主题：Research Paradigms in the AI Era: From Model Capabilities to Workflow Design
AI时代的研究范式：从模型能力到工作流设计
时间：2026年6月4日上午10:00-11:30
地点：：管理科研楼EMBA第一教室
主讲人：陶学臻
主持人：朱睿智
Bio:陶学臻，上海财经大学商学院产业经济系副教授。他的研究领域涵盖产品创新的结构实证估计以及不完全信息设定下的反垄断分析，研究议题包括医疗器械领域中的集中采购政策评估、产品创新传递效率评估以及跨区域就医的产品创新影响，不完全信息设定下企业的定价竞争和古诺竞争的信号均衡机制研究，以及AI市场竞争对创新和劳动力市场的影响。他的学术成果发表于American Economic Journal: Microeconomics, Journal of Industrial Economics, China Economic Review等国际期刊，并主持完成国家自然科学基金青年基金。
Tao Xuezhen, Associate Professor in the Department of Industrial Economics at the College of Business, Shanghai University of Finance and Economics. His research areas cover structural empirical estimation of product innovation and antitrust analysis under incomplete information settings. His research topics include evaluation of centralized procurement policies in the medical device sector, assessment of the transmission efficiency of product innovation, the impact of cross-regional medical treatment on product innovation, signaling equilibrium mechanisms in firms' pricing competition and Cournot competition under incomplete information, as well as the effects of AI market competition on innovation and the labor market. His academic work has been published in international journals such as the American Economic Journal: Microeconomics, the Journal of Industrial Economics, and China Economic Review, and he has led to completion a Young Scientists Fund project of the National Natural Science Foundation of China.
Abstract:
大语言模型的快速演进正在改变经管实证研究的工作方式，但研究产出的质量差异往往并非源于模型本身，而是取决于工作流的设计。本讲座围绕这一判断，结合APE全自动论文生产、NBER文献知识库、Auto_Research迭代实验等前沿案例，介绍智能体工作流的核心设计框架——项目宪法、任务拆解、质量门控与人类检查点，并深入讨论全自动化研究面临的风险结构，以及因果推断、制度判断、反事实评估等人类认知优势在AI时代的新定位。讲座面向有一定编程工具使用经验的经管学者与研究生，旨在提供一套可落地的工作流设计思路。
The rapid evolution of large language models is reshaping how empirical research in economics and management is conducted, yet differences in research output quality often stem not from the models themselves but from workflow design. Building on this premise, and drawing on frontier cases such as APE's fully automated paper production, the NBER literature knowledge base, and Auto_Research's iterative experimentation, this lecture introduces the core design framework for agentic workflows — project constitutions, task decomposition, quality gates, and human checkpoints. It then takes a closer look at the risk structure of fully automated research, and the new positioning of distinctively human cognitive strengths — causal inference, institutional judgment, and counterfactual evaluation — in the AI era. Intended for economics and management scholars and graduate students with some experience using programming tools, the lecture aims to offer a practical, implementable approach to workflow design.