AI2S (AI-to-Science) is a research initiative that advances a new paradigm for scientific discovery and engineering design.
AI2S is grounded in the observation that accurate AI models implicitly encode the governing principles of the scientific systems they represent. By analyzing internal representations and learned computational structures, AI2S extracts the mechanisms and rules captured by these models and expresses them as interpretable scientific principles suitable for validation, theory building, and engineering use.
This GitHub organization hosts:
- Research code for AI2S methods and case studies
- Tools for probing and interpreting AI model internals
- Domain-specific applications across science and engineering
- Experimental pipelines for validating extracted principles
AI2S is interdisciplinary, open, and evolving. Contributions and collaborations are welcome.