Fifth Generation Systems: East Meets West in Battle for Info Supremacy.
For the moment, the United States is ahead in the two areas critical to this latter-day revolution: exploitation of the semiconductor chip and development of a superior communications infrastructure. However, Japan and Europe appear determined to catch up and perhaps overtake the United Sttaes in both areas. European nations were slow to focus on the importance of communications, but are now making up for lost time by pushing through plans for an integrated services digital network. Japan has been a leader in communications technology for some time, and has recently moved ahead of the United States in a vital segment of the semiconductor industry, capturing over 70 per cent of the market for 64K RAM chips.
Now, the Japanese have set out to leapfrog USA computer technology by targeting the development of a "fifth-generation" system by 1990. In an effort reminiscent of the United States space program, Japan has gathered the youngest and brightest under a charismatic leader and backed the enterprise with considerable resources. The strategy is spelled out in an impressive national plan of the Ministry of International Trade and Indistry (Miti), called Fifth Generation Computer Systems. The plan documents a carefully staged ten-year research and development program on Knowledge Information Processing Systems (Kips) . . . artificially intelligent machines that can reason, draw conclusions, make judgments and even understand the written and spoken word . . . with initial funding and new laboratories in Tokyo provided by the Japanese government.
This ambitious undertaking is expected to cost as much as $1 billion over ten years. Compared with IBM's annual R&D budget of over $1.5 billion, the Japanese outlay may not sound that impressive, until one remembers that the money is for one project and a highly innovative one at that. However, the rewards are equally grand. If the plan is successful, Japan could not only dominate the traditional forms of the computer industry, but establish a "knowledge" industry in which knowledge itself would be a salable commodity.
In their book "The Fifth Generation: Artificial Intelligence and Japan's Computer Challenge to the World," Edward Feigenbaum and Pamela McCorduck claim that the wealth of nations, which depended upon land, labor and capital during its agricultural and industrial phases, will come in the future to depend upon information, knowledge and intelligence.
"That isn't to say that traditional forms of wealth will be unimportant," Feigenbaum and McCorduck state. "Humans must eat, and they use up energy, and the like manufactured goods. But in the control of all these processes will reside a new form of power, which will consist of facts, skills, codified experience, and large amounts of easily obtained data, all accessible in fact, powerful ways to anybody who wants it . . . scholar, manager, policy maker, professional or ordinary citizen. And it will be for sale."
Feigenbaum and McCorduck believe the Fifth Generation will be more than a technological breakthrough. "The Japanese expect these machines to change their lives . . . and everyone else's," they say. "Intelligent machines are not only to make Japans's society a better, richer one by the 1990s, but they are explicitly planned to be influential in other areas, such as managing energy or helping deal with the problems of an aging society." In addition, the new machines will help to increase productivity in the primary industries such as agriculture and fishing, as well as tertiary industries, such as services, design and general management, where productivity improvements have been difficult to achieve.
On a global scale, the Fifth Generation project, even if only partially successful, could vault Japan into a leadership position in the world's information processing business.
Recognizing this, individual European nations such as Britain and France have embarked on similar government-subsidized projects, while the European Economic Community has funded its own cooperative program. In the United States, the challenge is being mst by the government in the form of the Pentagon's Defense Advanced Research Projects Agency, as well as by industry. One of the more unusual efforts is a joint R&D venture, Microelectronics and Computer Technology Corporation, created an d funded privately by leading United States firms to help maintain United States technological preeminence and international competitiveness in microelectronics and computers. The firm's most complex and ambitious project is a ten-year effort on fifth generations systems. Going Beyond von Neumann
The transition from information processing top "knowledge" processing . . . from computers that calculate and store data to computers that reason and inform . . . will require a significant departure from conventional computer designs. The first four generations of computers . . . vacuum tube, transistorized, integrated circuit and very-large-scale integrated circuit (VLSI) computers . . . all followed the same general architecture, based around a central processor, memory, arithmetic unit and input/output devices. Known as von Neumann machines, after the computer pioneer and mathematician John von Neumann, these computers operate principally in serial fashion, step by step. In contrast, the fifth generation systems will use new parallel architectures (known collectively as non-von Neumann architectures), as well as new memory organizations, new programming languages, and new operations wired in for hanlding symbols and not just numbers.
With the von Neumann architecture, the processor-memory bottleneck ultimately creates a traffic jam that limits the speeds computers can attain even with the fastest microelectrics circuits. For the fifth generation systems, the Japanese are considering a "data flow" computer championed by Jack Dennis at the Massachusetts Institute of Technology. Data flow computers have a large number of processors, each with their own memory, and a routing network so that the processors and memories can communicate with each other and execute insturctions simultaneously.
The Japanese are also aiming for chips with ten million transistors, in contrast to today's limit of a few hundred thousand transistors. Such processors are being developed in the course of another Miti effort, the SuperSpeed Computing Project, and will be adapted into the fifth generation machines. In addition, the Fifth Generation depends on access to "knowledge" bases in many locations, so its technology will ultimately be fused with the most advanced communications technologies the Japanese can design.
Perhaps the biggest challenge the Japanese face is in the field of artificial intelligence, or AI, a discipline which focuses on machines that can deduce, infer and learn. In contrast to ordinary computers, which must be programmed to perform every step involved in solving a problem, AI machines can figure out how to solve problems on their own. They need only know what the problem is, and they can find that out by asking the user questions in languages that resemble human ones, in contrast to the arcane programming languages needed by conventional computers.
Artificial intelligence finds application in knowledge-based, or "expert" systems, which use AI methods to solve problems and to aid decision making by using a knowledge base along with rules of inference that apply to the specific field of knowledge. These expert systems not only replicate and multiply the value of human expertise, but also capture it and perpetuate it in computerized form, amking it possible to pass on knowledge and experience from generation to generation. Other AI work has given computers the ability to understand natural languages such as English, making it easy for computer novices to use the machines effectively and to develop new computer applications without programming. Evolution of Expert Systems
AI research dates from 1956 when the expression was coined by John McCarthy, then assistant professor of mathematics at Dartmouth College in Hanover, New Hampshire. Early research looked for general problem-solving solutions, which turned out to be overwhelming in scope. For instance, if a chess-playing program used AI techniques to examine every possible move before deciding on the best one, it would have to look at 10.sup.120 moves for each game. Obviously, the program would run faster if it could decide which of the possible moves would be considered "good" in a given context. When playing chess, or dealing with business problems, people use their accumulated knowledge of the world and their particular experience to solve problems rather than trying all possible alternatives. AI researchers came to realize that computers could be programmed to do the same and this gave birth to expert systems.
Today, expert systems are used as diagnostic aids in medicine, as planning tools for manufacturing and as decision support systems for dealing with oil drilling problems. According to Feigenbaum and McCorduck, knowledge is the key factor in the performance of such systems. They have impressive credentials to discuss the subject. Feigenbaum, a professor of computer sciences at Stanford University is a recognized pioneer in artificial intelligence; McCorduck, a science writer who teaches at Columbia University, has been interested in AI for two decades and recently authored a history of artificial intelligence called "Machines Who Think."
They explain that the knowledge needed in an expert system is of two types: the first comprises the widely shared knowledge that is written in text books and journals; the second is the knowledge of good practice and good judgment acquired by human experts over years of work. In addition to this knowledge, an expert system needs an "inference procedure," a method of reasoning used to understand and act upon the combination of knowledge and problem data.
Feigenbaum and McCorduck explain that the knowledge in the knowledge base must be represented in symbolic form and in memory structures that can be used efficiently by the problem-solving and inference subsystem (see figure). This representation can take many forms. One of the most common is the object, a cluster of attributes that describe a thing. Another common representation is the rule, which consists of a collection of statements, called the "if" part, and a conclusion or action to be taken, called the "then" part. To find out if a rule is relevant to the reasoning task at hand, the problem-solving program must scan over the store of "ifs" in the knowledge base.
In the Fifth Generation plan, knowledge will be stored electronically in a large file known as a relational data base. The job of automatically updating the knowledge in the file and in organizing appropriate searches for relevant knowledge will be performed by the knowledge-base management software. The interaction between the hardware file and the software file manager will be handled by a logical language called a relational algebra.
The Fifth Generation prototype knowledge-base subsystem will handle thousands of rules and thou sands of objects. Each object will be allotted a thousand characters of file storage space. Within the ten-year plan, the Japanese goal is to develop knowledge-base capacity that will handle tens of thousands of inference rules and 100 million objects . . . enough to store the entire Encyclopedia Britannica. Japanese planners want their machines to handle millions of logical inferences per second (Lips), where one logical inference is an "if/then" sequence of reasoning. The Japanese have also chosen Prolog as the language of interaction between the logic processing hardware and the software that implements the various problem-solving strategies. Milestones in Japan's Agenda
The initial milestone in the Fifth Generation plan is a single-user Prolog workstation, called a personal sequential-inference computer, which will be capable of performing one million Lips. Feigenbaum and McCorduck explain that it is intended to be both a prototype for later development, as well as an intermediate product that may be on the market by 1985. "This prototype would give an order of magnitude improvement over software-based Prolog implementations on today's common mainframe computers," they say. The final target for the subsystem is an inference supercomputer capable of performing 100 million to 1 billion Lips, a speed that Feigenbaum and McCorduck say can only be achieved by the "insightful use of a great deal of parallel processing in the computer hardware."
The work will be done by research teams at the Institute for New Generation Computer Technology (ICOT), supplemented by contract work done under ICOT's direction. ICOT was formed as an "instant" institute in april 1982 and staffed by 40 researchers under director Dr. Kazuhiro Fuchi, former head of the Information Sciences Division of Miti's Electrotechnical Laboratory and the main architect of the Fifth Generation project. Fuchi assembled the 40 researchers within two weeks of the start of the project, causing a controversy in the seniority-conscious country by demanding that everyone be under 35. Fuchi, who is in his mid-40s, justified his action by arguing that revolutions aren't made by the elderly.
The young researchers come from the eight firms backing ICOT . . . Fujitsu, Hitachi, Nippon Electric Corporation, Mitsubishi, Matsushita, Oki, Sharp, and Toshiba . . . and the two national laboratories that are also participating, the government-owned Nippon Telephone and Telegraph's Musashino Laboratories and Miti's own Electrotechnical Laboratory. After three or four yeas, the researchers will be rotated out of ICOT back to their company laboratories. Meanwhile, there are no proprietary considerations to limit collaboration among them while they are at ICOT. In addition, the researchers are routinely sent back to their firms to report on progress.
The rotation and the routine reports are intended to seed ideas throughout the participating firms systematically. "Such cooperation might agitate a Washington antitrust regulator were it to happen in the United States," note Feigenbaum and McCorduck, "but ICOT's mission is to foster such cooperation and to educate industrial scientists actively by joint project work."
Miti's announced commitment of $450 million over the ten-year period is spread rather lightly over the first three-year phase ($45 million) and then budgeted heavily for the years of expensive development engineering. The first phase will be funded fully by Miti. In the second and third phases, Miti expects the funding will be matched by the participating companies, bringing the total project budget to about $850 million. Feigenbaum and McCorduck note that other Miti-initiated national projects have seen higher ratios of industry-to-government spending, sometimes two or three to one. "It's very possible that if the project is meeting its intermediate targets at the end of the first phase, and if the Japanese economy is strong, the total budget could well escalate to more than $1 billion," the authors state.
The Fifth Generation project is structured over a ten-year period. The first three-year phase is intended for building the research teams and laboratories, learning the state-of-the-art, forming the concepts that will be needed in the later work, and building hardware and software tools for the later phases. The personal sequential-inference (PSI) computer is one of these tools. The workstation will be a prototype of later machines, as will be its problem-solving software. Early expert system prototype applications will also be written, according to Feigenbaum and McCorduck.
The second phase of four years is one of engineering experimentation, prototyping, continuing experiments with significant applications, and initial experiments at systems integration. The first thrust at the major problem of parallel processing will be done in these years. The final phase of three years will be devoted to advanced engineering, building final major engineering prototypes, and further systems integration work. During this phase, the R&D results will be distilled into a set of production specifications for the commercial products that are to be marketed by the participating companies. Making Machines Man-Like
One of the major undertakings will be the development of intelligent interfaces . . . the ability the machines will have to listen, see, understand and reply to human users . . . which will require extensive research and development in natural language processing, speech understanding and graphics and image understanding.
Because computer novices will be the largest groups of users, natural language processing is one of the Fifth Generation project's most important research goals. Natural language systems shift the burden of understanding from the user onto the machine; the natural language system must understand the idiosyncrasies of the user and his language rather than forcing the user to understand the idiosyncrasies of the computer. Such a system must accept and answer requests expressed in the user's natural language, phrased any way the user chooses to express himself. If any ambiguities exist in the request, a natural language system must deduce them and ask the user for clarification.
Natural language processing will also be put to use in the development of a highly ambitious machine translation program, initially between English and Japanese, with a vocabulary of 100,000 words, Feigenbaum and McCorduck report. The goal is 90 per cent accuracy, with the remaining 10 per cent to be processed by the user. Research is natural language processing will proceed in three stages, beginning with an experimental system, followed by a pilot model implementation stage that is connected with the inference and knowledge base machines, and concluding with prototype implementations. At that point, the machines will be expected to understand continuous human speech with a vocabulary of 50,000 words and 95 per cent accuracy with a few hundred or more speakers. The speech understanding system is also expected to be capable of running a voice-activated typewriter and of conducting a dialogue with users by means of synthesized speed in Japanese or English.
Picture and image processing are considered almost as important as language processing, according to Feigenbaum and McCorduck, especially as they contribute to computer-aided design and manufacture, and to the effective analysis of aerial and satellite images, medical images and the like. Here again the research will take place in three phases, beginning with an experimental phase, followed by the introduction of a pilot model and finally the implementation of the prototype and its integration into the Fifth Generation machine. One potential application is in robotics, where the goal would be to construct robots that can see, understand and act under differing circumstances.
The Japanese surprised many industry observers by selecting Prolog (for "programming in logic") as the language of interaction between the logic processing hardware and software that implements the various problem-solving strategies . . . effectively the machine language of the logic processor. Developed in Europe in the late 1970s, Prolog is used in AI work because its basic terms express logical relationships among objects, and not just equations as most programming languages do. However, few of the intelligent systems now in use rely on Prolog; most of them have been programmed in Lisp (for list processing), an older language which software engineers have developed more fully.
Prolog is derived from formal deductive logic, whereas Lisp comprises sets of equations through which a required function can be defined in terms of simpler, more primitive functions. The idea is to decompose the original statement into simpler tasks that can be easily executed in parallel by the computer. In fact, both languages are well-suited to parallel processing machines.
While development of Prolog machines has been slow, there are a number of computers and personal workstations to run Lisp, including machines from Digital Equipment Corp., Fujitsu, Lisp Machines, Incorporated, Symbolics, Incorporated, and Xerox Corporation, Apollo Computer, Incorporated recently extended its programming language base with Domain Lisp, allowing AI applications to run on its workstations.
Among natural-language interfaces, the best known is Intellect from Artificial Intelligence Corporation of Waltham, Massachusetts. Intellect gained a major milestone for AI technology last summer with its acceptance by IBM Corporation. Serving at present as an interface to IBM data base products, Intellect could become the primary interface to a host of software services ranging from communications to modeling to spreadsheets, allowing relatively inexperienced users to create sophisticated data base queries and other applications. Admiral Leads America's Team
Within the United States, the most concerted R&D effort in AI comes from Microelectronics & Computer Technology Corporation (MCC), based in Austin, Texas. MCC was created explicitly as a United States response to Japan's Fifth Generation project. Concerned about the Japanese challenge, William C. Norris, chairman of Control Data Corporation, convened a meeting of top computer and microelectronics industry executives in Orlando, florida in April, 1982. From the meeting came a decision to create a joint-venture corporation to fund research and also contribute researchers. Task forces then thrashed out a research agenda focusing on four programs . . . computer-aided design and manufacturing, software technology, packaging and advanced computer architectures.
The venture partly follows the Japanese model: the companies will donate scientists and researchers to MCC, loaning them for a minimum of three years. However, whereas Miti is helping to finance the Japanese effort, MCC will rely entire on private funding. Each shareholder company in MCC can invest money and personnel in any or all of the programs. In return, a company that funds research gets three years lead in being licensed at no cost to develop its own products and package them for the marketplace. After that, any company, member or not, can be licensed.
MCC is now owned by 15 American microelectronics and computer companies: Advanced Micro Devices, Allied, BMC Industries, Control Data, Digital Equipment, Harris, Honeywell, Martin Marietta Aerospace, Mostek, Motorola, National Semiconductor, NCR, RCA, Rockwell and Sperry. Despite initial antitrust fears, the Department of Justice has announced that it does not object to the creation of MCC. However, the Department will be looking at each of the four programs as they develop to see if the combination of firms cooperating in them could constitute a breach of antitrust laws.
The shortest program, dealing with the packaging of integrated circuits, will be funded at about $5 million a year for six years. MCC also plans to spend about $8 million annually on a seven-year program to develop new techniques, procedures and tools for improving the productivity of the software development process by one or two orders of magnitude. A further $11 million a year is targeted for an eight-year program to develop new computer-aided design and manufacturing tools for the production of tomorrow's VLSI chips.
The largest, most expensive and longest running program deals with advanced computer architectures and is the MCC equivalent to Japan's Fifth Generation program. Over its ten-year lifetime, the program will be funded at about $15 million annually for work in four main areas:
Parallel Processing--to develop the languages and architectures to allow computers to perform tasks simultaneously instead of sequentially, with corresponding increases in processing speed.
Data base system management--to improve data base design and storage methods and capacities to permit flexible storage and faster retrieval of a broader range of more complex information.
Human factors technology--to improve the relationship between man and computer by simplifying the use of computers through techniques such as improved voice or character recognition or use of natural languages.
Knowledge-based systems--to realize the computer's problem-solving potential by developing new ways to represent human knowledge and thought concepts, as well as new engineering models and tools to apply human expertise to a wide range of problems.
To run the corporation, MCC's directors chose retired Admiral Bobby Ray Inman, former director of the National Security Agency and former deputy director of the CIA. Inman believes the first three years of the program will be spent "squeezing out what we really know in all [four] areas . . . where we need additional basic research and where to press on with advanced research." The hope is that VLSI technology will have advanced to a stage by that time where it can be applied across all four areas. "Defining what [VLSI technology] will do for us, and what we decide on doing, will be a major effort that's probably going to take two or three yeas," Inman says. "So we'll probably be about the six-year point before we start dealing with the task of designs we want for computer architectures."
By next year, Inman envisions a staff of about 250 and a budget of about $75 million a year. By then he also expects to have detailed milestones for all four projects, though he has yet to decide whether to announce them or keep them proprietary.
More than any other single agency in the world, the Pentagon's Defense Advanced Research Projects Agency (DARPA) is responsible for the current state-of-the-art in AI. When no USA corporation or foundation chose to take AI seriously, or could afford to, DARPA supported it through two decades of vital but highly risk research. A 1981 study by the Defense Science Board ranked AI in the top ten military technologies for the 1980s. Today, AI work is funded through the agency's four-year, $600 million Strategic Computing Program, intended to put artificial intelligence into military equipment.
Among the systems under development: an "autonomous land vehicle," which will move around battlefields on legs, guided by a vision system and a supercomputer to help it distinguish among objects cluttering its path; a sophisticated command-control-and-communications battlefield management system that will utilize AI technology to predict battle scenarios and suggest appropriate strategies: and a pilot's associate to help identify incoming hostile targets or equipment malfunctions and to report them in a synthesized voice. Europe Adds Spirit to Contest
Western Europe has several projects in various stages of implementation and planning. In the United Kingdom, a fifth-generation computer project called Alvey is underway based on the Japanese model with the British government committed to funding more than half of the five-year $550 million budget. A collaborative effort on the part of government, industry, academia and other research organizations, the Alvey program will focus on four research areas: knowledge-based systems, software engineering, man-machine interface and VLSI technology.
In addition, government and industry are collaborating on a research institute devoted to artificial intelligence. Called the Turing Institute, in honor of British mathematician and computer theoretician Alan Turing, it will concentrate on fundamental research in computer architectures, automatic programming, knowledge-based systems and advanced robotics. Set up in collaboration with the University of Strathclyde, the institute has a number of industrial sponsors, including ICL, Sinclair Research, two Shell Oil Research laboratories and two government agencies.
France has also been paying close attention to the Japanese Fifth Generation Project. Inria, the French national information sciences laboratory, formed a group of scientists and industrialists from both the public and private sectors to plan a French response to the Japanese challenge. Its overall thrust is a national effort to design and manufacture software and hardware to compete with Japan's knowledge-based systems. Meanwhile, Schlumberger, the French oil field instrumentation specialist, considers artificial intelligence important enough to have established its own AI group.
The European Economic Community (EEC) als has an R&D plan, dubbed Esprit, for European Strategic Program for Research in Information Technology. A $1.2 billion joint venture among the ten EEC countries, Esprit will focus on three technologies . . . knowledge-based systems, software and microelectronics . . . as well as two applications areas . . . office automation and computer-integrated manufacturing. Espirit was set up with the help of the leading European information processing companies, and all projects require transnational cooperation between researchers from more than one EEC country.
European Commission vice president Viscount Etienne Davignon, architect of the Esprit program, managed to get the European Commission, European Parliament and national governments to agree to the program in record time . . . within a few months. However, although the level of funding was also agreed in principle, the go-ahead was not given because Esprit became entangled in the wider European community budgetary crisis. Neither the German nor the United Kingdom governments would agree to give long-term financial assurances to the Esprit program until the long-term financial future of the community funding was agreed. In February, however, the British and German governments finally gave the financial go-ahead for the first five-year phase of the ten-year program, with the $1.2 billion budget to be financed on an equal basis by the EEC and by industry.
To support the Esprit program, the EEC is establishing an advanced communications network, called the Esprit Information Exchange System. A team of major vendors is developing software to allow the Esprit participants not only to send each other files, documents, messages and software, but also to use each other's computer systems. Britain's ICL and GEC, West Germany's Siemens, France's Compagnie Des Machines Bull and Italy's Olivetti won the three-year contract to develop the network, whose aim is to bridge the incompatibility that exists between various machines used by Esprit participants and to provide a manufacturer-independent baseline for other Esprit developments. Various universities and research centers will also help to develop the network.
Meanwhile, not content with Esprit and the European national fifth-generation projects, Bull, ICL and Siemens, Europe's three lagest computer companies, established a joint research institute last September in southern Bavaria. The intitute is owned and financed equally by the three companies. Research will center on knowledge processing.
Britain's Alvey program is designed to complement rather than compete with Esprit. A direct response to the Japanese effort, the program is Britain's first large-scale collaborative R&D project between government and industry and represents a doubling of Britain's research efforts in information technology.
It began when a British delegation attended the Tokyo conference at which Japan's Fifth Generation program was unveiled. Spurred by the delegation's report, the British government immediately commissioned a working party under John Alvey, technical director of British Telecom, to advise on the scope of a similar collaborative R&D program. The result was the so-called Alvey program, a five-year plan to coordinate government, industry and university research nationwide. All industry work will be 50 per cent government-funded, while university research will get 100 per cent aid. To ensure that basic research leads to an end product, the program will identify a number of "demonstration projects" covering applications in industry, defense, medicine and the social services. To coordinate the program, the Alvey Directorate has been established within Britain's Department of Trade and Industry with Brian Oakley serving as program director. USA Choices for the 1990's
Faced with the government-backed Japanese and European challenge, what should be the United States response? In their book, Feigenbaum and McCorduck list various possibilities, such as joining with Japan, forming industrial R&D consortia protected from antitrust regulations, or relinquishing the hardware effort and concentrating solely on software--akin to the razor blade company that gives away razors because profits are in the blades. But what they would really like to see is a national center for knowledge technology. "It might be a mega-institute, like Los Alamos, embracing all forms of knowledge technology," they say. "Or it might be a smaller multiple-university-run laboratory, such as Brookhaven and Fermilab in physics." Whatever form it takes, the national laboratory should be newly created.
"We cannot look to the existing national laboratories for the kind of innovations a knowledge technology laboratory must produce, freighted as they are with tradition, stodginess and bureaucracy," they say. "Those three horsemen of the intellectual apocalypse will eventually come to the new laboratory, but while it is still new, it has at least a fighting chance to achieve brilliance."
Without some kind of plan, Feigenbaum and McCorduck conclude the United States should prepare "to become the first great agrarian post-industrial society."
|Printer friendly Cite/link Email Feedback|
|Date:||Jul 1, 1984|
|Previous Article:||Data-Base Access Made As Easy As 'Open Sesame' for Pacific Bell's Customer Service Representatives.|
|Next Article:||Digital PBXs Playing Key Role in Today's Info Management, High-Speed Data Needs.|