Empirical Research Assistance (ERA): An AI system that creates scientific software to optimize quality metrics, combining Large Language Models with Tree Search.
Key finding: The system discovered 40 novel bioinformatics methods that outperformed top human-developed approaches on a public leaderboard.
AI System Outperforms Human Experts in Scientific Software Development
Researchers have developed a new AI system called Empirical Research Assistance (ERA) that autonomously creates scientific software to optimize quality metrics. ERA combines a Large Language Model with Tree Search to systematically improve solutions, representing a potential leap forward in automated scientific discovery.
Breakthrough Results Across Multiple FieldsIn bioinformatics, ERA discovered 40 novel methods for single-cell data analysis that outperformed top human-developed methods on a public leaderboard. This demonstrates the system's ability to not just match, but exceed expert human performance in a complex domain.
For epidemiology, ERA generated 14 models that surpassed the CDC ensemble and all other individual models in forecasting COVID-19 hospitalizations.
The system also produced expert-level software in several other domains:
- Geospatial analysis
- Neural activity prediction in zebrafish
- Numerical solution of integrals
- A novel rule-based construction for time series forecasting
"The system represents a step towards accelerating scientific progress by automating the creation of computational experiments."
This development suggests that AI systems like ERA could dramatically speed up the pace of scientific discovery by handling the time-consuming process of developing and optimizing computational methods, freeing human researchers to focus on higher-level questions and interpretation.