Part of the TPC Seminar Series

---

Abstract:

Advances in generative artificial intelligence (genAI) systems are enabling entirely novel ways to abstract and reason on scientific data, including literature, simulations, (prior) experiments and structured databases. We posit that genAI models can probe, understand, and reason about complex relationships within scientific data that may not be directly evident from traditional approaches. In this talk, we will focus on preliminary analyses in building an agentic framework, namely, Scientia, that builds on Google’s co-scientist approach by enabling it to build novel hypotheses based on “reading” scientific literature, “infering” simulation and experimental evidence in support or against the hypotheses, while simultaneously tracking how “ideas” may be exchanged and developed further, emulating scientific discourse. We discuss some emerging issues in the context of scaling these workflows on supercomputers and discuss applications of the framework on real-world applications in protein engineering and design workflows.

Biography:

Arvind Ramanathan is a computational science lead at Argonne National Laboratory, focusing on the integration of generative AI, multiscale modeling and simulations, robotics and automation to understand how complex biological systems function.