There are few films that have had the influence of Kubrick's 2001: A Space Odyssey. It is difficult to say how many A.I. researchers were inspired to go into the field by the viewing of the film, but this book is good evidence that research in A.I. has undergone a sort of "public benchmarking" via a comparison with HAL, the arch computer-villan in the movie, and whose name is now ubiquitous in debates on robotics and artificial intelligence. This excellent book is a summary of just how far A.I. has come, as compared to the abilities of HAL. All of the authors acknowledge that the present level of A.I. is not what it is in HAL, but that progress is being made, in spite of, as one author remarked, the fact that A.I. is one of the most profoundly difficult problems in science. Marvin Minsky is interviewed in the book, and he believes it is the problem of knowledge representation that has slowed down the progress in A.I. Too much time has been spent on chess programs and character recognition, and not on fundamentals, Minsky argues. Minsky also believes that emotions are absolutely essential in intelligent problem solving, and this is interesting in light of recent work on computational models of emotion.
The large physical size of HAL is compared with present sizes of computers in the article of David Kuck. But, the author argues, the human brain is also much smaller than HAL, but such a marvelous computing machine has yet to be emulated by even the best of present-day supercomputers. Size does not matter in matters of intelligence.
The article on error correction by R. Iyer is particularly insightful, for he argues, quite correctly, that making mistakes is a characteristic of true intelligence. HAL's brag that it is "incapable of error" is a sign of stupidity, rather than intelligence. False leads, uncompleted strategies and goals, and inconsistent reasoning are all part of genuine problem-solving.
M.S. Campbell discusses the efforts in machine chess, with Deep Blue of course being the focal point. Comparisons of how HAL plays chess compared to Deep Blue illustrate Campbell's belief that a mere combinatorial approach to playing chess should not be considered an intelligent one. HAL is intelligent because it plays intelligence like a human does, but Deep Blue does not. Campbell also gives an interesting historical summary of computer chess. He cites that example of a "fortress" position as one that chess computers have severe difficulty with.
The ability of computers to speak, and not sound "machine-like" is the subject of the article by J.P. Olive. His discussion gives the reader both an historical and technical overview of the difficulties involved in computer speech. And yes, the song "Daisy, Daisy" is real, and was the first song ever sung by a computer.
R. Kurzweil gives an optimistic overview of computer speech recognition. He details the "lessons" one needs to perform speech recognition and repeats his well-known desire to "reverse-engineer" the human brain, once the computational ability becomes available.
R.C. Schank discusses why he thinks A.I. must be able to understand more than just words, if natural language processing is to become a reality. And, interestingly, he believes that intelligence and problem solving are not necessarily the same thing and gives the reader insight into the difficulties involved in language acquisition by a computer.
This book was first printed in 1997, but given the recent news of a common sense reasoning computer developed by the company that D.B. Lenat heads, his article is particularly interesting. The ability of computers to exercise common sense is considered a fundamental requirement for machine intelligence, and Lenat outlines "3 easy steps" for the building of a HAL-like computer. It remains to be seen whether Lenat has indeed achieved this.
The ability of HAL to recognize images is discussed in the article by A. Rosenfeld. In light of the recent need for face recognition software for security enhancement, this discussion is particularly interesting. Rosenfeld details just what is difficult and what is easy to do in this area.
The capability of a computer to perform speechreading is the topic of D.Stork's article. Noting that speech has both auditory and visual aspects, Stork argues that HAL's ability to perform speechreading is far advanced in comparison to the real computers of today. He discusses interestingly, some of the tools used in current research on speechreading, such as hidden Markov models and neural networks.
The article of D. Norman addresses the human factors involved in living with intelligent machines. Norman contends that 2001 is too optimistic a projection of the future, and, in addition, that its technology is too large and bulky to be of much use to humans.
Computational models of emotion have been a subject of intense interest of late, and R. Picard addresses the issue of emotion and computers in his article. The discussion is fascinating, and Picard clearly believes that emotional states can be recoginized and mimicked by a computer. "Emotional" intelligence is something that must be part of any notion of artificial intelligence.
The ability of HAL to manipulate circumstances and trick the crew members is addressed in D.Wilkin's article. The ability to plan is based on common sense issues, and this is no where near being realized, he argues.
D.Stork interviews Stephen Wolfram in another article, and the reader learns of Wolfram's opinions on A.I. and related issues. Wolfram believes that its the nature of intelligence that must be understood to develop thinking machines, and not just processing power.
When thinking machines are built, legal issues of responsibility will naturally arise. D. Dennett addresses this in the last article of the book. Dennet puts HALs behavior in the context of what a human might do when faced with the mission priorities given to him. Turning off HAL was justified, but so was HALs response to it.
