Strategic Computing: DARPA and the Quest for Machine Intelligence, 1983-1993 (review)

Although published by a different publisher and with a slightly different charter, this book may be read as a follow-on to the volume by Arthur Norberg and Judy O'Neill, Transforming Computer Technology: Information Processing for the Pentagon, 1962-1986 (Baltimore, Md.: Johns Hopkins University Press, 1996). That book chronicled the early support of computing science and technology by the Information Processing Techniques Office of the Defense Advanced Research Projects Agency (DARPA). It celebrated DARPA's success in advancing the state of the art of computing in the United States, and in doing so contributing to national security, even if the projects it supported did not always have "national security" as their explicit goal. Chief among those advances, of course, was DARPA's support of the concept of packet-switched computer networking, which led first of all to a network called ARPANET and then to its successor, the Internet.

Roland and Shiman's book likewise is the result of support and cooperation from DARPA, and, likewise, the authors were given access to internal documents and to key players in the story. But the two volumes are quite different. After being commissioned to write a history of DARPA's Strategic Computing Initiative (abbreviated here as SC) and promised access to materials, Roland and Shiman found themselves facing a stone wall when they requested certain documents, and even resorted to filing Freedom of Information Act requests for information from the very people who commissioned their study. In the Preface to the book Roland describes not only how he faced contentious negotiations with his patrons over the wording of the book contract; he also recounts how some of them were extremely dissatisfied with the early drafts of the study and hoped to have its publication stopped. For contractual reasons Shiman had to drop out of the project, and Roland finished it with other support.

Beginning with an account of the authors' troubles with DARPA is appropriate, as one of the book's central questions is whether the Strategic Computing Initiative was a "success" or not, and how difficult it was to come up with a definition of "success" that people could agree on. Historians of technology are quick to point out that they do not write the histories only of successes; furthermore, they argue that the careful historical analysis of a failure can tell us more about the nature of technological change than the study of a success. But DARPA did not want to commission a book that came to that conclusion, however well it might have been received by historians. One can find people who lament the development of the Internet, DARPA's earlier creation, as a "failure" of technology, but they are lonely people and it is hard to take their criticisms seriously. Not so with Strategic Computing. Critics abound. Part of the reason for that is in the way SC was introduced. DARPA did not set out to develop the Internet in the exact form that it eventually took; yet it can claim credit for setting in motion the forces that produced it. Strategic Computing, by contrast, listed a set of three specific, military systems that it would produce: an Autonomous Land Vehicle, a Pilot's Associate, and an Expert System to assist in Battle Management. None of the three were delivered, at least not in the exact form and on the exact date they were promised. No matter how much the SC program led to real advances—which Roland and Shiman document—the fact that the final "products" were not delivered as specified remains.

Strategic Computing advanced the state of the art in Expert Systems, computer vision, computer recognition and generation of speech, parallel-processing, and other base technologies that have found...