DARPA Strategic Computing, 1983: primary document

To meet the challenge of certain critical problems in defense, the Defense Advanced Research Projects Agency (DARPA) is initiating an important new program in Strategic Computing. By seizing an opportunity to leverage recent advances in artificial intelligence, computer science, and microelectronics, the Agency plans to create a new generation of "machine intelligence technology." This new technology will have unprecedented capabilities and promises to greatly increase our national security and our economic strength as it emerges during the coming decade.

THE CHALLENGE. Computers are increasingly employed in defense, and are relied on to help us hold the field against larger forces. But current computers, having inflexible program logic, are limited in their ability to adapt to unanticipated enemy behavior in the field. We are now challenged to produce adaptive, intelligent systems having capabilities far greater than current computers, for use in diverse applications including autonomous systems, personalized associates, and battle management systems.

THE OPPORTUNITY. Within the past few years, important advances have occurred in many separated areas of artificial intelligence, computer science, and microelectronics. Advances in "expert system" technology now enable the mechanization of the practical knowledge and the reasoning methods of human experts in many fields.

These separate advances can be jointly exploited to mechanize the thinking and reasoning processes of human experts into the form of powerful computing structures implemented in microelectronics, thus creating machine intelligence technology of unprecedented capabilities.

We now have a plan for action as we cross the threshold into a new generation of computing. It is a plan for creating a large array of machine intelligence technology that can be scaled and mixed in countless ways for diverse applications.

The timely, successful generation and application of intelligent computing technology will have profound effects. If technology is widely dispersed in applications throughout our society, Americans will have a significantly improved capability to handle complex tasks and to codify, mechanize, and propagate their knowledge. The technology will improve the capability of our industrial, military and political leaders to tap the nation's pool of knowledge and effectively manage large enterprises, even in times of great stress and change.

Successful achievement of the objectives of the Strategic Computing initiative will lead to deployment of a new generation of military systems containing machine intelligence technology. These systems will provide the United States with important new methods of defense against massed forces in the future, methods that can raise the threshold and diminish the likelihood of major conflict.

Autonomous systems, as used herein, are true robotic devices: they are able to sense and interpret their environment, to plan and reason using sensed and other data, to initiate actions to be taken, and to communicate with humans or other systems.

As an example, imagine a reconnaissance vehicle that could navigate up to 50 km cross-country from one designated position to another. It would be capable of planning an initial route from digital terrain data, updating its plan based on information derived from its sensors, resolving ambiguities between sensed and pre-stored terrain data, and incorporating landmark prediction and identification as a navigation means.

It is estimated that 1 trillion von Neumann equivalent computer operations per second are required to perform the vehicle vision task at a level that will satisfy the autonomous vehicle project's long-range objectives. At best, current machines of reasonable cost achieve processing rates below 100 million operations per second.

Pilots in combat are regularly overwhelmed by the quantity of incoming data and communications on which they must base life or death decisions. They can be equally overwhelmed by the dozens of switches, buttons, and knobs that cover their control handles demanding precise activation.

We have chosen to illustrate this concept by developing a personal associate within the context of the combat pilot. Called the Pilot's Associate, it is an intelligent system that assists the pilot in the air as well as on the ground, not replacing but complementing the pilot by off-loading lower-level chores and performing special functions so the pilot may focus his intellectual resources on tactical and strategic objectives.

The associate is personal to a specific pilot in that it is trained by that pilot to respond in certain ways and perform particular functions. For example, it might be instructed to automatically reconfigure the aircraft to a specific control sensitivity preferred by the pilot should the wing be damaged during combat.

A battle management system (BMS), as an example of a system to aid in the management of a large enterprise, would interact with the user at a high level through speech and natural language. It would be capable of comprehending uncertain data to produce forecasts of likely events, drawing on previous human and machine experience to generate potential courses of action, evaluating these options and explaining the supporting rationale for the evaluations to the decision maker.

It would generate hypotheses describing possible enemy intent, prioritize these according to their induced likelihood, and explain the reasons for the prioritization. Once the commander selects a course of action, the BMS would prepare and disseminate the operation plan, and compare the effects of option execution with those developed through the simulation both as a check on progress and as a means of identifying the need to replan.

It is estimated that in the aggregate the above functions define a distributed expert system requiring some 20,000 rules and processing speeds of 10 BIPs.