Blackboard Model Description by Lee Erman, Richard Hayes-Roth, Victor Lesser and D. Raj Reddy
In May 1980, Lee Erman, Richard Hayes-Roth, Victor Lesser and D. Raj Reddy published 'The Hearsay-II Speech-Understanding System: Integrating Knowledge to Resolve Uncertainty' in Artificial Intelligence, vol. 14, providing the canonical description of the blackboard model as a structured framework for cooperative problem-solving among independent knowledge sources.
Background
By the late 1970s, researchers trying to build speech-understanding systems had a coordination problem. A system that recognises spoken language cannot rely on acoustics alone. It needs to reason simultaneously about sounds, syllables, words, grammar and context, because each level of analysis can help correct errors at the others. If the acoustic signal is ambiguous, knowing that a particular word makes grammatical sense in that position can tip the balance. But getting all of that to work together was harder than it sounds.
Most approaches at the time were built as pipelines: one module would process the signal, hand its output to the next, and so on. That worked when each stage was reliable. Speech is not reliable. Noise, accent and ambiguity meant errors accumulated as they passed down the chain, and no later stage could reach back to correct an earlier one. Researchers knew they needed something more like a debate between experts than a production line, but there was no agreed way to build that.
Carnegie Mellon University had been running the ARPA Speech Understanding Research project through the 1970s, and the Hearsay-II system was its most ambitious product. Hearsay-II was built to understand continuous spoken English from a vocabulary of about a thousand words. Getting it to work required solving the coordination problem, not just for speech but in a way that might generalise.
What happened
In 1980, Lee Erman, Richard Hayes-Roth, Victor Lesser and D. Raj Reddy, working at Carnegie Mellon University and the RAND Corporation, published a paper in the journal Artificial Intelligence, volume 14, issues 1–2, that ran to 176 pages. The paper described Hearsay-II in full, and in doing so gave a precise, replicable account of what they called the blackboard model.
The architecture worked like this. A shared data structure, the blackboard, held the system’s current best picture of the utterance being analysed, organised into levels from raw acoustic segments up to word sequences. Separate modules, called knowledge sources, each specialised in one type of analysis. A knowledge source would watch the blackboard for a pattern it could act on, do its work, and write its conclusions back. No knowledge source talked directly to another. All coordination happened through the shared structure, and a scheduler decided which knowledge source to run next based on the state of the blackboard and estimates of how useful each potential action would be.
What this meant in practice was that the system could combine partial, uncertain evidence from many directions at once. An acoustic reading might suggest two possible words; a grammatical knowledge source might find that only one of them fits the context; the blackboard would be updated accordingly. Erman, Hayes-Roth, Lesser and Reddy documented the architecture, the scheduling mechanism and the knowledge sources in enough detail that other researchers could study, critique and build on the design directly. Before the paper, Hearsay-II had existed as a working system. After it, the blackboard model existed as a named, described abstraction.
Why it mattered
The blackboard model formalised a pattern, multiple specialised knowledge sources reading from and writing to a shared data structure, that became the architectural template for distributed AI systems well beyond speech recognition. It addressed the core challenge of combining heterogeneous, uncertain evidence incrementally, influencing later expert-system shells, real-time control architectures and multi-agent system design. The Hearsay-II paper gave researchers a replicable, named abstraction they could debate, refine and build upon, rather than reimplementing ad hoc coordination mechanisms from scratch.
People
Lee Daniel Erman, Richard Hayes Roth, Victor R. Lesser, D Raj Reddy
Organisations
Carnegie Mellon University, RAND Corporation
Sources
- The Hearsay-II Speech-Understanding System: Integrating Knowledge to Resolve Uncertainty.Stanford University Libraries (scan of Artificial Intelligence, vol. 14, nos. 1–2, 1980).Primary source
- The Hearsay-II Speech-Understanding System: Integrating Knowledge to Resolve Uncertainty.Elsevier via ScienceDirect (Artificial Intelligence, vol. 14, issues 1–2, pp. 1–176).Primary source
Cite this page
AI Achievements. (1980). Blackboard Model Description by Lee Erman, Richard Hayes-Roth, Victor Lesser and D. Raj Reddy. Retrieved 2026-08-22, from https://achievements.ai/milestone/blackboard-model-description-by-lee-rick
@misc{achievements_blackboard_model_description_by_lee_rick,
title = {Blackboard Model Description by Lee Erman, Richard Hayes-Roth, Victor Lesser and D. Raj Reddy},
author = {{AI Achievements}},
year = {1980},
url = {https://achievements.ai/milestone/blackboard-model-description-by-lee-rick}
}