Human Compatible: Artificial Intelligence and the Problem of Control.

• Author: Hemphill, Thomas A.

Human Compatible: Artificial Intelligence and the Problem of Control

Stuart Russell

New York: Viking Press, 2019, 354 pp.

Stuart Russell, professor of computer science and the SmithZadeh Chair in Engineering at the University of California-Berkeley, has succeeded in writing a very accessible book on artificial intelligence (AI), a truly revolutionary technology for society. As the book title indicates, Russell's treatise is focused on the problem of society controlling the development and application of AI. For those uninitiated to the technology, there are three stages of AI: (1) artificial narrow intelligence (ANI); (2) artificial general intelligence (AGI); and (3) artificial super intelligence (ASI). In the first stage, AI machines are capable of performing a singular task at the same level or better than performed by humans. Moving to the second stage, AI machines are able to reason, solve problems, think in abstractions, and make choices as easily as humans can, with equal or better results. Finally, in the third stage, AI involves systems ranging from slightly more capable at performing human cognitive tasks to those that are trillions of times smarter than humans. In his book, Russell is particularly concerned with the potentially lethal consequences to humanity of ASI.

Russell divides the book into three parts. The first part, covering Chapters 1 through 3, explores the general concept of intelligence in humans and in machines. He provides an excellent overview of AI in this section of the book, both historically as well as how AI will be utilized in such examples as self-driving cars, intelligent personal assistants, and smart homes and domestic robots. In the second part, covering Chapters 4 through 6, Russell identifies problems arising from imbuing machines with intelligence and recursive self-learning capabilities, focusing on the problem of retaining absolute power over super intelligent AI or ASI machines. The third part of the book, covering Chapters 7 through 10, reviews what Russell characterizes as a new way to think about AI that ensures machines remain beneficial to humans, now and forever. The latter half of his book deals with control problems with AI and policy issues.

Explaining what Russell calls the "gorilla problem" offers us insights into the threat of super intelligent AI machines to humanity. As Russell notes, "Around ten million years ago, the ancestors of the modern gorilla created (accidentally, to be sure) the genetic lineage leading to modern humans. How do gorillas feel about this? ... Their...