Human Compatible: Artificial Intelligence and the Problem of Control

Writer: Stuart J. Russell

In Human Compatible, Stuart J. Russell, a leading AI researcher, addresses the pressing challenge of aligning advanced artificial intelligence with human values. He critiques the traditional AI paradigm that focuses on optimizing fixed objectives, warning that such an approach can lead to unintended and potentially catastrophic outcomes if AI systems pursue goals misaligned with human intentions. Russell proposes a new framework where AI systems are designed to be inherently uncertain about human preferences, learning and adapting through continuous interaction. This approach emphasizes that AI should defer to human judgment and seek clarification when in doubt. The book delves into technical concepts like inverse reinforcement learning and discusses the broader implications of AI on society, ethics, and governance. Russell’s insights offer a roadmap for developing AI that is beneficial and controllable, ensuring that as machines become more capable, they remain aligned with human well-being.