SLAI Seminar 29th Session will be discussing the topic on " AI for Social Science and Economic World Models ",  from 10am to 11:30am, March 17th (Tuesday) at Room B401, online participation is welcome.

 (Tencent Meeting ID: 975-860-042)

报告主题：赋能社科的人工智能与经济世界模型

时间：2026年3月17日（周二）上午10:00 - 11:30

地点: 深圳河套学院B401教室

线上参与：腾讯会议号975-860-042

主讲嘉宾：Lin William Cong 丛林教授

主持人：黄建伟教授

讲者简介About the Speaker:

丛林教授现任深圳河套学院社会科学智能中心主任、新加坡南洋理工大学校长讲席教授，同时担任南洋商学院副院长、金融学讲席教授，以及计算与数据科学学院数据科学与人工智能教授，并任职于全球金融、技术与社会研究院。丛教授曾任康奈尔大学Rudd Family管理学和金融学讲席教授、芝加哥大学金融学教授。他在顶级管理科学与金融学期刊担任主编、副主编，是加密合约倡议计划的科研专家，同时身兼美国全国经济研究所、欧洲经济政策研究中心、亚洲金融与经济研究局的研究员、资深学者。

丛教授研究领域涵盖金融学、数字经济学、信息经济学、人工智能、金融科技、数字经济及创业创新，学术成果多次被彭博新闻社、美国有线电视新闻网、《经济学人》、《华盛顿邮报》等国际主流媒体报道。他所带领的团队开创了人工智能在经济学中的应用、金融信息设计研究，奠定了通证经济、经济世界模型、区块链经济学、预言机经济学、人工智能行为经济学等领域的理论基础，并开发出用于市场操纵监测和金融科技监管优化的数据分析工具。

作为金融科技领域发文量最多、引用率最高的学者之一，丛教授被IDEAS/RePEc评为全球五大年轻经济学家之一，其金融理论研究成果引用率在21世纪毕业学者中位列第六。他累计获得百余项国际会议最佳论文奖及竞争性科研基金，多次受邀在国际学术会议发表主旨演讲，并为初创企业、投资机构及非营利组织提供战略咨询，曾任Chainlink首任首席经济学家、HoloBit首席科学家。他还为中国证券投资基金业协会、美国证券交易委员会、美国司法部与财政部、纽约州金融服务局、加拿大央行等政府部门提供顾问咨询或培训支持。

丛教授拥有斯坦福大学金融学博士和统计学硕士学位，并以物理学系第一名、全优成绩毕业于哈佛大学，获物理学硕士学位、数学与物理学学士学位（主修）、经济学辅修学位及法语语言证书。

Lin William Cong is the President’s Chair Professor at Nanyang Technological University, with joint appointments at Nanyang Business School (Associate Dean and Professor in Finance), the College of Computing and Data Science (Professor in Data Science and AI), and the Global Institute of Finance, Technology, and Society. Previously, he served as the Rudd Family Endowed Chair Professor of Management and Finance at Cornell University, and as a Finance professor at the University of Chicago. He is an editor or associate editor at leading Management Science and Finance journals, a faculty scientist at the Initiative for Cryptocurrencies & Contracts (IC3), a research associate or senior fellow at the National Bureau of Economic Research (NBER), European Centre of Economic Policy Research (CEPR), and the Asian Bureau of Finance and Economic Research (ABFER).

Cong’s research spans finance, digital economics, information economics, AI, FinTech, digital economy, and entrepreneurship, and has been featured in Bloomberg, CNN, the Economist, Washington Post, etc. His team has pioneered AI for economics, information design in finance, and has laid the foundations of tokenomics, economic world models, blockchain economics, and oracle economics, behavioral economics of AI, and has developed data analytics for detecting market manipulation and better FinTech regulation, among others.

Cong is one of the most published and cited FinTech researchers, a top 5 young economist ranked by IDEA, and the 6th highest cited financial theorist among all those graduated in the 21st century. His research has been recognized with over 100 conference best paper prizes and competitive grants. He has also been invited to deliver keynote speeches at numerous international conferences and to advise various startups, investment firms, and non-profit organizations, including serving as the inaugural Senior Economist at Chainlink and the Chief Scientist at HoloBit. He has also been an advisor or trainer for government agencies such as Asset Management Association of China, the SEC, U.S. Department of Justice and Department of Treasury, New York State Superintendent’s Office and Department of Financial Services, and the Bank of Canada.

Cong earned a Ph.D. in Finance and a MS in Statistics from Stanford University. He also holds dual degrees from Harvard University where he graduated summa cum laude and top in the Physics department (perfect GPA), with an A.M. in Physics, an A.B. in Math & Physics, a minor in Economics, and a language citation in French.

报告摘要：

本讲座提出现代人工智能如何用于推动经济学研究以及更广泛社会科学研究的三个主要方向（或阶段）。整体而言，这些方向不仅构成了经济或金融学中首批超越文本生成式人工智能模型，也共同形成了一种区别于既有模型框架的全新研究范式——经济世界模型。

具体而言，

首先，我介绍在大规模建模空间中进行目标导向优化的算法方法，包括基于 Transformer 的强化学习框架，以及规模相对较小但具有良好可解释性的面板树模型。这类方法特别适用于最优决策问题（例如灵活投资管理与大规模个性化智能投顾），以及在面板数据中刻画分组异质性（例如构建测试组合或潜在因子以改进资产定价、评估既有模型，并分析不同资产类别或宏观经济阶段下的差异性因子结构及收益可预测性的异质性）。

（二）其次，我讨论如何将深度学习与鲁棒控制相结合，构建经济世界模型，以研究企业在复杂、高维、非线性随机控制环境下的决策问题。我们提出的 AlphaManager 不仅在样本内与样本外更好地解释和预测企业绩效结果，而且能够提出显著优于现有实践的关键管理决策建议。我们在一个经典的道德风险情境中验证了该框架的有效性，并在美国上市公司横截面与时间维度上，系统性地揭示了模型模糊性、预测性能以及政策有效性的丰富异质性。此外，我们通过逆向强化学习刻画企业管理者的目标函数，并利用模型模糊性度量来理解数据驱动方法的适用边界与内在限制。

（三）最后，我探讨如何进一步构建更丰富的经济世界模型，特别是结合智能体的建模与仿真方法：其一，通过在线借贷市场的案例，定义并刻画“数据驱动生成式均衡”，用于支持反事实分析；其二，借助心理学与行为经济学实验范式，研究人工智能智能体的行为特征——无论这些智能体是对人类行为的模拟，还是作为经济体系中新的主体群体存在；其三，通过提出“结构化知识增强神经网络”（SKINNs），在理论建模与数据驱动方法之间构建桥梁，实现两种研究范式的有机融合。

Abstract: I overview three major directions/stages of how modern AI can be used for advancing research in economics and, more broadly, in the social sciences. Collectively, they not only constitute the first non-text-based GenAI models in economics or finance, but also form a new research paradigm involving Economic World Models that differ from existing models.

Specifically,

(i) I first introduce goal-oriented algorithms in large modeling spaces involving transformer-based reinforcement learning or not-so-large but interpretable panel trees, which are particularly suited for optimal decision-making (e.g., flexible investment management and mass-customized robo-advising) and for characterizing grouped heterogeneity in panel data (e.g., generating test portfolios or latent factors for better asset pricing or evaluating extant models, and understanding differential factors across asset groups or macroeconomic regimes, and differential return predictability).

(ii) I then discuss combining deep learning with robust control to build Economic World Models to study corporate decision-making that entails complex, high-dimensional, and non-linear stochastic control. Our AlphaManager not only better explains and predicts corporate outcomes in- and out-of-sample but also prescribes key managerial actions that significantly outperform. We validate our framework in a well-known moral hazard setting, and empirically document rich heterogeneity in model ambiguity, prediction performance, and policy efficacy in the cross section of U.S. public firms and over time. Furthermore, we use inverse reinforcement learning to learn managerial objectives and use model ambiguity measures to understand the limit to data-driven approach in corporate finance research.

(iii) Finally, I discuss how we can move towards richer Economic World Models involving AI-agent-based modeling/simulation, by (a) introducing and characterizing data-driven generative equilibrium for counterfactual analyses using the example of online lending, (b) understanding the behavior of AI agents using experiments from the psychology and behavioral economics literature, regardless of whether they represent humans or constitute a new population in the economy, and (c) bridging theory and data-driven methodologies by inventing Structured-Knowedge-Informed Neural Networks.