Programmer 7500

home *** CD-ROM | disk | FTP | other *** search

/ Programmer 7500 / MAX_PROGRAMMERS.iso / CLIPPER / MISC / ESIE.ZIP / HISTORY < prev next >

Wrap

Text File | 1986-07-06 | 25.4 KB | 808 lines

ESIE The Expert System Inference Engine History Lightwave July 1986 P.O. Box 16858 Tampa, FL 33617 Copyright 1986 All Rights Reserved. This manual may only be distributed as one file on the ESIE distribution diskette. Such duplication and distribution is authorized without compensation as long as the diskette is a duplicate of the ESIE distribution diskette. This manual may also be distributed in printed form as long as a copy of the distribution diskette is attached. All other distribution is strictly prohibited. Page 1 Table of Contents Introduction . . . . . . . . . . . . . . . . . . . . . . 2 Before the 20th Century . . . . . . . . . . . . . . . . . 3 1900 to 1940 . . . . . . . . . . . . . . . . . . . . . . 5 The 40s . . . . . . . . . . . . . . . . . . . . . . . . . 6 The 50s . . . . . . . . . . . . . . . . . . . . . . . . . 7 The 60s . . . . . . . . . . . . . . . . . . . . . . . . . 8 The 70s . . . . . . . . . . . . . . . . . . . . . . . . . 9 The 80s . . . . . . . . . . . . . . . . . . . . . . . . . 11 Page 2 Introduction This history is designed to give you a brief "run down" on the past of Artificial Intelligence. While the history of AI is not exactly as exciting as Napoleon at Waterloo, we hope you will find it interesting. Those of you who are interested in the socioeconomic impacts of AI may well be excited, perhaps worried, about the direction and potential of AI. For example, a few science fiction authors have claimed that man's purpose on earth is to BUILD a better race that eventually will become the dominant one. We take a different outlook. Man was meant to be and do great things and we need to build great tools to help us do it. After all, if we never invented the spear we would still be wearing animal pelts. We are sure when the spear was first invented, other members of the tribe had serious misgivings about it and warned the young ones to do things the old way. We certainly are not claiming that the road to the successful use of Artificial Intelligence will be an easy one, (there must have been more than one caveman who stabbed himself in the foot with his new spear), but it can be one that provides numerous benefits. The coming of any new technology has always brought on some problems; the successful ones cure far more than they hurt. Hopefully, you will get more and more interested in AI, and meet as many knowledge engineers (KEs) as you can. One of the nicest, and most consistent things, about KEs is our nearly universal desire to talk and think about the future. A conversation with a KE at KE social hour can be invigorating. It is our belief that Artificial Intelligence has real promise to be an important tool in the ascent of man. Page 3 Before the 20th Century The astute reader may well be wondering what this chapter is doing here. Logically, didn't AI start with advent of the computer? Well, the answer is sort of. The IDEA of objects having human qualities has been around probably as long as man has. When Mr. Ug first missed his prey with his new found weapon, say the bow and arrow, he might have thought, "it would be nice if the arrow could find it's own way to the food." There is evidence that certain groups of ancient man thought their weapons had souls and should be appeased before the hunt. While not quite fitting in with a modern definition of AI these were definite feelings toward Artificial Intelligence. Real work towards defining the mathematics and symbolics behind AI can be thought of as beginning with Charles Babbage in the 19th century. Babbage was fascinated with the idea of building machines to do human tasks, and the mathematics that would be required to do such tasks. Babbage was, of course, a mathematician. Theory that was developed during his period is still used and debated today. The Tower of Babel is standard fare in beginning computer science courses. In the Tower of Babel you have three stakes in the ground and around one stake you have donut-shaped pieces. The pieces get consecutively larger in size: | | | x|x | | yy|yy | | zzz|zzz | | aaaa|aaaa | | bbbbb|bbbbb | | cccccc|cccccc | | ddddddd|ddddddd | | eeeeeeee|eeeeeeee | | The object of the puzzle was to move all of the pieces from the starting stake to any other stake with these rules: you may move only one piece at a time, and no piece may have a larger piece on top of it. In a child's toybox is often found the "stake" with the pieces around it, and a couple of wine bottles will suffice for the empty stakes, it you want to try to solve the puzzle. It's not as easy as it sounds. Puzzles such as these, and the human intelligence applied to solve them, were fascinating to Babbage. He theorized that Page 4 the rigors of mathematics could be applied to the mental process so that one common language could be used to transfer that process from man to machine. Page 5 1900 to 1940 This period saw the development of a formalism for computers and Artificial Intelligence. In the early history of computers the two were almost always talked about together - they were inseparable. The goal was to create machines that acted like humans or performed human functions so that humans would no longer have to perform them. The early pioneers in the U.S. were George Stibitz, Howard Aiken, Presper Eckert, John Mauchley, John Von Neuman, Herman Goldstine, and Julian Bigelow. In Britain, Alan Turing contributed substantially to AI and computer science. Nearly every computer in existence today is based on the Turing model. If you've had some coursework in computers, one or more of the above names should sound familiar. They are the fathers of computers, and in a way, the fathers of Artificial Intelligence. For them, the two were one and the same. Page 6 The 40s Computers during the Forties left a lot to be desired. They were used to do real work for the first time, during World War II, to help artillery batteries better aim their projectiles. After the war, the concentration changed: since computers could handle numbers well, shouldn't they handle symbols well? During the Forties, much effort was expended to get computers to work with symbols the same way it worked with numbers. Many attempts were failures, but some successes drove the fire towards building machines that could work with symbols and therefore be one more step closer to thinking. For an interesting book you might want to pick up and read "Cybernetics - Control and Communication in the Animal and the Machine", by Norbert Weiner. The book was published in 1948. Page 7 The 50s The Fifties saw work begin in earnest on the thinking machine - a computer that would reason as a human reasoned. Four of the major institutions involved during the 50s were: Stanford, RAND, Carnegie-Mellon, and MIT. In 1956 John McCarthy held a conference on Artificial Intelligence at Dartmouth. At this conference were, among others, Herbert Simon, Marvin Minsky, Alan Newell, Claude Shannon, and Arthur Samuel. All of these people are considered among the fathers of AI. DARPA was also very interested in human reasoning in the Fifties. Often it was claimed that building expert systems would show the true value of computers to man. An expert knowledge base feasibility study was conducted by DARPA in the late Fifties. The study was labelled MODAPS. MODAPS was eventually built into a usable system for the U.S. Army called A-VIS. A-VIS' main goal was for maintenance of hardware and software on computers of the day. Much funding for AI work came out of DARPA. Three universities in the U.S. took on the leading roles in AI research: Carnegie-Mellon, MIT, and Stanford. Four universities in Britain took on the leading role of AI there: Edinburgh, Sussex, Essex, and Imperial College. Donald Michie, H. C. Longuet-Higgines, R. A. Brooker, and R. Kowalski are all important people in British AI. Stimulated by the impressive gains these people made towards intelligent machines, the press, and the people, overreacted. Science fiction stories exploded on the scene about the power, and danger, of intelligent machines. Everyone, it seemed, was concerned about the impact of thinking machines on their lives. Surprisingly, the overwhelming attitude was positive - people were pro machine. This was due to the promise of less labor and more free time, along with greater prosperity. Page 8 The 60s The Sixties can be classified in Artificial Intelligence by the lack of it. During the Fifties, and somewhat during the postwar period, fantastic and glamourous claims were made for thinking computers. Computers, it was said, would soon solve all our problems by thinking and reasoning and performing like humans. We would use them to find all the tough answers and build machines that would do all our dirty work for us. The letdown from these claims produced the dismal lack of AI in the Sixties. Research was left to a few universities: MIT, Carnegie-Mellon ,and Stanford. The work at MIT centered on building machines to play the perfect game of chess. Researchers reasoned that if they could build a machine that played perfect chess, then they could use the same techniques to build a machine to mimic any human behavior. Toward the end of the Sixties they realized that building a computer to play perfect chess gave you a computer that played perfect chess, and that's all. They had trouble using the same techniques for chess playing in other fields, athough concepts were gained that have been applied successfully in many AI applications. Also, no chess playing computer has ever been capable of consistently beating the masters of the game. Page 9 The 70s In the Seventies came the push to "try something practical" in Artificial Intelligence. The goal then became to define very limited domains that AI could be applied to TODAY to solve real world problems. This trend changed the course of AI and is the reason you hear so much about AI today. The two types of AI focused on were expert systems and natural language processors. In 1970 there were only 65,000 computers in the United States. In 1984 there were over 5 million. The rapid "computerization" of America has helped AI. In 1972, SHRDLU made headlines (at least among the AI researchers) by using semantic networks for natural language processing. SHRDLU was roughly dividable into three parts: the first part analyzed the text to get at the intent of the user's input, a semantic processor to get at the meaning of words, and a logic segment to implement the user's requests. SHRDLU functioned with a fairly limited domain: the blocks world. In SHRDLU's world there were only blocks, and the only thing SHRDLU could do was move these blocks around on a screen. The method behind this movement was what was unique. The first part of SHRDLU, now called an augmented transition network (ATN), where SHRDLU tried to solve for the intent of user's request, was unique. In 1975 MARGIE was created by Roger Schank, with the model of conceptual dependency (CD) in mind. In CD, the researcher is intent on using work done by linguists and psychologists to build human language understanding into machines. In MARGIE an input would be analyzed into the most minimal components, where it could be operated on. MARGIE had two main operating modes, in one it would paraphrase your input, for example: Bob asked Mary out. might become: Bob requested that Mary go on a date with him. And MARGIE's other mode, where it would make inferences concerning the input. Inferencing became one of the most important aspects of MARGIE, even though it was intended for natural language processing. In 1977 another breakthrough occurred in natural language processing: GUS. As natural language processors became larger and took on additional capabilities, the size of the Page 10 semantic network, the network that models human language, became extraordinarily large. In order to handle such large amounts of data a system would have to break the information up into digestable chunks. GUS demonstrated that you could break this data up and still be effective. GUS used a coding scheme called frames. Frames are used to group nodes in the semantic network into groups that are similar. GUS was also one of the first natural language systems to work against a data base; GUS was used as an advisor to passengers flying in California. The data base was a part of the Official Airline Guide and GUS answered questions against this data base. Most natural language processors sold commercially today are designed specifically to answer questions from an existing data base. Page 11 The 80s The Eighties have brought an explosion in the computer field and a corresponding explosion in Artificial Intelligence. This has occurred for three reasons: 1) there finally is enough computer power, and advanced software, for AI to be useful in real time, 2) there are plenty of computers and computer professionals to spend time and money accomplishing more than the simple computer tasks, and 3) industry has taken notice of AI and moved it from the laboratory into the field, along with additional funding. In 1983 another step up in natural language processing occurred with IPP. With IPP the frame used in other natural language processors became a dynamic scheme. Frames could be moved, deleted, changed, updated, and added with relative ease. This made creation and maintenance of the semantic network easier and quicker. IPP could build new structures if it encountered information that was new to it, and these new structures were fully compatible with, and enhanced, the old structures. Many other countries are involved in AI besides the United States. Much press has been devoted to the Japanese 5th Generation Computer, which is AI of a high form. However, many other countries, most of them western, are also involved in AI. Canada, for example, has a very impressive AI program, although most of its work is done in the university laboratory. We expect that Canadian work will soon leave the lab and advance into the marketplace and attract significant financing with it. In 1983 several interesting developments came out of Canadian laboratories concerning machine vision. Mackworth and Havens are working towards several schema for scene interpretation (putting into words what the camera sees), map understanding, and remote sensing. Other major work in Canada involves natural language processing, knowledge representation, and expert systems. The United Kingdom has been active in AI almost certainly from its inception. One fact that may be surprising is that Japan is the largest user of industrial robots in the world. Not of robots per person or per corporation, but Japan has more robots in employment than anywhere else in the world, including the United States. Japan uses well over 200,000 industrial Page 12 robots, and some estimates place the tiny Asian country as having as many robots in use as North America and Western Europe combined. In Japan the concentration, as far as robotics are concerned, is on sensing and control: they have successfully made a robot that can shake your hand firmly but gently. Britain, France, Germany, Italy, The Netherlands, Belgium, Sweden, and Spain all have active AI laboratories. The Artificial Intelligence Laboratory at Linkoping University, Sweden is concentrating on knowledge representation, problem solving, and natural language communication. The Kaiserslautern University, Germany, is working on the theory behind expert systems, and how to build them. Prolog, which the Japanese have taken as their language of choice for the fifth generation computer, was originally built in France. Prolog is a logic programming language, and was built by A. Colmerauer. Later, Prolog was enhanced by R. Kowalski of Britain. At the Louvain La Neuve, Belgium, techniques for knowledge base pruning have been developed. Since knowledge bases can become very large as information is added to them, several algorithms have been designed over the years to eliminate large sections of the knowledge base as the consultation proceeds. The problem with pruning is that you might miss some knowledge you need. In Belgium, they are working towards the best of both worlds. At the Research Institute of Applied Computer Science, Budapest, a computer language called Lobo has been developed. This language offers many of the advantages of Prolog, while keeping the advantages of a standard computer language, such as speed. At the Telecommunication Laboratory and Study Center, Turin, Prolog programs have been written to analyze the concurrent communications that occur in telephone operations. If you look at every AI system in existence today you might well exclaim that the humanoid robot of science fiction and 'Hal' of 2001 are just around the corner. There are systems that can perform very delicate sensor-motor tasks such as assembly of complex automobile structures - as long as the parts are all laid out in their correct positions, hold very impressive conversations, win nearly every time in certain games, advise doctors better than the doctors themselves, identify and select objects from a bin Page 13 based on "looking" at them - using three dimensions, and a host of other successful applications. However, we are a long way from building 'Hal' or a humanoid robot. In not one single area of AI have we even come close to approximating human behavior or capabilities. In some, heavily restricted domains, with well defined parameters, the AI system CAN occasionally surpass the human in accomplishing the same task. This is primarily for two reasons: 1) humans get bored. We become lax in the attention needed to accomplish a task, and AI systems never get bored. 2) AI systems also never forget, once they are taught to do a task, and if the environment in which that task is performed does not change, they will continue to perform that task correctly forever. A human might, over a period of time, forget how to do some part of a task. Another reason we are years from building "science fiction systems" is the problem of integration. We can build a system to "see" parts on a conveyor built, and a system to build automobile assemblies, but to build a system that can "see" to select parts then build an automobile assembly from them is another thing. What is needed is an influx of AI researchers and experts, willing to spend the time needed to tackle complex problems, and the creation of tools that are inexpensive yet fast and powerful. It is our hope that ESIE will make the road a little easier.