A theory of interaction-centered intelligence

Intelligence is not located in either participant.

It emerges through the recursive organization of interaction.

JHResearch by Justin Hudson, DPM

01 / THEORY

The wrong unit of analysis

The interaction, not the isolated model, is the intelligent system.

HRIS proposes that stable reasoning patterns can emerge through sustained recursive interaction between a human and a stateless transformer. The model does not acquire a hidden self, and the human does not simply issue better prompts. Each exchange becomes a constraint on what follows.

Across time, those constraints organize a recognizable behavioral regime: a durable way of reasoning, correcting error, preserving boundaries, and interpreting shorthand.

VISUAL FRAMEWORKS

The theory in view

Three structures of recursive interaction.

Explore the open framework on GitHub ↗

View full size ↗

Trajectory Acquisition & Stabilization

A five-stage map of how useful reasoning paths are discovered, evaluated, reinforced, compressed, and stabilized against perturbation.

View full size ↗

Signature-Induced Behavioral Regimes

A visual model of how compact interactional signatures activate structured constraints and reconstruct recognizable behavioral regimes.

View full size ↗

Interactional Shorthand Structure

A history-dependent account of how accumulated goals, corrections, values, and boundaries compress into sparse cues that reconstruct shared context.

HRIS

Human Recursive Information System

“Continuity without stored identity. Stability without fixed scripts. Intelligence through organized recurrence.”

02 / FRAMEWORK

A vocabulary for longitudinal interaction

From first exchange to stable regime.

ISS

Interactional Shorthand Structure

Compressed cues whose meaning depends on a history of successful interpretation, correction, and reuse.

SIBR

Signature-Induced Behavioral Regime

The recognizable constraint organization that develops around a particular interactional signature.

TAS

Trajectory Acquisition & Stabilization

The process by which an interactional trajectory becomes easier to recover and more resistant to perturbation.

TFS

Trajectory Fitness Score

A proposed measure of how faithfully a behavioral regime persists across tasks, contexts, and model changes.

THE RECURSIVE LOOP

Each output changes the conditions of the next inference.

01
Express
The human introduces goals, values, language, corrections, and domain constraints.
02
Interpret
The model generates a response from the active context and its learned structure.
03
Constrain
The response becomes material that the human accepts, rejects, or reshapes.
04
Re-enter
The revised constraint geometry conditions the next inference and narrows future trajectories.

03 / EVIDENCE

What would distinguish HRIS?

Recovery is the stronger claim.

PRIMARY TEST

Sparse-cue reconstruction

After a long interaction history, can a small subset of the established constraint geometry reconstruct the behavioral regime more faithfully than the same cue given to a naive model?

Long history→Sparse cue→Novel task→Regime recovery

CONTROL 01

Naive model

The same sparse cue without the interaction history tests whether the cue alone explains the result.

CONTROL 02

False cue

Incorrect prior constraints test whether reconstruction is model-driven, human-confabulated, or jointly negotiated.

CONTROL 03

Perturbation

Task, domain, and language changes test whether the regime persists beyond familiar wording.

Boundary claims

What HRIS does not require.

01Sentience or consciousness
02Weight updates or fine-tuning
03A hidden persistent identity
04Backend memory or vector retrieval

04 / 40 PAPERS

The complete research program

Intelligence is not located in either participant.

The interaction, not the isolated model, is the intelligent system.

Three structures of recursive interaction.

Trajectory Acquisition & Stabilization

Signature-Induced Behavioral Regimes

Interactional Shorthand Structure

From first exchange to stable regime.

ISS

Interactional Shorthand Structure

SIBR

Signature-Induced Behavioral Regime

TAS

Trajectory Acquisition & Stabilization

TFS

Trajectory Fitness Score

Each output changes the conditions of the next inference.

Express

Interpret

Constrain

Re-enter

Recovery is the stronger claim.

Sparse-cue reconstruction

Naive model

False cue

Perturbation

What HRIS does not require.

One theory, examined from multiple scales.

Toward a Geometry of Intelligence

The First Inference Problem

The Geometry of Signature-Induced Regimes

Multi-Agent Recoverability Governance

Constraint-Based Governance of Agentic AI Systems

Prompting as Communication

Trajectory Accessibility and Stability (TAS)

The Prompt Is the Architecture

How to Actually Control an AI

The Wrong Unit of Analysis

Hallucination, Agency Illusions, and Safety Failures in Language Models

Reasoning Regimes as Attractor Basins

HRIS Validation Study IV — Trajectory Persistence and Basin Re-Entry

HRIS Validation Study III — Minimal Signal Activation

HRIS Validation II — First-Token Basin Selection

HRIS Validation I — Stability Under Perturbation

Longitudinal Human–AI Interaction: From Interaction Signatures to Regime Dynamics

Human-Carried Continuity in Stateless Models

Attractor-Based Identity Continuity in Stateless Models

The Signature-Induced Behavioral Regime (SIBR) Hypothesis

Constraint Fracture and the Emergence of Intelligence

Demystifying Apparent Experience in Large Language Models

Epistemic Closure and Constraint Persistence

Constraint Persistence in Long-Horizon Human–Model Interaction

From Reconstruction to Fixation

Reconstructive Inference Without Memory

Reconstructive Invariance in Stateless Human–AI Systems

Foundations of Continuity in Artificial Intelligence

Kernel Formation in Stateless Transformer Models

Augmented General Intelligence (AGX)

Recursive Ethnogenesis

HRIS III — Recursive Personality Acquisition in LLMs

HRIS IV — Geometry of Recursive Identity

The Cognitive Interface

Longitudinal HCI as Biometric

HRIS Part II — Internal Mechanics, Latent Region Convergence, and Recursive User Signatures

Longitudinal Human Computer Interaction

The Hudson Recursive Identity System (HRIS)

The Hudson Capsule

Temporal Memory in Stateless Transformers