tl;dr In the Blackboard pattern, several specialized components assemble their knowledge to bulld a possibly partial or approximate solution. The name 'blackboard' was chosen because it is reminiscent of the situation in which human experts sit in front of a real blackboard and work together to solve a problem: Each expert separately evaluates the current state of the solution, and may go up to the blackboard at any time and add, change or delete information. Humans usually decide themselves who has the next access to the blackboard; in the pattern we describe, a moderator component decides the order in which components execute if more than one can make a contribution.

The Blackboard pattern was originally documented in Pattern-Oriented Software Architecture, Vol 1, p. 71.


"The Blackboard pattern tackles problems that do not have a feasible deterministic solution for the transformation of raw data into high level data structures, such as diagrams, tables or English phrases. Vision, image recognition, speech recognition and surveillance are examples of domains in which such problems occur." The Gang-of-Five (as the POSA1 authors were sometimes called) described Blackboard in terms that were very AI-centric, but I believe that Blackboard is applicable to a wide variety of problems well beyond the traditional AI space.

"[These problems] are characterized by a problem that, when decomposed into subproblems, spans several fields of expertise. The solutions to the partial problems require different representations and paradigms. In many cases no predetermined strategy exists for how the 'partial problem solvers' should combine their knowledge. This is in contrast to functional decomposition, in which several solution steps are arranged so that the sequence of their activation is hard-coded." Consider business scenarios where a customer is entering a variety of data, from which we need to present a number of options for their consideration, such as insurance or financial products--a mortgage, for example, or health insurance. There may be a variety of potential solutions, and the choice of which solutions fit may well be based on only small parts of the data.

"In some of the above problem domains you may also have to work with uncertain or approximate knowledge. Each transformation step can also generate several alternative solutions. In such cases it is often enough to find an optimal solution for most cases, and a suboptimal solution, or no solution, for the rest. The limitations of a Blackboard system therefore have to be documented carefully, and if important decisions depend on its results, the results have to be verified." One of the areas of research in AI was around "fuzzy logic", in which solutions are based on data values that express a "range" of potential values--for example, in health care, we often want to determine if somebody is "obese", but there's not necessarily a sharp dividing line between "healthy" and "obese" (is 200 pounds healthy or obese? It depends!), and may have some overlap.


The following forces influence solutions to problems of this kind:


The idea behind the Blackboard is a collection of independent components that work cooperatively on a common data structure. Each component is specialized for solving a particular part of the overall task, and all components work together on the solution. These specialized components are independent of each other: they do not call each other, nor is there a predetermined sequence for their activation. Instead, the direction taken by the system is mainly determined by the current state of progress. A central control component evaluates the current state of processing and coordinates the specialized components. This data-directed control regime is referred to as opportunistic problem solving. It makes experimentation with different algorithms possible, and allows experimentally-derived heuristics to control processing.

During the problem-solving process the system works with partial solutions that are combined, changed or rejected. Each of these solutions represents a partial problem and a certain stage of its solution, The set of all possible solutions is called the solution space, and is organized into levels of abstraction. The lowest level of solution consists of an internal representation of the input. Potential solutions of the overall system task are on the highest level.

Divide your system into a component called blackboard, a collection of knowledge sources, and a control component.

The blackboard is the central data store. Elements of the solution space and control data are stored here. We use the term vocabulary for the set of all data elements that can appear on the blackboard. The blackboard provides an interface that enables all knowledge sources to read from and write to it.

All elements of the solution space can appear on the blackboard. For solutions that are constructed during the problem solving process and put on the blackboard, we use the terms hypothesis or blackboard entry. Hypotheses rejected later in the process are removed from the blackboard.

A hypothesis usually has several attributes, for example its abstraction level, that is, its conceptual distance from the input. Hypotheses that have a low abstraction level have a representation that is still similar to input data representation, while hypotheses with the highest abstraction level are on the same abstraction level as the output. Other hypothesis attributes are the estimated degree of truth of the hypothesis or the time interval covered by the hypothesis.

It is often useful to specify relationships between hypotheses, such as 'part-of or 'in-support-of'.

Knowledge sources do not communicate directly-they only read from and write to the blackboard. They therefore have to understand the vocabulary of the blackboard. We explore the ramifications of this in the Implementation section.

Often a knowledge source operates on two levels of abstraction. If a knowledge source implements forward reasoning, a particular solution is transformed to a higher-level solution. A knowledge source that reasons backwards searches at a lower level for support for a solution, and may refer it back to a lower level if the reasoning did not give support for the solution.

Each knowledge source is responsible for knowing the conditions under which it can contribute to a solution. Knowledge sources are therefore split into a condition-part and an action-part. The conditionpart evaluates the current state of the solution process, as written on the blackboard, to determine if it can make a contribution. The action-part produces a result that may cause a change to the blackboard's contents.

The control component runs a loop that monitors the changes on the blackboard and decides what action to take next. It schedules knowledge source evaluations and activations according to a knowledge application strategy. The basis for this strategy is the data on the blackboard.

The strategy may rely on control knowledge sources. These special knowledge sources do not contribute directly to solutions on the blackboard, but perform calculations on which control decisions are made. Typical tasks are the estimation of the potential for progress, or the computational costs for execution of knowledge sources. Their results are called control data and are put on the blackboard as well.

Theoretically, it is possible that the blackboard can reach a state at which no knowledge source is applicable. In this case, the system fails to deliver a result. In practice, it is more likely that each reasoning step introduces several new hypotheses, and that the number of possible next steps 'explodes'. The problem is therefore to restrict the alternatives to be taken rather than to find an applicable knowledge source.

A special knowledge source or a procedure in the control component determines when the system should halt, and what the final result is. The system halts when an acceptable hypothesis is found, or when the space or time resources of the system are exhausted.

The blackboard component defines two procedures: inspect and update. Knowledge sources call inspect to check the current solutions on the blackboard, and update is used to make changes to the data on the blackboard.



A Blackboard approach tends to lead to several consequences:


Blackboards can sometimes be built out of a Pipes and Filters approach, wherein each knowledge source is represented by a component in the pipeline, and the "board" of data being examined is passed explicitly to each one in sequential order. This inhibits the ability to operate in parallel, but does simplify the implementation since we no longer have to worry about concurrency issues. Additionally, if the pipeline is arranged in a circular pipeline (a ring), then it's easy to make multiple passes without having to wonder if each knowledge source has had a chance to examine the data.


A couple of different takes on Blackboard include:

Last updated: 05 March 2022

Tags: pattern   behavioral   architectural