Computers that reason and learn

T+D, Oct, 2003

It has been called the greatest learning tool since the printing press. The Internet, the foundation of the so-called new economy, began with the vision of scientists working for the Pentagon's Defense Advanced Research Projects Agency, its basic research think tank. So, what learning and training-related research is DARPA pursuing now? Quite a bit, actually.

According to new strategic plan issued earlier this year, the agency (US$2 billion annual budget) has a variety of activities up its sleeve, including a Training Superiority Program to transform military training. It is also embarking on a strategic thrust called "cognitive computing," a new generation of computer systems that know what they're doing. The goal of the initial three-year training superiority program is to "start a paradigm shift" in military training by providing continuously available, on-demand, mission-level training for all forces at all echelons.

"This goal will be attained by providing a new kind of cognitive training experience for units and individuals based on continuously available wars," says the plan. To reach that goal, the plan is to create scalable last-meter training systems that will train judgment and cognitive performance under stress and provide a Two-Sigma improvement over classroom training in student, user, and unit performance. It will also create an architecture to populate with and link stand-alone LMT systems. "This will combine real-time generation, maintenance, updating, and modification of the training war scenario," says a spokesperson. Ingredients of the system include ubiquitous PC-class computers, storage, graphics, and networks; simulation infrastructure and standards; progress in advanced distributed learning; massive multiplayer games; intelligent computer tutors; and standard ontology reusable-plugable domain-specific knowledge bases.

The cognitive computing initiative is aimed at furthering the vision of DARPA scientists of the 1960s and 1970s. J.C.R. Licklider envisioned computers and people working together symbiotically, with computers seamlessly adapting to people as partners. DARPA's information processing technology office sees it as a return to its roots to pursue Licklider's vision for computers that can reason and learn. It will focus on several core research areas: computational perception; representation and reasoning; learning, communications, and interaction; dynamic coordinated teams of cognitive systems; and robust software and hardware infrastructure for cognitive systems.

"If DARPA succeeds in this strategic thrust just now getting under way, then in another 10 to 20 years much of Licklider's vision may finally be realized, sparking a second powerful revolution in information technology," says the plan's principal author, program manager Ralph Chatham. Lastly, DARPA is also involved in research into the next generation of the Internet.