Blackboard system generator (BSG): an alternative distributed problem-solving paradigm

The classical blackboard model employs a number of relaxations of team decision theory that are commonly organized into three panels of AI heuristics, including: 1) a shared information panel that offers a capability for ensuring agent knowledge sharing, 2) a contract formalism for the agent and event scheduling, coordinating, and control panel, and 3) a blackboard panel for metalevel planning and guidance that offers whole situation recognition, top down reasoning, and adaptive learning. The nature and implications of these relaxations are explained in terms of the blackboard system generator (BSG) and via comparisons to what is done in other blackboard shells. Particular attention is paid to theoretical relaxations inherent in the classical blackboard model and to research opportunities arising as a result. Progress made to date to counteract adverse effects of some of these relaxations is described in terms of a project management/work breakdown paradigm adopted in BSG that: 1) alleviates the knowledge engineering bottlenecks of traditional blackboards and that provides BSG with a semantic rather than just syntactic understanding of blackboard control and scheduling; 2) allows a distributed problem-solving capability for connecting agents at virtual addresses on a logical network and that permits concurrent processing on any machine available on the network; 3) establishes an open architecture that includes techniques for integrating preexisting agent methods (e.g., expert systems, procedures, or data bases) while laying the foundation for assessing the impact of “black boxes” on the global and local objective functions; and 4) utilizes project management techniques for team agents planning as well as an analogical reasoner subsystem for BSG metaplanning and generic controlled learning. This latter item is supported by a connectionist scheme for its associative memory. The techniques of each of the three panels and of the four sets of paradigm-related advances are described along with selected results from classroom teaching experiments and from three applications using BSG to date. This journal article is available at ScholarlyCommons: http://repository.upenn.edu/ese_papers/197 334 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS, VOL. 19, NO, 2, MARCH/APRIL 1989 Blackboard System Generator (BSG) : An Alternative Distributed ProblemSolving Paradigm BARRY G. SILVERMAN, SENIOR MEMBER, IEEE, JOSEPH SHIH CHANG, AND KOSTAS FEGGOS Abstract -The classical blackboard model employs a number of relax-The classical blackboard model employs a number of relaxations of team decision theory that are commonly organized into three panels of AI heuristics, including: 1) a shared information panel that offers a capability for ensuring agent knowledge sharing, 2) a contract formalism for the agent and event scheduling, coordinating, and control panel, and 3) a blackboard panel for metalevel planning and guidance that offers whole situation recognition, top down reasoning, and adaptive learning. The nature and implications of these relaxations are explained in terms of the blackboard system generator (BSG) and via comparisons to what is done in other blackboard shells. Particular attention is paid to theoretical relaxations inherent in the classical blackboard model and to research opportunities arising as a result. Progress made to date to counteract adverse effects of some of these relaxations is described in terms of a project management/work breakdown paradigm adopted in BSG that: 1) alleviates the knowledge engineering bottlenecks of traditional blackboards and that provides BSG with a semantic rather than just syntactic understanding of blackboard control and scheduling; 2) allows a distributed problem-solving capability for connecting agents at virtual addresses on a logical network and that permits concurrent processing on any machine available on the network; 3) establishes an open architecture that includes techniques for integrating preexisting agent methods (e.g., expert systems, procedures, or data bases) while laying the foundation for assessing the impact of “black boxes’’ on the global and local objective functions; and 4) utilizes project management techniques for team agents planning as well as an analogical reasoner subsystem for BSG metaplanning and generic controlled learning. This latter item is supported by a connectionist scheme for its associative memory. The techniques of each of the three panels and of the four sets of paradigm-related advances are described along with selected results from classroom teaching experiments and from three applications using BSG to date. “Give me a fruitful error anytime, full of seeds, bursting with its own corrections, you can keep your sterile truth to yourse[f:” Comment on Kepler, Virfredo Pareto, circa 1900. Manuscript received June 15, 1987; revised January 28, 1988 and February 9, 1989. This work was supported in part by NASA JPL Small Business Innovative Research (SBIR), NASA GSFC SBIR, and the GSFC code 522 Contracts. B. G. Silverman is with the Institute for Artificial Intelligence, George Washington University, Washington, DC 20052 and IntelliTek, Inc., Rockville, MD 20852. J. Chang is with IntelliTek, Inc., Rockville, MD 20852. K. Feggos is with the Institute for Artificial Intelligence, George IEEE Log Number 8820137. Washington University, Washington, DC 20052.