Building AI Co-Scientist System for Accelerating Scientific Discoveries

Abstract

Given the fast development of technology and automation systems, researchers are generating tons of data from scientific experiments to facilitate discoveries, especially in the areas of biology and medicine. However, large-scale scientific datasets make it challenging for scientists to more effectively utilize these data with robust and powerful methods for novel and reproducible discoveries, and thus, we build AI scientists to make our lives easier. In this talk, I will share two recent progresses from our team to pioneer the study of AI scientists and related benchmarks. I will discuss our progress in building an AI Bioengineer (SAGA) for accelerating scientific discoveries, which facilitates generation tasks in multiple areas to enclose this robust and extensive system, followed by two benchmark studies (SciAgentArena and OSWorld-Science) we extend with a multi-step and complicated reasoning process to evaluate the capacities of AI scientists in solving complex and computer-using tasks.

線上視訊請點選連結