제4회 세미나
발표자 Fengli Xu (Tshinghua University)
발표 주제 Agent for Innovation: from Specialized Tools to Agentic Research Partners
개요 The increasing scale and complexity of scientific research have motivated systematic inquiries into the mechanisms underlying research productivity, collaboration, and innovation. Over the past decade, AI for Science has emerged as a major force reshaping how scientific knowledge is produced. This talk presents our large-scale empirical study of its impact on the scientific ecosystem. We find that AI for Science substantially accelerates individual researchers’ progress and amplifies their scientific impact, yet paradoxically narrows the collective scope of scientific exploration. This tension highlights a structural limitation of current AI for Science tools, which are predominantly optimized for isolated tasks rather than the broader dynamics of scientific discovery.
Building on these findings, we design OmniScientist and AutoSOTA, two AI scientist systems designed to support the full research lifecycle. By incorporating insights from the science of science into the architecture of AI agents, these systems provide integrated support for literature review, idea generation, experimental execution, scientific writing, and automated peer review. Together, they explore a path for transforming AI from a task-specific tool into a collaborative partner in scientific discovery.
장소 온라인 (https://us06web.zoom.us/j/
일시 2026년 8월 3일(월) 17:00-18:30