diverse researchers into existing institutions and existing problem framings. It is to resource and centre organisations led by and accountable to the communities they serve. It is to fund women and people of colour not as diversity hires but as the most qualified people to lead work that requires their expertise.