Portland, Ore. The Pentagon is funding artificial intelligence research under a $29 million program called the Perceptive Assistant that Learns (PAL) program.
On Wednesday (July 16), the Defense Advanced Research Projects Agency (Darpa) said it would fund research on cognitive computing by more than 22 companies.
The five-year PAL program aims to spawn AIs that attempt to encapsulate the knowledge that an administrative assistant might amass while helping an executive. Knowledge would be gathered by learning rather than using a historically-brittle knowledge base past AI efforts have failed because their knowledge bases could not deal with novelty, and were therefore scorned as "brittle".
By learning to help around the office by observing and interacting with office workers, PAL aims to automatically configure itself to the user. DARPA hopes PALs will be crafted for military, business and academic
users where decision-makers need help managing multiple,
simultaneous tasks and unexpected obstacles.
"PAL isn't as ambitious as a true complete autonomous thinking
artificial intelligence, but nevertheless is an ambitious project and we don't know whether we will succeed...its a high risk project, but if we win, we win big," said director Ronald Brachman of Darpa's Information Processing Technology Office (IPTO).
IPTO awarded two primary research contracts for PAL. Carnegie
Mellon University's School of Computer Science (Pittsburgh) received a $7 million contract and $22 million to SRI International (Menlo Park, Calif.) received a $22 million award to be divided among 20 subcontractors, including top U.S. universities and Boeing Phantom Works.
Darpa hopes the research will lead to systems that can reason, learn from experience, take advice and respond intelligently to new situations never, according to Brachman.The agency hopes
PAL will also strengthen the computing infrastructure, allowing
cognitive computers to handle their own maintenance, help ward off security attacks, manage their own internal resources dynamically and reduce development and deployment time by assisting in their own debugging.