By Louis DiPietro, Cornell Ann S. Bowers College of Computing and Information Science
On


What will scientific discovery look like in the age of artificial intelligence (AI)?

Carla Gomes, the Ronald C. and Antonia V. Nielsen Professor of Computing and Information Science in the Cornell Ann S. Bowers College of Computing and Information Science, takes aim at this critical question in an essay published in the winter/spring 2026 edition of Daedalus.

In “Knowledge-Centric AI for Scientific Discovery,” Gomes – a pioneer in computational sustainability – posits that today’s data-driven AI approaches, such as large language models (LLMs) fail to reason like actual scientists. LLMs may know Newton’s Laws but may fail to apply them consistently. 

“Despite their remarkable success, relying solely on purely data-driven methods has intrinsic limitations for scientific discovery,” she said. 

Can Data Alone Drive Discovery?

Carla Gomes named Schmidt AI2050 Senior Fellow

Despite their remarkable success, relying solely on purely data-driven methods has intrinsic limitations for scientific discovery.

Carla Gomes
The Ronald C. and Antonia V. Nielsen Professor of Computing and Information Science

In the essay, Gomes presents a new framework called knowledge-centric AI, a kind of supercharged LLM that can find patterns in mountains of data like modern-day models and also follow and apply scientific rules like real scientists. This method, she argues, could fundamentally accelerate discoveries across science, in areas ranging from climate change and sustainability to medicine and conservation.

To create AI that thinks more like scientists, Gomes turned to deep reasoning networks, or DRNets, a deep learning architecture that integrates domain knowledge and breaks down reasoning processes into mathematical steps computer systems can follow rigorously.

Gomes and her team have shown this model can work. They developed and applied knowledge-centric AI to discover new solar fuel materials, created continental-scale bird distribution maps to aid in bird conservation decision-making, and predict soil carbon cycles critical for climate modeling. 

“These results illustrate how combining deep learning with first-principles, knowledge-­centric reasoning can automate data interpretation and significantly accelerate scientific discovery,” she wrote in the essay. "Looking ahead, scientific progress is poised to be shaped by communities of knowledge-centric scientific AI agents that learn, reason, and collaborate across domains, close the loop between inference and action in (semi-)autonomous laboratories, and navigate trade-offs transparently. In this vision, artificial intelligence becomes a valuable scientific collaborator.”

Louis DiPietro is a writer for the Cornell Ann S. Bowers College of Computing and Information Science.