Times piece "Nobody's Smart About Intelligence" (March 1, 1998), he offers this lament: "IQs are up. S.A.T.s are down. Americans flunk math and prosper. Somebody with brains should figure this out." The best anyone can offer, Johnson claims, is a conjecture that the complexity of everyday life (programming small electronic devices or calculating the latest projection of your net worth when you retire) has stretched and exercised our brains into faster and more agile computing engines. This might explain our increasing IQs while MTV and video-game overload might explain our increasing ignorance and declining capabilities at the logical plodding deductive thought of traditional intelligence. So what type of intelligence is AI trying to create? Astro Teller (New ~rk 7qmes, Op-Ed, March 21, 1998), suggests that no matter the type, building intelligences will make our world better as we learn more about our minds and who we are. But what will we understand? How better to exploit our neighbors or sell them goods and services at ever increasing profits? Will we understand the difference between Gandhi and Saddam? Mozart and Madonna? Or just what is it that everyone finds funny about Seinfield? The recent pinnacle of AI achievement has not come from our half century long quest to pass the Turing Test, but from our fascination at a machine beating a human at the complex task of playing chess. Deep Blue, a parallel supercomputing creation from IBM for processing hundreds of millions of chess moves per second, is the hardware and software that realized this • e.,o • oe. eeeoeleQo e*e eolee • oo t • • e#= The recent pinnacle of AI achievement has come from Deep Blue, a machine that beat a human at playing chess. But what kind of intelligence is this? impressive accomplishment. An ancient game, long attacked by AI'ers and now empirically conquered. Everyone can honestly admit that Deep Blue doesn't have a clue about what it is doing, so self awareness is not an issue. It just "knows" the next best move from an intensive search. So what kind of intelligence is this? It is dearly '~AI Intelligence," a smart machine; honored byAI associations and foundations with a small pile of cash (compared to IBM expenses!) and a Newell Research Medal. New York Times piece on Deep Blue, asked whether Deep Blue is indeed intelligent. He offered that although human chess grandmasters don't do exactly …
T wenty years ago, if you wanted to work on computer graphics, you had to have access to expensive graphics terminals. The enterprise has evolved to the extent that one rarely considers writing an application that is devoid of a GUI. Well, mostly. A large number of us continue to persist on command line interfaces. However, the introduction of Java is quickly taking care of that. Yet another Java article? Nod This installment is about a similar evolution that has been quietly taking place in the area of robotics. It used to be that, besides lots of money, one needed to have expertise in electrical engineering , electronic control, and a good mechanical engineering background to put together a robot like SHAKEY for purposes of modeling AI theories about intelligent agents. Do not get me wrong, one still needs these disciplines. Just look at the effort going into the building of COG in Rodney Brooks' lab at MIT. However, there has also been an evolution of platforms that can be used for building small mobile robots that are about as expensive as adding an additional disk drive to your desktop computer , and about as easy to use (or easier!) as Microsoft Word. Besides, they give you functionality that is not too distant from that employed in the Mars rover. Several schools have started integrating these platforms into their AI courses. In my mind, the next revolution in computer science education is here. Teaching AI using an Embedded Agent theme is an extension of the agent-centered approach. As discussed in the last installment of this column, the problem of AI is seen as describing and building agents that receive perceptions as input and then output appropriate actions based on them. Consequently, the study of AI centers around how best to implement this mapping from perceptions to actions. By using small robots, the embedded agent perspective takes the agent-oriented approach one step further; rather than study software agents in a simulated environment, we embed physical agents in the real world. This adds a dimension of complexity as well as excitement to the AI course. The complexity has to do with additional demands of learning robot building techniques but can be overcome by the introduction of kits that are easy to assemble. Additionally, the kits are lightweight, inexpensive to maintain, programmable through the standard interfaces provided on most computers , and yet …
In his keynote address to the Autonomous Agents 97 conference, Danny Hillis, vice president of research and development at Walt Disney Imagineering, listed four "holy grail" items with respect to entertainment agents:1. A computable science of emotion,2. Virtual actors,3. Agent evolution, and4. Computable storytelling.
Finding information is a common and fundamental task for Internet users. Search engines such as AltaVista and indices such as Yahoo! are the tools commonly used for this task. Users type in queries using keywords to indicate their interest or navigate through hierarchical directories of topics, eventually obtaining (usually quite large) collections of Web pages.
This brief book will not be much help to the day-to-day AI practitioner but could be used as a point of departure. The author is Distinguished Professor of Computer Science at the Center for Advanced Computer Studies of the University of Southwestern Louisiana. He has written four previous books about computer science. I am in sympathy with the author's intent, but I found that he cracked the concrete of his conceptual support and then left a flabby interior of amorphous specifics to be gutted by rather too sharp conclusions. The author states that creativity, especially in technology, can be discussed within a computational framework. Specifically, he concerns himself with a plausible explanation of Wilkes' invention of microprogramming in 1951. In general, the author puts forth hypotheses related to the specific which corroborate findings by others that creativity is not significantly differing from other everyday mental processes.
Traditionally, visualization is the transformation of data into information that can be rendered using computer graphics techniques. Visualization combines techniques and representations from computer graphics, computer vision, and image processing. In distributed and hierarchical operations and processes, such as for command and control or logistics and planning, visualization is the central mechanism for communicating the state of the situation and operations. The major challenge of visualization is to filter, tailor, and present the information in compact forms that can be efficiently created and displayed. In contrast to many visualization problems, these domains have what seems to be an overwhelming variety and quantity of information. We address aspects of this problem with the graphical visualization of abstract temporal information in a concrete spatio-temporal framework.
This book is a reprint from a special issue of the journal Machine Learning on Genetic Algorithms (GAs). The fact that it was the third special issue of the journal on GAs shows the continuous increase of interest of people in the field. The five selected articles were presented at the Fourth International Conference on Genetic Algorithms, in June 1991 (San Diego), and at a Special Workshop for Machine Learning at the same conference. The intended book audience is both researchers and practitioners in the field.
Those of us who are newly arrived on the agent scene have a variety of books available with which to increase our agent vocabulary. In this article, I provide a short review of five books related to agent research. I also discuss several overview papers and the recently published proceedings of the 1997 Autonomous Agents conference. Rather than providing an in-depth, critical analysis of these books (most of which are edited collections of papers), I describe their overall approach, attitude, and quality. These books are not ones that I chose from an exhaustive search of all potentially relevant available books but are those that quality publishers of AI/philosophy texts provided to me for review or those that I purchased for my own education. I have not selected books that treat only one aspect of agent operation in detail (e.g., Rosenschein & Zlotkin, 1994) in order that I may present you with books that significantly expand your agent literacy.
Here we have a set of papers by a number of researchers (students and faculty) at the time of the revival of connectionism, and some of the students have since made their names in the field (e.g., Bookman, Miikkulainen, and Regier). The proceedings is from the second Connectionist Models Summer School held at Carnegie Mellon University in 1988 and organized by Dave Touretzky with Geoffrey Hinton and Terrence Sejnowski as advisors.
In order to design and realize intelligent autonomous agents, a very active research trend has concentrated its attention on the study of models of mental activity, encompassing the explicit representation of mental attitudes such as beliefs, desires and intentions. In such studies mental attitudes are normally represented as data structures on which a so-called interpreter operates determining the overall agent behavior. In this article we propose an original point of view about mental activity modeling, founded on two basic claims: i) mental attitudes should be regarded as autonomous active entities; ii) an intelligent agent should be conceived as distributed structure, where global behavior is produced by interactions among active mental entities.
Description logics have a history of success in configuration applications in major companies including AT&T (mentioned in this paper) and the Ford Motor Company. While we have produced a number of commercial configurators, we find a demonstration application to be the best expository tool for describing how description logics can be leveraged effectively in tasks such as configuration.
The ISO Prolog standard took 10 ars to produce. For a language II that, on its face, is both simple and logical this seems like a long time to spend on standardization. While this book is in no sense a history of the standardization of Prolog, it does provide some insight into why the process took so long. Anyone who has tried to learn Prolog knows that to use the language effectively one must have a sound understanding of the computational model. (Of course, this is true of any programming language, but for Protog it seems to be especially so.) The standards group was to find plenty of room for dispute over the computational model. A major service that this book performs is to provide a clear description of the Prolog inference engine.
The book is based on the author's Ph.D. dissertation and is written primarily for researchers in AI and practitioners in the field of knowledge-based systems development who, in addition to developing knowledge-based systems, also want to understand the model-semantic foundation in the development of knowledge modeling languages. It can also be used as an advanced textbook for knowledge modeling.
This book is primarily an exposition of theories of mind and of recent attempts to invent artificial minds. It reviews a large body of work in the fields of artificial intelligence (AI), artificial life, cognitive science, and neuroscience. It is also informed by work in philosophy and in biology, especially neuroscience and the study of animal behavior. The author's own view of the nature of mind emerges largely through his assessments of the various theories he presents. He believes that recent research is leading toward a "new paradigm of mind" (p. 421), a viewpoint which will be appraised later in this review. Franklin writes with a clear and pleasant style, and his expository descriptions are aimed at a broad audience. The reader should have some general mathematical and scientific knowledge, including the basic ideas of how computers and programming languages work, but otherwise no special technical background is required. The level of difficulty of the book is a little higher than that of a Scientific American article.
John Haugeland, professor of philosophy at the University of Pittsburgh, is well known in the Artificial Intelligence community. With interests --- among others --- in the philosophy of mind and philosophical psychology, he has produced numerous thoughtful works in the past and continues to do so. The title under review,Mind Design II,is the revised version of his 1981 bookMind Design.Other Haugeland books includeArtificial Intelligence: The Very Idea[Haugeland 1985] andHaving Thought[Haugeland 1997].
A rtificial Intelligence and Mobile Robots is a compilation of 13 chapters by well respected robotics researchers describing some of their most recent work. The book aims to provide a "how-to" guide to building the control system of a mobile robot. These case studies are definitely worth studying: seven of the 13 have placed highly in various robotics competitions. In the introduction, the editors give a good, yet brief, history of intelligent robots. The book is then divided into three sections.
This book describes various neural networks and their behavior based on David Marr's theory of vision and information processing systems [Marr82]. The author, Richard Golden, is an Associate Professor in the Program of Cognition and Neuro-science at the University of Texas at Dallas. He has published a number of papers in the area of artificial neural network (ANN) analysis within the past 10 years.
Practical Application": That should be the goal of all theory and technology development. This review offers opinions written from that perspective. The review started with the question, "Can this book be used as a guide to implementing a new technology by someone with related knowledge?" Generally, the reader will find a very useful book with application insights and extensions to the author's previous developments. Each chapter gives concept, theory, and application. However, except for expert peers in fuzzy engineering, statistics, and communications engineering, it does not explain the applications in sufficient detail to give a clear path to implementation. The concepts discussed in this book are part of the more general data mining technology.
Suppose that you want to use your intelligent programmer's assistant to write a program, for example, one that will insert elements into a balanced binary tree. One way you could proceed would be to provide your assistant with a complete specification for the program. This is the approach, influenced by the "water-fall model," taken in classical software engineering. But here is another idea. You give your assistant some examples of how you want the program to work, and ask him to write the program. If the assistant is a bright undergraduate, the examples should be enough. What if the assistant is itself a program?
