报告题目：人工智能对齐 (AI Alignment )

报告人：陆海兵

时间：2024年6月16日下午14:30-16:30

地点：西安交通大学雁塔校区财经主楼806会议室

报告人简介：

陆海兵教授现就职于美国克拉拉大学。于2002年和2005年在西安交通大学获得学士和硕士学位，并于2011年在Rutgers大学管理专业（信息科学方向）获得博士学位，曾在新加坡管理大学任研究员（2005-2006）。陆海兵教授的研究兴趣主要集中在数据挖掘、隐私和安全以及优化等方向的交叉领域，已经在Manufacturing & Service Operations Management、IEEE Transactions on Big Data、IEEE International Conference on Data Engineering、IEEE Transactions on Dependable and Secure Computing等期刊及会议上发表多篇论文。

摘要：

As artificial intelligence (AI) systems become increasingly integrated into various aspects of society, ensuring that these systems align with human values, goals, and ethical standards is of paramount importance. This presentation explores the key concepts and principles in AI safety and alignment, a critical field dedicated to developing AI technologies that act in ways consistent with intended ethical guidelines and societal expectations. The presentation will address the technical and ethical challenges associated with aligning AI systems, including value alignment across diverse contexts, mitigating unintended consequences, and addressing conflicts between AI goals and human objectives.

We will examine several cutting-edge methodologies and approaches, such as reinforcement learning for human feedback, scalable oversight mechanisms, and mechanistic interpretability. In addition to technical solutions, the presentation will briefly cover recent developments in AI regulations, such as the EU AI Act, highlighting their impact on AI governance and the importance of regulatory frameworks in promoting responsible AI deployment. We will provide an opportunity for participants to engage with, evaluate, and discuss these concepts.

经济与金融学院

2024年6月6日