Blackboard Architecture

Origin. Developed at Stanford (HEARSAY-II) for speech recognition; Japanese AI research adopted and extended blackboard systems for multi-agent coordination, particularly in manufacturing and robotics at universities like Tokyo, Osaka, and the FGCP-affiliated institutes.

Mechanism. Multiple specialized knowledge sources (agents) collaborate by reading from and writing to a shared data structure (the blackboard). No knowledge source communicates directly with another; all communication is through the blackboard. A control component selects which knowledge source to activate based on the blackboard state. This enables opportunistic, data-driven problem solving.

Procedure. Define the blackboard structure: levels of abstraction, data types, regions. Implement knowledge sources as independent modules that watch for patterns on the blackboard and contribute when they can. Implement the control component that decides which knowledge source to activate (agenda-based or opportunistic). Initialize the blackboard with the problem. Knowledge sources fire, contribute, and enable other knowledge sources. The solution emerges from their collective contributions.

Applies to. Multi-sensor fusion, distributed problem solving, complex systems with heterogeneous expertise, real-time systems where information arrives asynchronously.

Limitations. Control is the hard problem. Without good control, knowledge sources thrash or starve. The blackboard can become a bottleneck in systems with many knowledge sources. Debugging is difficult because causation is indirect — effects on the blackboard emerge from multiple contributors. Japanese researchers found that careful engineering of the control component was essential but domain-specific.