In non-intelligent systems, interconnections are passive

To understand this principle, it is most illustrative to view active interconnections through the lens of buildings or engineering systems. In conventional approaches, components of a system are highly modular and exchange only sparse, predetermined information. These systems are “nearly decomposable”, as Herbert Simon noted in 1969, with clear boundaries between functional units. However, this modularity comes at a cost: such systems can only respond to circumstances that their designers have anticipated. In contrast, the principle of active interconnections proposes that the connections between components should themselves be active participants in information processing, dynamically shaping the information that flows between components based on the overall system state and context.

Active interconnections set themselves apart from passive connections in their ability to broker, adapt, and transform information as it passes between components. In other words, unlike passive connections that merely transmit data, active interconnections can emphasize certain information, suppress other information, or combine information in novel ways depending on the situation. They function as “second-order components”–components that operate not directly on sensorimotor or environmental data, but on the components that process that data. This creates a system where the whole becomes significantly more than the sum of its parts, as the rich combinatorial possibilities of adjusting information flow between components allows for responses to novel situations that weren’t explicitly programmed. In essence, by using active interconnections, a system moves away from rigid, predetermined responses and gains the ability to adapt through the dynamic interplay of its components.

Active interconnections in SCIoI and related projects

If active interconnections are a recurring pattern across all instances of intelligent systems, this must also occur in the many projects we conduct at SCIoI. The matrix below illustrates this, by linking properties (rows) associated with active interconnections (such as type of behavior, domain of system building, etc.) with several projects (columns).