2025年第104期(总第1145期)

演讲主题：Augmenting the Operations Manager with a Prediction Machine

主讲人：芦涛 康涅狄格大学副教授

主持人：关旭 供应链管理与系统工程系主任、教授

活动时间：2026年01月07日（周三）09:00-10:30

活动地址: 管院大楼205教室

主讲人简介：

芦涛是University of Connecticut商学院运营与信息管理系副教授。他在香港科技大学工业工程和物流管理系取得博士学位，曾任教于荷兰Erasmus University鹿特丹管理学院。他的研究兴趣涵盖供应链管理、社会责任和可持续运营、平台经济、运输物流等领域。他的研究成果发表在Management Science, Manufacturing & Service Operations Management, Operations Research, Information Systems Research, Production and Operations Management等顶级期刊。他目前担任Management Science和Service Science等杂志的副主编。

活动简介：

Firms increasingly use Artificial Intelligence (AI) enabled forecasting engines (＂prediction machines＂) to augment their managers' own forecasting capabilities and thus improve sales-and-operations planning outcomes. Deployment of a prediction machine may cause an unintended reduction in a manager's own forecasting effort which in turn diminishes the value of machine adoption. We model a firm facing uncertain demand that delegates a procurement quantity decision to a human manager who can exert effort to generate a demand prediction. The firm deploys a machine that provides the manager with a demand-prediction signal. We establish the conditions under which managerial effort reduction occurs and thus reduces the machine's potential value. Adopting a Bayesian persuasion approach, we show that partially disclosing the machine's prediction, either downplaying high predictions or exaggerating low predictions, can be optimal, depending on the product's cost-to-revenue ratio. A strategy of minimal obfuscation (to achieve effort) is optimal if the machine is more accurate than the human; however, maximal obfuscation (while maintaining effort) can be optimal if the human is more accurate. Our results imply that the firm may be better off tuning a machine to be less informative than its maximum capability.