“The Hearsay-II Speech-Understanding System: Integrating Knowledge to Resolve Uncertainty” was written by Lee D. Erman, Frederick Hayes-Roth, Victor R. Lesser, and D. Raj Reddy and published in ACM Computing Surveys, volume 12, number 2, in June 1980. Hearsay-II was built at Carnegie Mellon University as part of a major US program to make computers understand connected, spoken sentences, and the paper became the definitive description of an idea far bigger than speech: the blackboard architecture.
The problem Hearsay-II faced was that understanding speech requires many different kinds of knowledge at once, about acoustics, phonemes, words, syntax, and meaning, and each is uncertain on its own. The blackboard model handles this by giving each kind of knowledge its own independent module, a knowledge source, and a shared, structured workspace called the blackboard. Knowledge sources do not call each other directly. Instead each watches the blackboard and posts hypotheses when it can contribute, so a tentative word recognized acoustically can be confirmed or rejected by syntactic and semantic experts. A scheduler decides which knowledge source should act next, focusing attention where progress looks most promising. The metaphor is a group of specialists writing partial solutions on a common blackboard until a consistent interpretation emerges.
Blackboard systems became a standard way to organize complex AI applications that must combine many reasoning components, from signal interpretation to planning. The architecture’s clean separation between independent experts and a shared state remains influential in how multi-component intelligent systems are built.
Why business readers should care: the blackboard pattern, many specialized components contributing to a shared, evolving solution under a controller that decides what runs next, is an early blueprint for the multi-agent and tool-orchestration systems being built around today’s AI.