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

The difference between a computer salesman and a used car salesman is that the used car salesman knows when he's lying. The Strategic Computing Initiative promised to revolutionize air, sea, and land warfare with artificial intelligence, leading to devices that would replace many human functions— if not humans themselves—in war. A billion dollars later, that hadn't happened, but the program did not die; like an old soldier, it just faded away.

Alex Roland, who could complement the high level of access to documents and support provided by a DARPA contract with the breadth of analysis and the depth of knowledge of a seasoned historian of technology, was the ideal candidate to write this follow-on study to Arthur Norberg's fine examination of the DARPA Information Processing Techniques Office, Transforming Computer Technology (1996). Unfortunately, the same thing happened to him and his coauthor, Philip Shiman, that happened to many other DARPA contractors in the program; the project was prematurely terminated, and the final product was not quite what its patrons expected. Although this is not unusual for historians who, in order to write history at all, must seek access and patronage from the very agencies they would like to study, Roland makes it clear from the outset that, once launched, the project remained in flight without visible means of support. We have the volume in hand as a result.

Coping deftly with complexities of bureaucracy, language, personalities, and technologies—a six-page list of acronyms provides the most indispensable tool for the reader—Roland and Shiman have managed to fill in the gaps in the material DARPA supplied with interviews and research to provide a coherent and readable account of a program the Department of Defense and the Congress would just as soon forget. Their analysis deploys the weapons of twenty-first-century history—systems analysis, social construction, technological trajectories—to penetrate the heart of DARPA's patronage of advanced research projects.

The story is one of humans rather than machines, despite the technological determinism of its subjects, and builds on an in-depth appraisal of the planners and promoters of strategic computing, from directors to program managers, to explain why some projects succeeded and others failed. The godfather of the effort, Robert Kahn, was, like many DARPA research managers, a researcher himself. Like many of them, his allegiance to the standards and methods of academic science was premised on a faith that good men with good equipment would produce good technology, and that this in turn would provide the impetus for innovation in military and industrial applications. Robert Cooper, director of DARPA, who brought with him from NASA a belief that applications would be stimulated by a goal (as, for example, the Apollo Project spawned the Saturn V and its payload), sought deliverable applications that would pull scientists in academia and industry to do the same for machine intelligence. If their strategies varied, they nevertheless created a program that appeared to use both while retaining an interpretive flexibility that permitted Congress and the Department of Defense to envision the project in terms of their own goals and lend it unprecedented support. Lynn Conway, who "rationalized" the two strategies at Cooper's behest, was among the early victims of the program's schizophrenia.

Of course, as Everett Dirksen is supposed to have said, "a million here and a million there, and pretty soon you're talking real money." The size of the program attracted close scrutiny within DARPA and without, premised on expectations that were generated in the sales process and easily disappointed when the system crashed. Roland uses a judicious selection of projects supported by the program to illustrate the conflict between "technology push" and "applications pull" that undermined the need for connection of the parts to a whole. He explains both...