From a perusal of the title, it might appear that this book is one of a few that could be classified as "futurism" or "future-projected technology". These books, which have mostly appeared in the last five years or so, have an extremely optimistic view of future developments in artificial intelligence, but most of them do not justify this optimism with rigorous scientific evidence or attempt to quantify what is means for a machine to exhibit intelligence.
This book, first published in 1997, and appearing in paperback last year, is however different in this regard. In the book the author attempts, and in general succeeds, in giving the reader an overview of the status of artificial intelligence as it was in 1997. It does project these developments out to the future, even to the year 2050, but it does so in a way that is free of the overindulgences of media hype and Hollywood exaggerations that frequently accompany "semi-popular" works on artificial intelligence. Even though it is targeted at readers that are not specialists in artificial intelligence, the book does enable readers with a general education to understand just how advanced machine intelligence was during that time. Most importantly, the author strives to identify what it means for a machine to be intelligent, and his proposals for defining and measuring machine intelligence are quite interesting and show keen insight.
Indeed, the author's views on intelligence, machine or otherwise, are quite refreshing, for he does not make them human-centric. Other species exhibit intelligence in ways that are unique to them and highly suitable for their survival. The author emphasizes that life forms or machines have a degree of intelligence that is appropriate to themselves and the contexts and environments in which they are situated. Humans he says, via technological development, are bringing about machines that may very soon exhibit intelligence that is highly competitive to that of human intelligence, but this is to be measured relative to the needs of each, and these needs may conflict. The author is concerned with this potential conflict, and he devotes a sizable portion of the book in elaborating on just how it may come about.
Throughout the book the author endeavors to contrast the differences between human and machine intelligence. The fact that humans behave and perform differently makes any comparison between machines and humans problematic he believes. The absence of a `typical' human as a standard of comparison for machine intelligence implies that other measures must be devised for estimating this intelligence. And, just as there is high variability among human performance and ability, it is to be expected that this would also be the case for machines. The machines will differ in their respective abilities and with respect to humans. In some instances these machines will "outperform" humans on various tasks, as they have done in many cases up to the time of publication of this book and at the present time.
Another interesting difference between human and machine intelligence that the author points out concerns what has been called `domain-specificity' by many researchers. In the author's view, machines that are performing "intelligent acts" do so only in certain domains that are highly specified. A machine adept at chess for example may not be good at doing network management. Humans though can think and accomplish goals in many different domains: they can be good chess players as well as good network managers. However currently there is much debate among cognitive scientists as to whether human expertise is the result of a collection of specialized modules that interact in some way or whether it is the result of a "general" type of module that can think in many different domains. The author does not indulge himself in this debate, but instead emphasizes that machines and humans in general exhibit different types of intelligence. It is only when their performance on specific tasks is compared can one say whether a machine is "smarter" than a human, or vice versa.
In the author's view, both humans and machines can learn both in a "passive" and in an "active" sense. Passive learning is closer to what one would describe as "memorization", whereas active learning involves the deliberate initiation of the learning process. Scientific investigation would be an excellent example of active learning, for it involves setting up equipment, taking measurements, etc, in order to test a particular hypothesis or hunch on the part of the investigator. Clearly some machines can do passive learning, via the addition of extra rules or data, but not all can, says the author. Machines can also perform active learning and the author discusses an example of how this is done. While doing this he diverges into a discussion of the `frame problem' in artificial intelligence, which he dismisses as not being a limitation of machine intelligence, giving examples of just why he takes this viewpoint. The frame problem, he concludes, is just as much a problem for humans as well as other life forms.
Particularly insightful is the author's discussion on the advantages of using neural networks for learning rather than depending on expert systems. He is careful to point out that artificial neurons are not exactly the same as human neurons, and therefore that artificial neural network brains will be different from human brains. The performance of these artificial brains will therefore be different, and thus their intelligence will be. The author then asks the reader to consider what goals are to be accomplished using these artificial brains. Since the construction of these brains is done to get something different from human brains, the advantages in using them must be delineated. In the author's view they must go beyond the limitations imposed by the human brain. The author spends over half of the book describing what has been accomplished in the actual construction of artificial brains, with emphasis on the activities of his laboratory at the University of Reading in the UK. All of this discussion is fascinating reading.