Stuart J. Russell

I probably shouldn’t have wasted my time reading this book which was written in about 2014. The basic problem is that AI technology is moving so fast that a book like this is out of date before it hits the bookstores. And that’s the basic problem with this book. Apart from that I’m not sure that he really comes to grips with the real issues of safety with superintelligent machines running the world. Certainly he teases out a large number of issues. But I’m still fearful that no show more matter how clever we are in initially programming super intelligent machines they will rapidly outsmart us and we will become like the gorillas in his example.
Still a few segments caught my interest and I’ve extracted some of these below: (Oh, only fair to state, that I’ve only read this in the Blinkist summary version ...which means one misses a lot of the nuances).
Rethink your fundamental assumptions about AI.....The risks and downsides of new technologies are far too often left unexplored, as scientists and engineers fixate on their feverish quest to realize the utopias of the future.
We need several breakthroughs in software before AI surpasses human intelligence. [I suspect that since he wrote this we have nearly made up the ground]. In terms of raw computing power, the Summit Machine actually slightly exceeds the human brain, although it requires a warehouse full of hardware and a million times more energy.
The real problem in designing intelligence is in the software. We can never really say when conceptual breakthroughs will take place. But one thing’s for sure–we shouldn’t underestimate human ingenuity.
We’ve been operating under a misguided conception of intelligence......The concern is that in a world controlled by superintelligent AI, we’d be in much the same position as the gorillas.
An AI’s intelligence is measured simply by how well it can achieve a pre-given objective.
We might ask a superintelligent AI to find a cure for cancer, only for it to start giving people cancer in order to do experiments on them......Instead of just intelligent machines, we should be designing beneficial machines.....
1. The key message here is: Instead of just intelligent machines, we should be designing beneficial machines......AI should only have one objective, which is the maximal fulfillment of human preferences.
2. The second principle is that the AI should initially be uncertain about what those preferences are.....This is the humbleness principle.
3. Uncertain AI will allow themselves to be turned off. This is because it would interpret a human trying to turn it off as having the preference that it be turned off.......Their ultimate source of information about human preferences should be human behaviour. This is called the learning principle.
These three principles represent an alternative understanding of what genuine intelligence involves:.....AI with this kind of intelligence would be much closer to human intelligence, since we’re also capable of examining and altering the goals that we strive towards.
We can expect AI to benefit us in many ways......Virtual lawyers, for example, are already vastly outperforming real lawyers at sourcing legal information quickly......With the help of superintelligent AI, scientists will no longer have to sort through immense amounts of published research,
AI is going to make life less secure for everyone.
Mass automation will either liberate humanity’s potential or debilitate it.
My overall take on the book. Raises some serious questions. Doesn’t really have the answers and is now seriously out of date. AI is with us now. Three stars from me. show less

Here's a book from which any reader can get an authentic glimpse of current AI research from a practicing expert in the field. The prospect that true artificial superintelligence may someday be realized raises concerns about what Russell calls the "gorilla problem" (humans' status vis-à-vis robots could become like gorillas' present status vis-à-vis humans!) and the "King Midas problem". He discusses a precautionary approach he calls "beneficial AI", wherein (1) an AI's only objective is show more the realization of human preferences, (2) those preferences are initially unknown to the AI, and (3) the AI's inferences about the preferences ultimately must come from human behavior. Still needed, he says, are major advances in computational technology ("deep learning", for example, won't suffice) and in moral philosophy (produced, I presume, by philosophers who believe in reaching definite conclusions). show less

Important book that covers three main areas:
- a timeline of AI with suggestions on how it will likely evolve and affect us
- an articulation about the risks of AI and what might mitigate them
- an extended discussion that shows how hard it will be to actually implement these controls

The first two areas are extremely well covered and original. The last part of the book is less so and seems more muddled and confusing while trying to advocate for further research in problems which likely do not show more have a clear solution. show less

I had to read this as part of my Artificial Intelligence course in Georgia Tech's Online Master's in Computer Science program, and as an AI textbook it was excellent. It provides detailed, and easy to follow, algorithms ranging from minimax and alpha-beta to Bayes Nets, Hidden Markov Models, A*, Neural Nets, and plenty more. I did not read every page of this book, but I can attest that I would not have done nearly as well in my course without it and if I need to look up an AI algo, I'll turn show more here first to read what Russell and Norvig have to say first and then check other resources. show less