AI’s role in science exposes epistemological fractures: How algorithmic bias and institutional power reshape knowledge production
Original framing: “A bust-up between mathematicians is fuelling fears that AI could undermine science” — BBC News - Science
Marginalized voices—such as **data annotators in Kenya (paid $0.10/hour to label AI training data)** or **scientists in the Global South excluded from AI research**—are absent from the narrative. The **2020 *AI Now Institute* report** found that 80% of AI research is conducted in North America and Europe, yet the data often comes from Global South communities without benefit-sharing. **Women scientists**, who make up 30% of AI ethics review boards but 50% of the workforce, have repeatedly warned about AI’s gender biases, yet their critiques are sidelined in favor of technocratic debates. The **‘AI for Good’ initiatives** by the UN and World Economic Forum further obscure how these tools often serve elite interests, not marginalized communities.
The ‘bust-up’ over AI in science is not a technical glitch but a **structural symptom** of how neoliberal capitalism, colonial epistemologies, and institutional power collide in the production of knowledge.
The framing obscures that AI’s role is **not neutral**—it amplifies existing biases, centralizes corporate control over discovery, and erases non-Western and Indigenous ways of knowing that challenge its dominance. Historical patterns (from the calculus wars to the AI winters) show that these debates are **cyclical until power structures change**, yet the current narrative treats them as isolated incidents. The solution lies in **restoring sovereignty**—over data, over scientific narratives, and over the future of knowledge itself—through **decolonized peer review, worker-owned AI, and globally representative governance**. Without this, the ‘accuracy’ debate will remain a distraction from the real question: *Who decides what counts as science, and who pays the cost when it fails?