Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP Path: utzoo!watmath!clyde!floyd!vax135!cornell!uw-beaver!tektronix!hplabs!sri-unix!HADDAD@SU-SCORE.ARPA From: HADDAD@SU-SCORE.ARPA Newsgroups: net.ai Subject: fifth generation Message-ID: <526@sri-arpa.UUCP> Date: Wed, 21-Mar-84 23:47:07 EST Article-I.D.: sri-arpa.526 Posted: Wed Mar 21 23:47:07 1984 Date-Received: Mon, 9-Apr-84 05:27:41 EST Lines: 165 From: Ramsey Haddad [Forwarded from the Stanford bboard by Laws@SRI-AI.] For anyone interested in these things, there is a review by John McCarthy of Feigenbaum and McCorduck's "The Fifth Generation: Artificial Intelligence and Japan's Computer Challenge to the World" in the April 1984 issue of REASON magazine. [The following is a copy of Dr. McCarthy's text, reprinted with his permission. -- KIL] The Fifth Generation - Artificial Intelligence and Japan's Computer Challenge to the World - by Edward Feigenbaum and Pamela McCorduck, Addison-Wesley Publishing Co. Review of Feigenbaum and McCorduck - for Reason Japan has replaced the Soviet Union as the world's second place industrial power. (Look at the globe and be impressed). However, many people, Japanese included, consider that this success has relied too much on imported science and technology - too much for the respect of the rest of the world, too much for Japanese self-respect, and too much for the technological independence needed for Japan to continue to advance at previous rates. The Fifth Generation computer project is one Japanese attempt to break out of the habit of copying and generate Japan's own share of scientific and technological innovations. The idea is that the 1990s should see a new generation of computers based on "knowledge information processing" rather than "data processing". "Knowledge information processing" is a vague term that promises important advances in the direction of artificial intelligence but is noncommittal about specific performance. Edward Feigenbaum describes this project in The Fifth Generation - Artificial Intelligence and Japan's Computer Challenge to the World, predicts substantial success in meeting its goals, and argues that the U.S. will fall behind in computing unless we make a similar coherent effort. The Fifth Generation Project (ICOT) is the brainchild of Kazuhiro Fuchi of the Japanese government's Electro-Technical Laboratory. ICOT, while supported by industry and government, is an independent institution. Fuchi has borrowed about 40 engineers and computer scientists, all under 35, for periods of three years, from the leading Japanese computer companies. Thus the organization and management of the project is as innovative as one could ask. With only 40 people, the project is so far a tiny part of the total Japanese computer effort, but it is scheduled to grow in subsequent phases. The project is planned to take about 10 years,during which time participants will design computers based on "logic programming", an invention of Alain Colmerauer of the University of Marseilles in France and Robert Kowalski of Imperial College in London, and implemented in a computer programming language called Prolog. They want to use additional ideas of "dataflow" developed at M.I.T. and to make machines consisting of many procesors working in parallel. Some Japanese university scientists consider that the project still has too much tendency to look to the West for scientific ideas. Making parallel machines based on logic programming is a straightforward engineering task, and there is little doubt that this part of the project will succeed. The grander goal of shifting the center of gravity of computer use to the intelligent processing of knowledge is more doubtful as a 10 year effort. The level of intelligence to be achieved is ill-defined. The applications are also ill-defined. Some of the goals, such as common sense knowledge and reasoning ability, require fundamental scientific discoveries that cannot be scheduled in advance. My own scientific field is making computer programs with common sense, and when I visited ICOT, I asked who was working on the problem. It was disappointing to learn that the answer was "no-one". This is a subject to which the Japanese have made few contributions, and it probably isn't suited to people borrowed from computer companies for three years. Therefore, one can't be optimistic that this important part of the project goals will be achieved in the time set. The Fifth Generation Project was announced at a time when the Western industrial countries were ready for another bout of viewing with alarm; the journalists have tired of the "energy crisis" - not that it has been solved. Even apart from the recession, industrial productivity has stagnated; it has actually declined in industries heavily affected by environmental and safety innovations. Meanwhile Japan has taken the lead in automobile production and in some other industries. At the same time, artificial intelligence research was getting a new round of publicity that seems to go in a seven-year cycle. For a while every editor wants a story on Artificial Intelligence and the free lancers oblige, and then suddenly the editors get tired of it. This round of publicity has more new facts behind it than before, because expert systems are beginning to achieve practical results, i.e. results that companies will pay money for. Therefore, the Fifth Generation Project has received enormous publicity, and Western computer scientists have taken it as an occasion for spurring on their colleagues and their governments. Apocalyptic language is used that suggests that there is a battle to the death - only one computer industry can survive, theirs or ours. Either we solve all the problems of artificial intelligence right away or they walk all over us. Edward Feigenbaum is the leader of one of the major groups that has pioneered expert systems -- with programs applicable to chemistry and medicine. He is also one of the American computer scientists with extensive Japanese contacts and extensive interaction with the Fifth Generation Project. Pamela McCorduck is a science writer with a previous book, Machines Who Think, about the history of artificial intelligence research. The Fifth Generation contains much interesting description of the Japanese project and American work in related areas. However, Feigenbaum and McCorduck concentrate on two main points. First, knowledge engineering will dominate computing by the 1990s. Second, America is in deep trouble if we don't organize a systematic effort to compete with the Japanese in this area. While knowledge engineering will increase in importance, many of its goals will require fundamental scientific advances that cannot be scheduled to a fixed time frame. Unfortunately, even in the United States and Britain, the hope of quick applications has lured too many students away from basic research. Moreover, our industrial system has serious weaknesses, some of which the Japanese have avoided. For example, if we were to match their 40 engineer project according to output of our educational system, our project would have 20 engineers and 20 lawyers. The authors are properly cautious about what kind of an American project is called for. It simply cannot be an Apollo-style project, because that depended on having a rather precise plan in the beginning that could see all the way to the end and did not depend on new scientific discoveries. Activities that were part of the plan were pushed, and everything that was not part of it was ruthlessly trimmed. This would be disastrous when it is impossible to predict what research will be relevant to the goal. Moreover, if it is correct that good new ideas are more likely to be decisive in this field at this time than systematic work on existing ideas, we will make the most progress if there is money to support unsolicited proposals. The researcher should propose goals and the funders should decide how he and his project compare with the competition. A unified government-initiated plan imposed on industry has great potential for disaster. The group with the best political skills might get their ideas adopted. We should remember that present day integrated circuits are based on an approach rejected for government support in 1960. Until recently, the federal government has provided virtually the only source of funding for basic research in computer technology. However, the establishment of industry-supported basic research through consortia like the Microelectronics and Computer Technology Corporation (MCC), set up in Austin, Texas under the leadership of Admiral Bobby Inman, represents a welcome trend--one that enhances the chances of making the innovations required.