Responsibility Across the AI Lifecycle

Responsible AI in global health cannot end at design, validation, or ethical intent. The real test begins after deployment.

In practice, AI systems drift. Data quality changes. Workflows evolve. Infrastructure fails. Users adapt tools in unintended ways. Harms may also emerge unevenly across gender, language, geography, disability, migration status, or socioeconomic position. Without post-market surveillance, these risks remain invisible until they become embedded in care delivery or public health practice.

Accountability is equally underdeveloped. If an AI-supported decision delays care, misclassifies risk, excludes someone from services, or produces unsafe guidance, who is answerable? More importantly, what pathway exists for affected individuals to report harm, seek review, receive correction, or obtain remedy?

Responsible AI must be judged not only by how it is built, but by how it is governed after it is used.