What do you think?
Rate this book
Singularity Hypotheses: A Scientific and Philosophical Assessment
Book by
450 pages, Hardcover
First published November 30, 2012
9 people are currently reading
178 people want to read
Ratings & Reviews
Friends & Following
Create a free account to discover what your friends think of this book!
Community Reviews
Displaying 1 - 2 of 2 reviews
November 6, 2022
All you ever wanted to know about singularity hypotheses and were too afraid to ask.
-1 star because it is already outdated and wayyy too long =)
-1 star because it is already outdated and wayyy too long =)
September 11, 2016
After reading a lot of messianic hyperbole about the Singularity it is refreshing to have a collection of scientifically sound essays about the topic at hand.
Or so I thought.
As a very brief summary: Even if the essays in this book make a claim to scientific treatment, sadly this is nothing but pretense. In fact, the sloppy reasoning of these authors, who are supposed at the forefront of singularity research, exposes Singularitarianism as the religion it is, with only nominal connections to actual science.
The book is divided into four parts. Essays in the first part argue that an intelligence explosion is likely and examine some of its consequences. The second part concerns itself with the values of superhuman intelligences. Part 3 is about the impact of an assumed singularity on humans, while the fourth part brings forward arguments why a singularity seems unlikely.
The first part consists of arguments that an intelligence explosion is inevitable. Even if the arguments seem compelling on a first read, they are quite sloppily reasoned. The authors have a depressingly one-dimensional view of intelligence. In fact, no author bothers to even define what they mean by intelligence. Any singularity hypothesis hinges on the implicit assumption that it is natural for a general AI to want to enhance its own intelligence. But this is only true for a very narrowly defined motivation of that AI, which is generally assumed by singularity apologists to want to gain power. But what if an AI is quite content with whatever power it possesses and instead optimizes its own happiness (however that is defined)? It has been shown over and over that power and happiness in humans don't follow a simple linear correlation. That singularity researchers fail to see this is a major shortcoming in AI research IMO. It is also a major failing of this book that the question of motivation for AIs is not addressed. And finally there is this gaping hole in the argumentation: Humans are, evidently, not able to build machines of superhuman intelligence. So why should machines of human intelligence (the so-called human level AIs) be able to do that?
Part 2 argues for the need to imbue an AI with human-compatible values and discusses constraints on this process. Although authors in this part often have a more nuanced and multidimensional view of intelligence than the authors presented in part 1, their discussion of values and motivations is (sometimes pitifully) underestimating the complexity of the world, economy and the human, let alone superhuman, mind. All authors seem to assume that a general AI will come into being by just flicking a switch, and attune its behaviour according to a human-supplied utility function (a mathematical formulation akin to what motivates humans). This, in my opinion, is a highly mistaken perception of how intelligence actually develops - by education. Part 2 calls for a thorough examination of human values and the human mind, but fails to give answers. I also find it weird that none of the authors propose the obvious solution "ensure that an AI follows human laws." Even if it is not clear that this solution would actually work, it at least deserves critical discussion.
I'll talk about the other parts as I read them.
In summary I find the field of singularity research in a frankly pitiful state, with rampant ideologism and little scientific merit. Almost all the authors display disregard for the actual complexity of intelligence, motivation and values. They seem very convinced of their own hypotheses with little to back them.
I still rate this book quite highly because it has given me a deep insight into the field. Also I cannot deny that I found several ideas to think about.
One remark about the layout quality of the ebook. I got this book for free when Springer Verlag made its entire ebook portfolio available, so I cannot complain. If I had spend money on it I'd been majorly pissed off at the low quality of the layout. Tables, images and mathematical formulas are laid out so badly that large portions of text are hardly readable. From a quality publication I expect better.
Or so I thought.
As a very brief summary: Even if the essays in this book make a claim to scientific treatment, sadly this is nothing but pretense. In fact, the sloppy reasoning of these authors, who are supposed at the forefront of singularity research, exposes Singularitarianism as the religion it is, with only nominal connections to actual science.
The book is divided into four parts. Essays in the first part argue that an intelligence explosion is likely and examine some of its consequences. The second part concerns itself with the values of superhuman intelligences. Part 3 is about the impact of an assumed singularity on humans, while the fourth part brings forward arguments why a singularity seems unlikely.
The first part consists of arguments that an intelligence explosion is inevitable. Even if the arguments seem compelling on a first read, they are quite sloppily reasoned. The authors have a depressingly one-dimensional view of intelligence. In fact, no author bothers to even define what they mean by intelligence. Any singularity hypothesis hinges on the implicit assumption that it is natural for a general AI to want to enhance its own intelligence. But this is only true for a very narrowly defined motivation of that AI, which is generally assumed by singularity apologists to want to gain power. But what if an AI is quite content with whatever power it possesses and instead optimizes its own happiness (however that is defined)? It has been shown over and over that power and happiness in humans don't follow a simple linear correlation. That singularity researchers fail to see this is a major shortcoming in AI research IMO. It is also a major failing of this book that the question of motivation for AIs is not addressed. And finally there is this gaping hole in the argumentation: Humans are, evidently, not able to build machines of superhuman intelligence. So why should machines of human intelligence (the so-called human level AIs) be able to do that?
Part 2 argues for the need to imbue an AI with human-compatible values and discusses constraints on this process. Although authors in this part often have a more nuanced and multidimensional view of intelligence than the authors presented in part 1, their discussion of values and motivations is (sometimes pitifully) underestimating the complexity of the world, economy and the human, let alone superhuman, mind. All authors seem to assume that a general AI will come into being by just flicking a switch, and attune its behaviour according to a human-supplied utility function (a mathematical formulation akin to what motivates humans). This, in my opinion, is a highly mistaken perception of how intelligence actually develops - by education. Part 2 calls for a thorough examination of human values and the human mind, but fails to give answers. I also find it weird that none of the authors propose the obvious solution "ensure that an AI follows human laws." Even if it is not clear that this solution would actually work, it at least deserves critical discussion.
I'll talk about the other parts as I read them.
In summary I find the field of singularity research in a frankly pitiful state, with rampant ideologism and little scientific merit. Almost all the authors display disregard for the actual complexity of intelligence, motivation and values. They seem very convinced of their own hypotheses with little to back them.
I still rate this book quite highly because it has given me a deep insight into the field. Also I cannot deny that I found several ideas to think about.
One remark about the layout quality of the ebook. I got this book for free when Springer Verlag made its entire ebook portfolio available, so I cannot complain. If I had spend money on it I'd been majorly pissed off at the low quality of the layout. Tables, images and mathematical formulas are laid out so badly that large portions of text are hardly readable. From a quality publication I expect better.
Displaying 1 - 2 of 2 reviews