LLM Scientific Reasoning - How to Make AI Capable of Nobel Prize Discoveries

Explore how large language models can be developed to achieve breakthrough scientific discoveries comparable to Nobel Prize-winning research in this conference talk. Examine the complex reasoning processes behind major scientific breakthroughs, using examples like RNA interference that required connecting distant areas of biology with different vocabularies and methodologies. Learn about the classification of problems tackled by Retrieval-Augmented Generation (RAG) systems and discover a new family of NLP problems specifically related to scientific discovery. Understand the proposed benchmarking template for evaluating AI systems' ability to identify novel relationships between phenomena from disparate scientific fields. Investigate whether RAG systems, Agentic RAG, or fully-fledged AI agents can effectively tackle high-level scientific reasoning challenges. Follow the systematic approach covering reasoning processes before retrieval, finding optimal components for discovery, understanding the nature of scientific reasoning, problem-solving methodologies, design choices for AI systems, and testing frameworks for evaluating scientific reasoning capabilities in artificial intelligence.