Machines Who Think: A Personal Inquiry into the History and Prospects of Artificial Intelligence

Rating:
Summary: An insightful view of AI from an insider in the field.
Review: The field of artificial intelligence, in terms of its research content, and the confidence it expresses in the results of this research, has executed a roller coaster ride in the last five decades. There have been many proposals, many leads not going anywhere, but just as many leads showing great promise but were abandoned. The reasons why they were abandoned are unclear, but many researchers in artificial intelligence have let them themselves be persuaded that their results do not reflect real intelligence. This has thwarted the development of many promising areas in artificial intelligence, which could have been highly developed by now.

The author, in this new edition of her book, has given the reader her opinions of the status of artificial intelligence in the twenty-five years after the first edition of the book. Her assessment of the last twenty-five years is in general optimistic, but her review concentrates mostly on research in the academic setting. There have also been dramatic advances in artificial intelligence in the commercial sector in the last twenty-five years, but many of these are difficult to document, since issues of propriety arise in the business environment. The many applications that are used by business and industry are practical proof of the rise of machine intelligence in the last twenty-five years, and many of these make use of the academic developments that the author discusses in this book.

The self-doubts and concentrated attention expressed by various researchers are well documented by the author, and some interesting historical anecdotes are included. The author describes the "odd paradox" in artificial intelligence as one where the its practical successes are absorbed into the domains in which they found application. Once assimilated, they become "silent partners" alongside other (non-intelligent) approaches. This reinforces the belief that the intelligent applications were not intelligent in the first place, and are then viewed merely as "valuable automatic helpers". This scenario has been played out many times in the history of artificial intelligence, and, even worse, the fact that the workings of these applications were understood made many assert that this was proof of their non-intelligence. If a process or algorithm is understood, it cannot be intelligent. This bias, the author correctly observes, continues to this day. Regardless of these beliefs or prejudices, the fact remains though that many of today's computers and machines are packed full of intelligence, albeit in different levels, and these levels will dramatically increase in the next twenty-five years.

Researchers in artificial intelligence have been accused of exaggerating the status of machine intelligence, and similar to the same exaggerations that occur in other fields, which arise many times from pressures to obtain funding, these accusations do have some truth to them. But the author points out a case where the funding was cancelled due to the project not being "extravagant enough." This is an interesting historical fact, and one that illustrates the large swings in confidence that have plagued AI research from the beginning.

The strong emphasis on emulating human intelligence has been dampened in recent years, with researchers realizing, refreshingly, that human intelligence is not the only kind in nature. It is in retrospect quite surprising that silicon-based machines were thought to be able to mimic the processes and powers of biological systems. The author quotes one researcher as saying that "Silicon intelligence would surely be different from human intelligence". This is indeed correct, and one can expect many different types of intelligence to reside in future machines, each of these types emphasizing particular tasks, but being general enough to think in many domains. Maybe a better word for describing the field would be to call it 'Alien Intelligence', so as to emphasize the (non-human) idiosyncrasies of these different intelligences.

With very exceptions, the philosophical community has been against the possibility of artificial intelligence. This continues to this day, and the author discusses some of the philosophical tirades leveled against artificial intelligence since the first edition of the book. Researchers in AI have taken the time (unfortunately) to answer some of these criticisms, but there is also a trend, which hopefully will continue, to ignore them and instead spend time on what is important, namely the design and construction of intelligent machines. There is no penalty in ignoring philosophical criticism; it lends no constructive insight into artificial intelligence. However there is a great penalty taken in the form of wasted hours in attempting to answer the vague and impractical claims of philosophers. Ironically, there have been a few renowned philosophers that have left the practice of philosophy and have entered into research into artificial intelligence (and have done a fine job in this regard).

The author also shares with the reader her personal insights into artificial intelligence, and these are interesting considering her involvement with some of the major academic experts in AI. She describes her bias in thinking of (mobile) robots as the sole representative of artificial intelligence. This bias has been alleviated to a large degree in the last decade, but many still equate artificial intelligence to the presence of bipedal robots wandering around performing useful tasks or possibly acting as adversaries to human beings. The latter view of course is very popular in Hollywood interpretations of artificial intelligence. The real truth though is that (immobile) machines, be they servers in networks, laptop computers, or other types of machines, can exhibit high levels of intelligence, depending on what kind of "software" or "mind" is overlaid on them.

The most important thing to be settled for the field of artificial intelligence, and this is brought out also in many of the author's remarks, is a general methodology for gauging machine intelligence. The Turing test is too subjective and tied too much to measures of human intelligence. The AI community definitely needs to arrive at quantitative measures of machine intelligence in order to assess progress and allow the business community to judge more accurately whether a certain level of machine intelligence is needed for their organizations.

Rating:
Summary: An Update to a Classic
Review: Twenty five years ago Pamela McCorduck wrote a definitive book on Artificial Intelligence. She first went among the artificial intelligentsia when the field was fresh and new, and she asked the scientists engaged in it what they were doing and why. She saw artificial intelligence as the scientific apotheosis of one of the most enduring, glorious, often amusing, and sometimes alarming, traditions of human culture: the endless fascination with artifacts that think.

Long out of print, it became a classic, often quoted, but not often read. Now it's back and in a new edition with an extended afterword that brings the field up to date in the last quarter century, including its scientific and public faces. McCorduck shows how, from a slightly dubious fringe science, artificial intelligence has moved slowly to a central place in our everyday lives, and how it will be even more crucial as the World Wide Web moves into its next generation.