Blackboard Architecture

The Blackboard is the central, shared data structure that represents the current state of the problem and its evolving solution. It acts as a global workspace where:

• Hypotheses and partial solutions are posted.
• Data is organized into levels of abstraction (e.g., raw input, features, hypotheses, final answer).
• Knowledge Sources monitor the blackboard for relevant changes to trigger their execution. Its structured nature allows different specialized agents to contribute without direct coordination.

Knowledge Sources (KS) are independent, specialized modules or agents that contain the expertise needed to solve part of the overall problem. Each KS is:

• Event-driven: It monitors the blackboard for specific conditions or patterns that match its domain of expertise.
• Opportunistic: It activates autonomously when it can contribute to the solution.
• Encapsulated: It has its own reasoning mechanism and does not call other KS directly. Examples include a signal processing KS for raw data, a hypothesis generator, and a validator KS.

The Control Component manages the flow of problem-solving by deciding which Knowledge Source should execute next. It is responsible for:

• Scheduling: Evaluating the potential contributions of all triggered KS and selecting the most promising one.
• Conflict Resolution: Managing situations where multiple KS are applicable or have competing solutions.
• Focus of Attention: Directing computational resources to the most productive areas of the problem space. This component ensures efficient and directed collaboration among the independent agents.

The Problem-Solving State is the complete representation of progress on the blackboard at any given moment. It evolves incrementally through KS contributions and includes:

• Input Data: The raw problem statement or sensor data.
• Intermediate Results: Partial solutions, annotations, and hypotheses at various abstraction levels.
• Solution Space: The set of all candidate solutions being explored.
• Control Data: Information used by the control component, such as confidence scores or execution history. This state is the sole medium of indirect communication between KS.

The Hearsay-II system is the canonical example of blackboard architecture, designed for speech recognition in the 1970s. It demonstrated the pattern's power for complex signal interpretation:

• Blackboard Levels: Included spectral features, phonemes, syllables, words, and phrases.
• Specialized KS: Independent agents worked on acoustic analysis, lexical lookup, and syntactic parsing.
• Opportunistic Execution: A KS for word detection could trigger a KS for phrase formation, which in turn might refine earlier phonetic hypotheses. This system showed how fragmented expertise could integrate to solve a problem no single algorithm could.

Choreography is a related but distinct distributed coordination pattern. In choreography, control logic is distributed; each participant knows when to act based on observed events or messages, without a central controller. Contrast with Blackboard Architecture:

• Blackboard: Centralized data (blackboard) and often a centralized controller. KS are decoupled but coordinated via the shared state.
• Choreography: Fully decentralized. Participants communicate directly via events (e.g., using a message broker), and the workflow emerges from their collective, pre-defined reactions. Choreography is often used in microservices, while blackboard is suited for collaborative reasoning systems.

A classic task allocation and negotiation protocol inspired by economic contracting. A manager agent announces a task. Contractor agents evaluate the announcement and submit bids. The manager evaluates bids, awards the contract to the best-suited contractor, and manages the result. This protocol provides a structured alternative to the opportunistic problem-solving of the blackboard, introducing explicit negotiation phases.

• Key Phases: Task Announcement, Bidding, Awarding, Execution.
• Contrast with Blackboard: More structured and explicit negotiation vs. the blackboard's opportunistic, data-driven collaboration.

A messaging pattern central to decoupled, event-driven agent communication. Agents (publishers) send messages categorized by topics without knowing the specific recipients. Other agents (subscribers) express interest in one or more topics and receive relevant messages asynchronously. This pattern is often used to implement the notification mechanism in a blackboard system, where knowledge sources subscribe to changes on specific areas of the blackboard.

• Core Mechanism: Topic-based routing of messages.
• Blackboard Application: Knowledge sources can subscribe to 'events' on the blackboard (e.g., 'new-hypothesis-added', 'partial-solution-updated').

A decentralized coordination pattern where control logic is distributed among participating agents. Each agent knows when to execute its operations based on the observed events or messages from other agents, without a central orchestrator. This contrasts with orchestration, which uses a central controller. Blackboard architecture is a form of choreography; the control component coordinates, but the decision of when a knowledge source acts is driven by the state of the shared blackboard, not direct commands.

• Key Trait: Distributed control, emergent workflow.
• Relation to Blackboard: The blackboard pattern is a specific, structured instance of choreography focused on collaborative problem-solving.

The software infrastructure that enables reliable, asynchronous message exchange between distributed components, including agents. It provides the 'plumbing' for patterns like publish-subscribe and message queues. A blackboard system, especially in a distributed deployment, relies on MOM to handle the underlying communication between knowledge sources and the blackboard service, ensuring messages (contributions, notifications) are delivered reliably.

• Common Implementations: Apache Kafka, RabbitMQ, Amazon SQS/SNS.
• Provides: Decoupling, reliability, scalability, and often message persistence.

The set of techniques and protocols used to maintain consistency of shared information across a distributed set of agents. In a distributed blackboard system, this is a critical challenge: ensuring all knowledge sources have a consistent view of the blackboard's state. Mechanisms include optimistic concurrency control, conflict-free replicated data types (CRDTs), and distributed consensus algorithms to resolve conflicting updates.

• Core Challenge: Maintaining a single, logical 'truth' across replicas.
• Blackboard Relevance: Directly impacts the integrity of the shared problem-solving workspace.

The systematic process by which agents in a network advertise their capabilities and locate other agents. In a dynamic blackboard system, knowledge sources must register with the control component or a central directory, declaring the types of data they can process or produce. This allows the control component to intelligently route notifications or for agents to find collaborators. It enables plug-and-play extensibility of the system.

• Mechanisms: Service registries, yellow/white page services, multicast discovery.
• Enables: Dynamic addition and removal of specialized problem-solving agents.