Artificial Minds

Rating:
Summary: Excellent Survey of Mechanisms for Minds
Review: As a former student of Dr. Franklin, I am probably somewhat biased. I feel, however, that I can still provide an objective overview of this well written book.

The goal of Dr. Franklin's book is to put forth his concept of a general mechanism of mind. As an early proponent of autonomous agents, his writing leans heavily in this direction. His new paradigm of mind, described in the final chapter, can be seen to view the mind as a multitude of disparate mechanisms operating on sensations to create information and re-creating prior information with the primary purpose of selecting the next action, within its limits, for an agent to take.

In building toward his new paradigm, Dr. Franklin makes stops on an admittedly biased tour to briefly explain other concepts for mechanisms of mind. Patti Maes, Marvin Minsky, Rodney Brooks, Douglas Hofstadter, and Robert Ornstein are just a few of the pioneers in this field whose works are touched on. The bibliography serves as a must-read list for anyone interested in expanding their horizons on this new frontier.

Along the way, Dr. Franklin also sheds light on three debates central to Artificial Intelligence. They are 1) Can machines truly have minds, 2) Which approach, symbolism or connectionism, will yield a workable mechanism of mind, and 3) What role should representations play in intelligence? While presenting both sides of each argument, it is fairly clear which camp Dr. Franklin is in.

While written for the lay reader, Dr. Franklin's highly accessible writing should provide much food for thought for those practicing in AI as well. I highly recommend this book.

Rating:
Summary: Great sightseeing, but where do I go afterwards?
Review: As is explained by other reviewers, this book is well-written, humorous and thought-provoking. It introduces you (at least, if you're like me, not (yet?) an AI professional) to a vastness of fascinating ideas. So far, it's a great book. Therefore I would like to sum the disadvantages:

* In the last chapters, Franklin loses contact with the ground. Not that he gets too speculative, but that you feel that you haven't really gripped some foundational issues enough to understand, and discriminate between, the more advanced ideas in the latter part of the book. I suspect this is inevitable, that you need some hands-on experience prior to appreciate what's the fuzz about these ideas. This criticism goes only for the last chapters.

* The philosophy chapter is uninspired.

* When I had finished the book, i wanted some advice on how to go on exeperimenting with the ideas and techniques introduced in the book myself. There was none.

But, propably, there is no better book.

Rating:
Summary: What a wonderfully fascinating and thought provoking book!
Review: I was originally searching for a book to fuel my thirst for Visual Basic computer-programming, thinking that this book would give clues to how an artificial mind could be implemented by someone like me. And on that basis I should have given this book 3 stars, but I have realised that the sphere of AI is monumentally vast!

When I first began to read, I thought it was quite hard going, but I became accustomed to the author's formal but chatty narrative. I found the chapter about Animal Intelligence riveting and truly eye-opening.

If you have even a passing interest in either psychology or ambitious computer programming, then you cannot live without this book. To everyone else: you cannot live without this book!

In a word: Inspirational

Rating:
Summary: Good book for the time of publication
Review: This book brings some refreshing new ideas and perspectives to the field of artificial intelligence. Although qualitative in his approach, and written for a "general audience", the author offers many insights into a field whose view of intelligence has been interpreted solely in terms of human capabilities. The author offers several alternatives for defining and measuring machine intelligence, all of them interesting, but needing more integration into practical applications in order to test their efficacy. His ideas though serve as strong perturbations that lift the field out of the local equilibrium it frequently finds itself in. He does occasionally wade into the muddy waters of speculative philosophy in the guise of the "mind-body problem", but the time spent in doing this is more than made up for by his efforts to define and give advice on how to build an artificial mind.

The author is fully aware of the roller-coast ride of confidence and false optimism that the field of machine intelligence has taken in the last five decades. He describes the "brittleness" of AI systems that are usually confined to a specific domain, such as chess or natural language processing, but cannot be used with drastic modification in other domains. It is the opinion of this reviewer that such domain-specific systems are intelligent, but should be classified as a form of "low-level intelligence". Systems that can cross learn across domains have a higher level of intelligence. The author could easily incorporate such a classification since his view of the mind is that it is "continuous" rather than a "Boolean" notion. He wants the reader to consider the proposition that there are "degrees of mind" rather than a discrete, discontinuous jump from not having a mind to having one. Mind can be implemented in machines he says, how much depends on several factors, with the most question to answer is how it can be done, not that it can be done.

Readers who do not want to engage in philosophical speculation may be tempted to skip the reading of chapter 2, which discusses the "mind-body problem". However, there are many interesting discussions in this chapter, and in omitting its perusal such a reader would miss in particular the highly insightful overview of the ideas of Aaron Sloman on free will. Sloman has some interesting things to say about free will, and his ideas are pertinent in an engineering setting, as the reader will discover when reading the author's summarization. The notion of free will as a "Boolean" is dismissed by Sloman, and the author shows how Sloman's view of free will is directly applicable to the design of (intelligent) agents. The view of Sloman, who is a philosopher, is a good example of the trend of many philosophers to move into the field of artificial intelligence. There are many examples of this trend, and this century will no doubt see the rise of many more of what may be called "industrial philosophers."

One of the more refreshing ideas in the book is the author's view that human intelligence should not be viewed as the sole one in existence. Quoting the commandment from ethology that "thou shalt not anthropomorphize" in his discussion of "animal minds" the author then makes a good case for the existence of different types of intelligences. The most interesting consequence of his discussion is the difficulty of defining and measuring intelligence in any kind of organism, entity, or system. Before progress can be gauged in artificial intelligence, it is important that there be tools developed that measure the intelligence gains involved in this progress. So far, these tools have been conspicuously absent.

It is clear that the author has a strong background in dynamical systems, since throughout the book one can see ideas from this field appear again and again. For example, in his discussion on 'symbolic AI', the author defines a 'production system' as a discrete dynamical system that consists of three components: a database, the set of production rules, and the control structure which directs the production rules. Dynamical systems also appear in the 'connectionist' paradigm in AI and thus their use may bring together two proposals for machine intelligence that have traditionally been at conflict with each other.

Along with what the author would call real examples of artificial minds, such as the SOAR project, he discusses other interesting proposals for their construction. One of these is the 'pandemonium model of mind', which makes use of an 'association engine' and 'concept demons'. Such terminology may seem alien to the reader, but the author puts any objections as to its impracticality by pointing out the computational interpretation of this model and its actual implementation in Pascal code. Still another project for creating an artificial mind that he visits is the work of Jose Brustoloni on a formal theory of agents. The author describes his work as one of the first that attempts to give a classification of autonomous agents. Interestingly, Brustoloni believes that one type of agent, namely the problem-solving agent, should not be viewed as intelligent, due to its need for maintaining a world model and the subsequent issues of the frame problem and non-monotonic logics. Also interesting is Brustolini's views on a 'hierarchy of behaviors', beginning with 'instinctive behaviors', and ending with 'theory making'. The author states that he knows of no artificial agent that has shown signs of theory making. As of the time of publication, this was an accurate statement, but since then the artificial intelligence community has seen the rise of intelligent machines that can engage in the construction of scientific theories. Based on inductive logic programming, these machines have indicated abilities that are competitive with human scientists.

The author ends the book with optimistic notes of encouragement for the reader. Considering the content of the book and the developments that have taken place since its publication, there is every reason to be very optimistic about the continued development of artificial minds.

Rating:
Summary: Muddled
Review: This book was pretty disappointing. Not really an enjoyable read. Doesn't seem to be any reason to believe any of the things he says. It all comes across as a bunch of opinions with his own thrown in there as well. Also has the annoying habit of saying "Next we're going to talk about X". You get to the next section and he says, "Now we're going to talk about X". I've read "The Cambridge Quintet" which was a very enjoyable discussion of the philosophy of AI. Also Godel, Escher, Bach is an excellent read for philosophy. I was hoping to get a concise overview of current practices and the reasoning behind them. It didn't happen.

Rating:
Summary: The best history and review of AI I've seen
Review: This is a tremendously witty and scholarly book. It would make a great text for a one-semester undergrad AI survey course in a computer science, psychology or philosophy sequence. It is also tremendously entertaining reading for the educated reader. I found it an excellent reintroduction to areas I've been away from for some years.

Rating:
Summary: The best survey
Review: This is the best survey of AI I've seen. However, I think that it really should have more information on very innovative projects such as copycat. Towards this, I can only recommend Douglas Hofstadter's Fluid Concepts and Creative Analogies, Melanie Mitchell's Analogy-making as perception, and Robert French's Subtlety of Sameness.

Rating:
Summary: Excellent introduction to the subject
Review: This is the perfect introduction for the lay reader who has more than a passing interest in AI. The author is able to explain issues in reasonable depth while keeping his explanations accessible. Some readers might find the author's digressions stray too far from the point at hand, but this reader doesn't fault the author, since the digressions are always fascinating. Comments, footnotes and biblio combine to give you a clear idea of what you'll want to read next.