PORTLAND, Ore. SRI International recently has won the role of system integrator for a new approach to AI funded by the Defense Advanced Research Project Agency (Darpa).
Called Bootstrapped Learning, the initial phase begins immediately, funded by a $10 million, 15-month plan to develop the cornerstone for the program: a learning system called Phased Learning through Analyzing, Teaching and Observation (Plato). If all phases of the development program are completed, up to $27 million will be invested in the program by Darpa over the next 3 1/4 years.
Bootstrapped Leaning scraps the unachievable goals originally set for AI in favor of a feasible one: learning a task from a human instructor. Instead of trying to fulfill the AI objective of encapsulating knowledge in an "expert system," the program models an "electronic student," learning tasks from a human instructor. In the examples given by Darpa while soliciting proposals, a human teacher teaches an electronic student to fly an unmanned aerial vehicle (UAV).
SRI will craft Plato as the core engine managing all algorithms stemming from the multi-organization effort funded by Darpa. Plato will integrate all Bootstrap Learning functions, enabling new learning modules to be explored. Plato will incorporate the efforts of Boeing, Stony Brook University, University of Illinois at Urbana-Champaign, University of Maryland-Baltimore County, University of Massachusetts, University of Southern California, University of Texas at Austin, University of Washington and University of Wisconsin-Madison.
In addition to UAV flight control applications, SRI will seek to create test environments for space station diagnosis, Robocup soccer and battlefield control. SRI also heads technology integration initiatives for Darpa's PAL (Personalized Assistant that Learns), GALE (Global Autonomous Language Exploitation), and AURA (Automated User-centered Reasoning and Acquisition) programs.