人工智慧學系

【AI Seminar】2026.09.22 The Human Side of AI Training – WashU Department of Computer Science and Engineering Prof. Chien-Ju Ho

【文號/Document number：CGUD00000907171】

1. 講者: 美國聖路易斯華盛頓大學電腦科學與工程學系 何建儒副教授
2. 題目: The Human Side of AI Training 人工智慧訓練中的人類行為與相互影響
3. 參加對象: 全校大三以上學生、教職員與長庚醫師
4. 實體地點: 管理大樓11樓AI講堂
5. 演講時間: 2026年9月22日(二) 13:00-15:00
6. 線上連結：https://reurl.cc/5XD8Kn或 掃描海報 QR code
演講無需報名， 請於演講時間進入Teams線上會議室

何建儒副教授為現任美國聖路易斯華盛頓大學 (WashU)電腦科學與工程學系副教授，曾於康乃爾大學擔任博士後研究員，目前擔任長庚大學人工智慧學系客座副教授。他的研究主要探討機器學習、人機互動與協作、人類運算，以及演算法與行為經濟學之間的交會，特別關注人類行為如何影響 AI 系統的學習與決策，以及 AI 系統又如何反過來影響人類。他的研究成果發表於 PNAS、ICML、NeurIPS、WWW、AAAI、IJCAI、ACL、NAACL、AAMAS 等國際期刊與會議。他亦擔任 ACM Conference on Human-AI Complementarity and Alignment（HCOMP）2026 議程共同主席（Program Co-Chair）。

演講摘要：
近年來，人工智慧快速發展，而這些進展在很大程度上仍仰賴人類所提供的資料、回饋、判斷與示範。然而，人類並非只是被動的資料來源：人類的行為會受到 AI 系統與訓練機制的影響，而人類行為本身的特性，也可能進一步影響 AI 學習的結果。本演講將以「人工智慧訓練中的人類行為與相互影響」為主題，探討在人工智慧訓練過程中，人類行為及人與人工智慧之間的相互影響，如何改變我們蒐集與使用訓練資料的方式。我將介紹一系列相關研究，包括人們在知道自己的行為將被用來訓練人工智慧時，如何改變自身的決策行為；在從人類示範中學習時，如何考量人類行為的可學習性與脆弱性，以改善人工智慧的學習成果；以及如何結合人類與大型語言模型，更有效率地取得與估計人類的判斷。這些研究共同指出，在訓練人工智慧時，我們不僅需要關注所使用的資料，也需要理解這些資料背後的人類行為，以及人與人工智慧互動所產生的相互影響。

公告單位：智慧運算學院
承辦人：洪語彤
聯絡分機：409-2501

Speaker: Prof. Chien-Ju Ho, WashU Department of Computer Science and Engineering
Date and Time: Tuesday, Septermber 22 , 2026, 13:00–15:00
Venue: 11th Floor, Management Building AI Lecture Hall,
Meeting Link: https://reurl.cc/5XD8Kn or scan the QR code on the poster

About the Speaker:
Chien-Ju Ho is an Associate Professor of Computer Science and Engineering at Washington University in St. Louis (WashU). He is currently on sabbatical and serving as a Visiting Associate Professor at Chang Gung University. His research lies at the intersection of machine learning, human-AI interaction and collaboration, human computation, and algorithmic and behavioral economics. In particular, his work examines how human behavior shapes the learning and decision making of AI systems, and how AI systems in turn influence people. His research has appeared in venues including PNAS, ICML, NeurIPS, ACL, NAACL, WWW, AAAI, and AAMAS. He also serves as Program Co-Chair of ACM Conference on Human-AI Complementarity and Alignment (HCOMP) 2026.

Lecture Abstract：
Recent advances in artificial intelligence depend heavily on data, feedback, judgments, and demonstrations provided by humans. Yet humans are not simply passive sources of training data: their behavior can be shaped by AI systems and by the process of training them, while the characteristics of human behavior can in turn affect what AI systems learn. In this talk, I will examine human behavior and mutual influence in AI training, and discuss how taking these factors into account can change the way we collect and use training data. I will present a series of studies examining how people change their decisions when they know their behavior will be used to train AI; how accounting for the learnability and fragility of human behavior can improve learning from human demonstrations; and how humans and large language models can be combined to more efficiently elicit and estimate human judgments. These studies highlight the importance of understanding not only the data used to train AI, but also the human behaviors and interactions through which those data are generated.


Contact person：Tiffany Hung
Contact number：409-2501

---