Google AI Predicts Flu and COVID Better Than Top Models

Scientists at Google have created an AI assistant that's changing how we predict disease outbreaks, and early results show it's working better than many existing tools.

The system, called Empirical Research Assistance (ERA), started making weekly predictions for flu hospitalizations across every U.S. state in November. By December, it expanded to forecast COVID-19 and RSV cases too, submitting real-time forecasts to CDC tracking hubs every single week.

Public leaderboards maintained by the University of Massachusetts Amherst show Google's AI performing at or near the top for both flu and COVID-19 predictions since it started competing. The tool accurately predicts hospital admissions up to four weeks in advance, giving health systems crucial time to prepare.

But ERA isn't just about tracking respiratory illnesses. Google researchers have been testing it across wildly different scientific fields, from studying cosmic strings in space to monitoring carbon dioxide from weather satellites.

In one breakthrough, the team combined ERA with another AI system to solve a longstanding cosmology problem about gravitational energy from cosmic strings. Previous attempts using other AI tools had only cracked the simplest version, but ERA found six general solutions and a complete formula.

The Ripple Effect

What makes this especially exciting is how it could democratize science itself. Complex computational modeling has traditionally required specialized expertise and expensive resources, limiting who can do cutting-edge research.

ERA changes that equation. The tool can generate expert-level scientific software for researchers who might not have access to large teams or advanced computing power.

For public health, this means smaller health departments in rural areas or developing countries could potentially access forecasting tools as powerful as those used by major research institutions. A visiting faculty researcher described the experience of using ERA as inspiring, watching it tackle problems that would normally require entire research teams.

The system works by helping scientists create and test models faster, turning weeks of programming into hours of guided problem-solving. It doesn't replace human expertise but amplifies it, letting researchers focus on asking the right questions instead of wrestling with technical barriers.

Google plans to make ERA more widely available to researchers globally, potentially accelerating discoveries in fields from neuroscience to climate science. The weekly forecasts for flu, COVID-19, and RSV continue, with each prediction helping refine the system's accuracy.

For now, health officials are watching closely as the AI proves itself week after week, turning raw data into potentially life-saving predictions.