Browse Books

Reviewer: Fred J. Damerau

This is a history of the Defense Advanced Research Products Agency (DARPA) Strategic Computing Initiative (SCI), which began with considerable fanfare in 1983 and quietly vanished from view in 1993, after an expenditure of approximately one billion dollars. SCI was a project to develop intelligent machines by advancing all of the required research fronts at once. Beyond the technology vision, the driving forces behind SCI were the Reagan era defense buildup, and the Japanese Fifth Generation computer program. Japanese manufacturers had already captured much of the market for memory chips, and now threatened to do the same for high-performance computing and artificial intelligence. The first three chapters of the book focus on the interplay of the people involved, as they tried to realize their vision, fund it, and manage it. Chapter 4 discusses the work on very large scale integration (VLSI), particularly the metal oxide semiconductor implementation service (MOSIS). The number of wafers processed by MOSIS (wafers have many chips on them) rose from 258 in 1981 to 1880 in 1989, implementing more than 12,000 projects. DARPA also funded the acquisition of computers specifically designed for the LISP programming language, the lingua franca of the artificial intelligence community at the time. These became obsolete with the development of high performance general-purpose workstations, which were faster, and could run LISP equally well. Chapter 5 discusses the related topic of machine architectures, particularly parallel machines, with a focus on the Connection Machine, from Thinking Machines Corporation. This was a massively parallel machine, originally intended as a back end to a LISP machine, to run semantic networks. A number of the machines were sold, but not enough to sustain the company, which filed for bankruptcy in 1994. Chapter 6 discusses artificial intelligence, and in particular expert systems, which at the beginning of the 1980s seemed to many to finally demonstrate the utility of artificial intelligence. DARPA’s focus was on building generic expert systems, which could be tailored by domain experts rather than knowledge engineers. The two main contractors for this project were IntelliCorp and Teknowledge, both of which had ties to Edward Feigenbaum of Stanford. Neither the Teknowledge project nor the IntelliCorp product ever became true generic tools, though both companies still survive as consulting businesses. The programs in speech recognition and natural language understanding were deemed successes, and DARPA has continued to support both technologies. Chapter 7 is largely a discussion of the travails of the autonomous land vehicle (ALV). This project involved numerous contractors, both universities and industrial companies. Eventually, the contractors realized that there would be no all-purpose vision system, but that perception had to be tailored to a specific circumstance. The DARPA program was canceled in 1988. Chapters 8 and 9 are largely concerned with the management personalities and the bureaucratic maneuvering that went on as the SCI began to wind down. In chapter 10, “Conclusion,” the authors make a number of points. Two important ones are “SC was managed erratically. The turnover in personnel was itself enough to disrupt the careful orchestration that its ambitious agenda required,” and “one can design and manage complex technical systems such as the Apollo spacecraft. The key is that all the components are in hand or within reach...Research and development works differently. The components want more than connection. They want invention...No rational plan can hope to connect them in advance. SC promised to do that and failed.” Although it is concerned with technology, this is not a technology book; it is much more a history book, and a sociology book. Because DARPA is a government agency, sensitive to the vagaries of politics and government spending priorities, it is hard to draw many general lessons from their experience, although the value of having realistic goals and a consistent management philosophy clearly comes through. The demise of SCI does not seem to have changed DARPA very much. Throughout the 1990s, DARPA continued to fund artificial intelligence research, using many of the same contractors that it had used since the 1960s. Some of the language programs originally supported by the Information Processing Technology Office (IPTO) are now under the Information Awareness Office, whose primary mission is intelligence directed toward terrorism. IPTO has itself been reconstituted, with much the same vision it had when originally founded by J.C.R. Licklider: to make computer systems able to reason, to learn, and to respond intelligently to novel situations. The difference is that no one has announced an expensive new program to accomplish these goals, many of which were outlined at the DARPATech 2002 Symposium (http://www.darpa.mil/DARPATech2002/presentation.html). I certainly recommend this book to anyone associated with the DARPA programs. They will recognize at least some of the players, and perhaps themselves. Beyond that, I found the details of how individuals and bureaucracies influenced the course of technological development to be very interesting. Knowing that it happens is not really the same as being able to follow the events in detail. Online Computing Reviews Service