The Future of AI Scientists: Emerging Directions and Fundamental Challenges

Abstract

Modern AI systems are rapidly evolving from passive tools into agentic systems that can reason, learn, collaborate, and interact with human scientists in the research process. This talk discusses emerging directions in AI scientist systems, drawing inspiration from human-centered principles such as lifelong learning, teaching, collaboration, objective alignment, and auditing. The goal is to identify fundamental research challenges and stimulate new perspectives on building steerable and productive AI systems in complex real-world AI-for-Science settings.

Biography

丁杰，明尼苏达大学统计学院副教授，电子工程系、计算机系客座教授，AI中心核心成员，明尼苏达大学AI健康医疗工作小组联合组长，本科毕业于清华大学，博士毕业于哈佛大学。其研究位于人工智能、统计学、信息科学的交叉领域，当前重点关注可扩展与可信赖的人工智能、AI for Science，以及人机协作系统。曾任 Amazon Scholar，与 Amazon Alexa 及 AGI 团队合作开展联邦学习与多模态大模型相关研究。其工作获得美国国家科学基金会 CAREER Award、ARO Young Investigator Award、Cisco Research Award、Meta Research Award 及多项最佳论文奖。他还发起了明尼苏达大学首个面向全校的人工智能开源课程。近年来，他进一步关注AI智能体系统的工程化与落地实践，探索如何让科研人员或各领域专家在短时间内构建可控的AI智能体，并围绕其联合创立的AI-for-Science平台MorphMind及相关开源项目WorldSeed开展系统性探索。