Researchers built a multi-agent AI framework for public health data collection. It handles source discovery, assessment, evidence structuring, event linking, and validation. The system recovered all 13 cases in a hantavirus outbreak. It linked 85 publisher domains versus 28 in the reference benchmark. It reconstructed Canadian measles counts 0.33% below the national reference. Higher cumulative counts appeared on 79.8% of comparable days. It raised the share of days with recent Sierra Leone mpox reports from 28.2% to 65.6%. The workflow uses attributed media to improve coverage. Limitations in record completeness and accuracy remain. Prospective timeliness gains are untested. The authors propose agentic AI as a complementary layer for public health data infrastructure.
What it unlocks
Automating discovery, assessment, and linking of heterogeneous public health data sources.
- 13 individual hantavirus cases recovered
- 85 vs 28 publisher domains linked
- 0.33% below national measles reference
- 79.8% of days with higher cumulative counts
Sources