Regime-Adaptive Gating

In the Dimensional Emergence Framework (DEF), intelligence is not about running every process at maximum capacity all the time. Instead, it is about activating the appropriate Regime Level (R) for the task at hand. Regime-Adaptive Gating is the mechanism that manages this complexity.

The Router Mechanism

At the heart of a DEF-aligned architecture lies the Router (or Controller). This component acts as a traffic manager for information flow, deciding which dimensional streams (A, B, C, or D) are necessary for the current operation.

The Router evaluates several inputs to determine the required regime:

• Instruction Modus: Is there an active command or communicative intent?
• Planning Horizon: How far into the future or how abstract is the current goal?
• Uncertainty/Error: High TD-error or reward uncertainty may trigger higher-order processing.
• Memory Budget: Balancing the computational cost of high-dimensional streams.
• Narrative Drift: Monitoring the coherence of the language stream to trigger deconstruction if needed.

Regime Transitions

The gating logic follows a strictly hierarchical path to ensure grounding:

1. R2 Regime (Physicality): Only Streams A and B (World State & Dynamics) are active. This is used for “sub-symbolic” tasks like basic navigation or reactive obstacle avoidance.
2. R3 Regime (Sentience): Stream C (Valence & Reference) is activated. The agent now operates with goals and preferences, allowing for autonomous decision-making.
3. R4 Regime (Consciousness): Stream D (Meaning & Narrative) is activated. This is the most resource-intensive mode, used for language generation, complex reasoning, and long-term planning.

Biological Inspiration: Task-Positive vs. Task-Negative

This gating logic is biomimetic. In the human brain, we observe a similar “Regime-Switching” where the Default Mode Network (DMN)—which corresponds to our Narrative stream—is naturally suppressed when the brain is focused on immediate physical or motor tasks (Task-Positive).

By implementing this as a hard constraint, we prevent the AI from “over-thinking” or hallucinating during simple execution tasks, ensuring that high-level reasoning only occurs when it is contextually relevant.