I just saw this in the DARPA Strategic Plan for 2009:

Computer systems are essential to military logistics and planning, command and control, and battlefield operations. However, as computing systems have become pervasive in DoD, they have also become increasingly more complex, fragile, vulnerable to attack, and difficult to maintain. The computing challenges facing DoD in the future – autonomous platforms that behave reliably without constant human intervention, intelligence systems that effectively integrate and interpret massive sensor streams, and decision support systems that can adapt rapidly – will depend on creating more flexible, competent, and autonomous software.

DoD needs revolutionary new computer technology to overcome these challenges. DARPA has embarked on an ambitious mission to create a new generation of computing systems – cognitive computers – to dramatically reduce military manpower and extend the capabilities of commanders and warfighters. Cognitive computing systems can be thought of as systems that “know what they’re doing.” DARPA’s cognitive computing research is developing technologies that will enable computer systems to learn, reason and apply knowledge gained through experience, and respond intelligently to new and unforeseen events.

Success will have enormous benefits for our military. In the real-time environment of military operations, cognitive systems that can learn, reason, and draw on their experience to assist their users will make a huge difference. Cognitive systems will give military commanders and their staffs better access to a wide array of rapidly changing information, reduce the need for skilled computer system administrators, and dramatically reduce the cost of system maintenance.

For example, today’s computers handle low-level processing of large amounts of raw data and numeric computations extremely well. However, they perform poorly when trying to turn raw data into high-level actionable information because they lack the capabilities we call “reasoning,” “interpretation,” and “judgment.” Without learning through experience or instruction, our systems will remain manpower-intensive and prone to repeat mistakes, and their performance will not improve. The DoD needs computer systems that can behave like experienced executive assistants, while also retaining their ability to process data like today’s computational machines.

The Personalized Assistant that Learns (PAL) program has been developing integrated cognitive systems to act as personalized executive-style assistants to military commanders and decision-makers. PAL is creating a new generation of machine learning technology so information systems automatically adjust to new environments and new users, help commanders maintain the battle rhythm and adapt to new enemy tactics, evolving situations and priorities, and accelerate the incorporation of new personnel into command operations, while making more effective use of resources.

PAL technologies are being implemented in several U.S. defense information management and operations systems. PAL technologies are being used by the Army’s Command Post of the Future (CPOF) to amplify the capabilities of overworked combat command and control staffs. Working with CPOF, PAL learns significant battlefield activities, organizes and locates them on maps, and helps users collect information, plan, and execute operations. Evaluations at the Army Battle Command Battle Lab (BCBL) were highly successful. An officer returning from Iraq remarked, “PAL could be an incredibly powerful tool for Tactical Operations Center operations. It has the potential to save countless man-hours by performing routine, repetitive tasks… Those man-hours could then be reallocated to other tasks … or even free up Soldiers to conduct combat operations.” In a head-to-head evaluation at the BCBL, a PAL-enhanced CPOF prototype strongly outperformed the current, manually controlled CPOF.

Senior leadership uses the Strategic Knowledge Integration Web (SKIWeb) to share intelligence and to stay abreast of events unfolding throughout the world in real time. SKIWeb uses threaded discussions, or blogging, to share ideas, and encourages collaboration by providing up-to-the-minute situational awareness. PAL learning technology will help SKIWeb recognize and respond to critical event information. Outstanding performance has been demonstrated in an experiment with real SKIWeb data.

Inside command posts, operators perform many complex planning tasks using computers, including developing air tasking order and medical logistics planning. DARPA’s Integrated Learning program has demonstrated software that can learn these planning tasks by watching examples. Once the system learns a planning task, it can then support other operators who are perhaps less expert by guiding them through the task. This software will eventually make it practical to create many sophisticated decision support systems that will make our operators faster and more effective at their tasks and help keep our observe-orient-decide-act loops faster than the enemy’s.