Data Cascade Prevalence

92% of AI practitioners experienced at least one data cascade. 45.3% experienced two or more cascades in a given project.

This finding comes from a qualitative study of 53 AI practitioners across India, East and West African countries, and the US, all working in high-stakes domains including healthcare, conservation, credit assessment, and public safety.

The high prevalence, even among practitioners who understood data quality’s importance and had deep moral commitment to the communities they served, points to systemic issues rather than individual negligence. The problem isn’t awareness; it’s that conventional AI practices and incentive structures make cascades nearly inevitable.

Related: 04-atom—data-cascades-definition, 05-atom—model-valorization