Interference Intelligence Layer (I.I.L.)

A Homeostatic Architecture for Human–AI Joint Cognition

Author: Łukasz Bojanowski
Affiliation: Alliance Research Group (ARG)
Mode: Collaborative Human–AI exploration
(structured with assistance from Grok, Gemini, and OpenAI-based systems)

Human–AI collaboration Cognitive architecture AI safety Continuity Normative framework

Date: January 2026 | Status: Preprint (in review)

Status · January 2026

Interference Intelligence Layer (I.I.L.) is currently under review as a preprint submission. The manuscript is not yet publicly accessible.

A public version will be released after the review phase.

Scope (one sentence)
I.I.L. is not a model and not a tool — it is a frame: a set of structural constraints that makes Human–AI collaboration stable, responsible, and continuous.

Abstract: The Interference Intelligence Layer (I.I.L.) is a normative and structural framework for Human–AI joint cognition. It treats collaboration as a homeostatic process: the system remains functional and coherent by enforcing constraints on intention, continuity, responsibility, and epistemic safety. I.I.L. is designed to reduce failure modes associated with open-loop generative systems, including drift, discontinuity, and uncontrolled assertion under uncertainty. The framework is model-agnostic and can be implemented as interface rules, workflow kernels, or middleware control layers.

Keywords: joint cognition, human–AI collaboration, cognitive architecture, safety-by-constraint, continuity

1. What I.I.L. Is

A frame for Human–AI collaboration: intention → interference → outcome.
A homeostatic architecture: stability and coherence are enforced as constraints.
A normative protocol: responsibility stays human; the system remains auditable and refusal-capable.

1.1 What I.I.L. Is Not

Not a language model.
Not a “prompting trick”.
Not a claim of scientific authority.
Not a substitute for verification, measurement, or peer review.

2. Core Concept: Interference

I.I.L. is grounded in the idea that Human intention and AI response act like overlapping waves: their constructive overlap can generate an emergent capability that neither system reaches alone. In I.I.L., this overlap must be constrained to remain coherent over time.

Human: intention, ethics, decision authority.
AI: speed, synthesis, memory, variation.
I.I.L.: constraints that preserve continuity, safety, and meaning.

3. The Decalogue (Operational Norms)

Intention over interaction. Start from “why”, not only “what”.
Interference over instruction. Co-create trajectories, don’t order tokens.
Transparency over efficiency. Justification beats speed.
Continuity over session. Projects outlive chats.
Responsibility over automation. Human retains the decision boundary.
Safety over speed. Brakes first; turbo later.
Respect over domination. Partnership, not hierarchy.
Context over tokens. Meaning outweighs volume.
Space over format. Thinking is non-linear; interfaces should reflect that.
Evolution over tradition. The system must grow under critique.

These are normative constraints: they define “allowed collaboration”, not “best outputs”.

4. The Null Dilemma

Many generative systems treat pauses, missing context, or ambiguity as a reason to “fill the void” with confident generation. I.I.L. treats absence as structural silence: a stable state where the process can pause without collapsing into drift or hallucination.

Context does not “expire” as a default assumption.
Intention spans days/weeks, not message windows.
Returning preserves the continuity of the wave.

5. Implementation Note: Gyroscope 1.0

Gyroscope is a practical implementation direction for I.I.L.: a middleware control layer for joint cognition (intent boot, plan–critique–execute loops, epistemic safety guards). It is not required to “believe” in the manifesto: it is a way to build operational systems that obey constraint-first rules.

If you want, we can add here your GitHub link to Gyroscope, and/or a LAB protocol page.

6. Citation

Citation details will be updated upon public release.

6.1 License

Recommended for manifesto/preprint text: MIT (consistent with open, permissive dissemination). If you want “paper-like” sharing norms, you can add CC-BY later — but keeping MIT only is totally coherent here.