Human Compatible: AI and the Problem of Control

Define como debe ser el futuro de las maquinas inteligentes que deben tener objetivos compatibles con los nuestros y por tanto, como los nuestros son a veces inciertos y cambiantes, deben ser flexibles y preguntarnos sobre ellos.

Good 👍👍👍

I feel like this book should be mandatory reading in this day and age. There is so much hype, ignorance, alarmism and progress happening all at once with AI that would do society in general a great service to include the thoughts and questions included in this book into the current zeitgeist of our time. AI is not going away. In fact, its very nature and nearly limitless possible applications to human society and endeavor are likely to make it one of the most significant technologies humanity has produced. So, I believe we owe ourselves the kindness of engaging with its development and implementation on every level to ensure a safe and prosperous future. This book presents an excellent, accessible and incredibly information dense foundation for that.

The consensus among AI practitioners has been that general AI; a computer system with human level capabilities, is unlikely in the next 25 years. So, nothing to worry about, right? Wrong says Russell and in this exceedingly well written, accessible and thought provoking book, he lays out the 'what if we succeed' scenario. Russell argues that whether we think the super intelligence outcome likely or not, the downside is literally the loss of human civilisation. Strangely, perhaps, we are inured to such talk. Russell isn't and it seems that the AI industry has taken his warnings to heart. The book is practical, it doesn't simply itemise the dangers, it proposes design principles to contain the threat. This is a wake up call, not mere scaremongering. Highly recommended.

An impressive general analysis of the situation and field of Ai through the modern society (even including a mid 2023 foreword tackling ChatGPT and other large language models). Small but significant writing about the possible outcomes of Ai, what could go wrong, how it could go wrong and how me might try to use Ai for our own benefit (if that’s even possible we don’t quite know yet). A must-read for anyone slightly familiar with the field, engaging as it’s informative.