Foundation models learn to discover new scientific problems and knowledge
Discovery Foundation Models: Toward Open-Ended Discovery Intelligence
Computation and Language
Summary
Usually, AI systems work by solving problems humans give them or by learning from existing facts. This paper talks about an advanced kind of AI, called Discovery Foundation Models, that can help find new problems, come up with ideas, and test them to make new knowledge. The authors built systems that mix computer reasoning and real lab experiments to do this kind of discovery. This means AI can not just answer questions but take part in inventing new questions and solutions all on its own.
What this means in practice
- •For biotech research teams: Coordinate AI and robotic systems to accelerate new drug discovery through iterative hypothesis testing and lab experiments.$Commercial implications: Enables creation of AI-driven drug discovery platforms combining computational reasoning with wet-lab automation for pharma companies.
- •For software developers: Develop AI tools that automatically generate and refine research plans and ideas beyond static problem solving.
Authors
Ling Yang, Zhenfei Yin, Yingcheng Wu
Abstract
Foundation models have progressed from learning and reasoning over existing knowledge, to increasingly learning through action, tool use, and outcome feedback. We argue that the next frontier is a further transition: from solving and acting within problems specified by humans to participating in the process by which new problems, representations, explanations, and knowledge are created. We refer to this capability as Discovery Intelligence. We formulate Discovery Foundation Models (DFMs) as general-purpose model systems for open-ended discovery. A DFM operates over a revisable research state and supports seven coupled capabilities spanning problem discovery, formulation, representation construction, hypothesis formation, intervention, evidence-grounded revision, and continual discovery improvement. We instantiate this framework with Zetema, which couples explicit research-state dynamics, verification and experimental gating, external grounding, and cross-task Discovery Skill evolution. We further ground the framework with GALILEO, a real therapeutic-discovery system in which Dry-Lab reasoning, robotic and hands-on Wet-Lab experimentation, external biological evidence, and iterative hypothesis and design revision form a closed physical discovery loop. We then formulate a unified approach to capability formation and process-centered evaluation, enabling discovery behavior to be trained, improved, and measured beyond final-answer performance. Together, these components establish discovery as a learnable, executable, and evaluable capability of foundation-model systems. We view this shift as a broader progression in intelligence scaling: from learning over existing knowledge, to learning from action outcomes, and ultimately to participating in the construction, testing, and revision of the structures through which new knowledge is discovered. Code: https://github.com/Gen-Verse/DFM-Plans