Paradigms of Artificial Intelligence: A Methodological & Computational Analysis

Rating:
Summary: A very thought provoking book, but could be improved.
Review: Paradigms of AI is comprised of three parts: symbolic-paradigm overview, connectionist-paradigm overview, and a critique of both paradigms. The first two parts make up over half the book, and after reading them I was left wondering why the author bothered to spend that many pages on them. Someone new to AI will not gain much from the terse overviews, and someone well versed in AI will probably want to skip these first two parts. On the other hand, the connectionist overview had a few interesting sections on subjects that were new to me: adaptive resonance theory, self-organizing maps, and the integration of symbols into connectionist models. That being said, I felt the third part should have been expanded (i.e. made up more of the book) in a way that gave more concrete examples of the sometimes vague concepts the author attempted to relate. For example, why not place the mathematical treatment related to the computational limitations of connectionist networks from the appendix to an earlier chapter, where it can be well-developed and read so as to support the hand waving? But even more annoying may have been the author's lack of consideration of the computational complexity of connectionist networks, and his failure to mention how unlikely it is that a Turing machine could succeed in simulating a human brain (yes time matters!). He mentions this in a footnote, but does not spend time on the matter. May be the most interesting part of the book was his exploration of the meaning of symbols, and presenting the connectionist/symbolic debate from a phenomenological point of view. The main message seemed to be that AI systems need to take the scientist or engineer into consideration. In other words, what good is a paradigm if it cannot be understood and controlled? Thus, there must be third-person symbols for helping engineers to manage the complexity of the computing paradigm. Although I agree with this from a "practical" point of view, from an investigative point of view focusing only on systems whose computations and modus operandi can be understood seems to be missing an important perspective of AI; that being that the most intelligent systems (humans) living on our planet are sometimes unpredictable, yet are very useful. I believe there will be a time when humans work side-by-side with artificial machines that are not easily predictable or understood, but useful none-the-less. In closing, I recommend part three of this book to anyone who has an interest in the philosophy of arificial intelligence. The first two parts can be safely be omitted.

Rating:
Summary: A detailed review
Review: This book takes a look at the paradigms of two different approaches to artificial intelligence; the symbolic paradigm and the connectionist paradigm. The symbolic approach uses a mathematically oriented way of abstractly describing processes leading to intelligent behaviour, while the rather physiologically oriented connectionist approach, favoured the modelling of brain functions in order to reverse-engineer intelligence. The symbolic paradigm has always been the dominant one following the highly influential analysis of the capabilities and limitations of the perceptron. However the brain-oriented connectionist paradigm emerged to challenge the traditional symbolic paradigm, which was said to be unsuccessful since symbols are insufficient to model crucial aspects of cognition and intelligence.

This book is quite unique in that it takes a step back from the heated debate between the advocates of both paradigms, and gives a neutral standpoint encouraging the reader to make a decision himself. It delivers a methodological analysis of the virtues and vices of both the symbolic and the connectionist paradigm, which most books have often neither appreciated nor really addressed. But the book is much more than a mere checklist of the differences between the two paradigms. The book sets out to develop criteria, which any successful method for building AI systems and any successful theory for understanding cognition has to fulfil. The major theme of this book are methodological considerations regarding the form and purpose of a theory, which could and should be the outcome of scientific endeavours in AI and cognitive science.

Furthermore the author addresses the human subject who is to perform the design or who wants to understand a theory of cognition. The specific capabilities and limitations of the human subject to understand a theory or a number of design steps needs to be an instrumental criterion in deciding which of the paradigms is more appropriate. Furthermore, the human subject's capabilities and limitations have to provide the guideline for the development of more suitable frameworks for AI and cognitive science.

I have only given praise to the author and his book, and why shouldn't I? I am his student, and so know what a great teacher he is. Every Thursday afternoon I attend his Machine Learning lecture and in the same evening I attend his Neural Networks lecture at a world-renowned university. He is a very kind and gentle person, and his lectures are never lacking in humour. While always encouraging the students to think and actively participate, his lectures are well planned and thought provoking, very much like the book in question. I hope this book will give you the opportunity to have a taste of what Dr Achim's lectures are like.