← back
LLM Scientific Reasoning: How to Make AI Capable of Nobel Prize Discoveries: Hubert Misztela
Takeaway
Scientific discovery RAG needs reasoning before retrieval — decompose the question, restructure the data into graphs, and pick a reasoning type that matches question complexity.
Summary
- Frames challenge with the 1990s petunia color-flip paradox that took 8 years to resolve into the RNAi Nobel Prize — could RAG over literature accelerate this?
- Categorizes question complexity: 1-to-1 (single needle), 1-to-n (multi-needle), n-to-n (multiple concepts hidden across chunks) — the messier the question, the more reasoning before retrieval is needed.
- Pre-retrieval reasoning options include routing, HYDE, graph readers/graph RAG to restructure data, and decomposition; post-retrieval reasoning is the standard LLM step.
- Compares RAG-with-reasoning pipeline to chemistry's encoder/decoder + latent-space navigation used for molecule generation at Pharma Novartis.
- Talk pitches reasoning types — aggregation, logical, causal, probabilistic, structured, algorithmic — as the toolkit for scientific RAG.
ragreasoningscience
Original description
Do you remember that feeling when you realized who was Jon Snow's mother? Or who was the Batman really? Those 'aha' moments define scientific reasoning: of many steps and non-obvious. Even though the scientific discovery using LLMs is becoming more popular recently, there is little if any discussion about high level reasoning process behind breakthroughs in science. Some of the discoveries like RNA interference required connecting of distanced areas of biology, described in different journals and with distinct vocabulary. In this talk we would like to discuss if LLMs would be able to spot those novel relationships behind phenomenas from distanced areas? What would it take: RAG, Agentic RAG or fully fledged AI agent? We will first try to classify problems tackled by RAGs. Then we would define a new family of NLP problems related to the scientific discovery and propose a template for benchmarking of thereof. We would discuss them on the examples of Nobel Prize discoveries. Then the question would be if and how RAGs or more general AI Agents might help us in tackling those problems, so we will try to entertain you with some attempts to solve it. Recorded live in San Francisco at the AI Engineer World's Fair. See the full schedule of talks at https://www.ai.engineer/worldsfair/2024/schedule & join us at the AI Engineer World's Fair in 2025! Get your tickets today at https://ai.engineer/2025 About Hubert Cryptographer who became Apps Developer, Apps Developer who became a Data Scientist, Data Scientist who became a Consultant, Consultant who became AI Researcher for drug design. AI Researcher who is flirting with AI Engineering. I have over a decade of experience spanning from programming and research to business stakeholder management. Delivering solutions in AI, software development, mathematics and cryptography across industries: pharma, high-tech, government and education. For the last six years I have been at Novartis: first two working on AI applications to commercial activities optimisation in close alignment with region Europe leadership team and the last 4 building AI solutions for small molecule design in collaboration with Microsoft Research. Lecturer, conference speaker, internal trainings organizer. Recently I have organized an internal company LLM workshop and hackathon with 140+ participants.