Human Compatible: AI and the Problem of Control
人类兼容:人工智能及其控制问题
Author: Stuart Russell
This book presents a high-level overview of the concepts of AI, related technical challenges, and the long-term risks and solutions involved. It hit the perfect level of technical depth for me and is available as an audiobook, which is rare in books on AI. This is an outstanding book in the AI genre. It presents a clear description of the challenges, but unlike other AI books, it doesn’t expound on the problem but rather proffers solutions.
 

Stuart clearly explains a variety of concepts, from AI, to blockchain, to quantum computing, using a sense of humour. He also suggests many exciting approaches to AI research, while also considering the implications. Stuart basically proposes an alternative to specifying aims into software and describes how a machine would do better in inferring and continually updating human preferences. Human Compatible is a great introduction to the AI alignment problem, written for a larger audience than are many other AI books.
 

The book explains how superhuman artificial intelligence is approaching an apocalypse that threatens not just human relationships and jobs, but the essence of civilization itself. It is now clear that a conflict between humans and machines is inevitable, and its outcome very predictable. This is because in this age, there is more appraisal of intelligence and little application of it.
 

Stuart argues that this apocalypse can be avoided. Although humans are misusing AI, and if the predicted breakthroughs occur and superhuman AI emerges, we would have created entities that are more powerful than we are. But if AI is rebuilt on a new foundation where machines are designed to be uncertain about the human preferences they are to satisfy, they would be altruistic and committed to pursuing our objectives, not theirs. This new foundation would allow us to create machines that are provably deferential and beneficial.
 

This book teaches that solving the problem of control over AI is not just possible, but is the key to unlocking the door that opens into a future of unlimited promise.
 

本书作者Stuart Russell是加州大学伯克利分校计算机科学系教授，也是美国人工智能学会的成员。他的这本书可谓具有开创意义，书中介绍了对于人工智能技术的一些误用以及人工智能技术所能为人类带来的近期利益。与此同时，他还解释了为何我们应该在新的基础之上重建人工智能系统，在这个系统之中，人工智能机器应该为我们的目标和我们人类服务。这是一本以事实和真理为基础的著作，作者对人类在人工智能时代所要体验的旅程持乐观态度。