“Human Compatible” by Stuart Russell is a book that will get you thinking at the same time as giving you an insightful peek into AI thought leadership. Russell explores the challenges and risks of developing artificial intelligence, balanced with the values and goals of humans. It’s not a technical book, and so should be accessible to both industry insiders, as well as laymen such as the average digital marketer.
The book begins by bringing the reader up to speed with the current state of AI and talks about how it has rapidly advanced in recent years, much like the rest of the technological revolution we have been experiencing. Russell’s perspective is that whilst these advancements bring significant potential benefits, there is also serious risk involved when development is misaligned with human values or goals. Hence, there is a need for AI that is human compatible, designed from scratch with human harmony in mind. Russell emphasises the need to avoid development that conflicts with humanity.
Russell does provide some technical details around AI and machine learning. This can give you a good overview of the key concepts without overwhelming with too much jargon. There are also discussions on the limitations of AI, and how difficult it is to include human values in the design of machine learning algorithms.
One of the key insights of the book is the importance of uncertainty in AI systems. Russell argues that machines that are too certain in their actions can be dangerous, as they may fail to consider alternative possibilities or outcomes. He advocates for the development of AI systems that are capable of expressing their uncertainty, and of seeking clarification and guidance from humans when needed.
The book excels at providing a balanced perspective on a tricky and misunderstood topic. Popular media has thrived on demonstrating the worst case scenarios of AI that overcomes its creator’s expectations, but there is a groundlessness to the arguments, highlighting the fact that AI is developed with a specific goal in mind. It is rarely the case that AI can be programmed to learn in the exact way humans do.
Overall, “Human Compatible” should be present on the bookshelf for anyone interested in the future of AI and the potential impact it could have on society. Russell’s insights and perspectives are both thought-provoking and informative, and his book provides a valuable contribution to the ongoing discussion about the development of safe and beneficial AI systems.
For us marketers, it’s extremely helpful to have such an insight into AI and where the future lies. We’re experts in our marketing field, and AI is increasingly important to us. Big media companies (Google and Facebook) rely on AI like a black box, not revealing their secrets to any of the advertisers trying to use their products. As such, “Human Compatible” provides a fantastic view over and above the industry constraints, and with plenty of entertainment to boot.