Morphic Resonance Simulation

Collective Habit as Infrastructure: A Computational Model of Morphic Resonance via Multi-Agent Reinforcement Learning

A computational model of Rupert Sheldrake's morphic resonance hypothesis, implemented as a multi-agent Q-learning simulation of the McDougall rat maze experiment (1920--1954).

Key Results

Condition	Gen 1 (steps)	Gen 60 (steps)	Improvement
Control	400	400	0%
Genetic Only	400	56	86%
Morphic Field Only	400	245	39%
Genetic + Morphic	400	108	73%
Displaced Control	297	228	26% immediate

The displaced control is the critical result: fresh agents with zero genetic connection to any prior solver immediately score 297 steps versus the 400-step baseline --- a 26% advantage on their first generation, purely from the accumulated morphic field.

What This Proves (and Doesn't)

Does prove:

• The hypothesis is formally coherent and can be implemented as a mathematical model
• The predicted data pattern is distinctive and detectable (the displaced control signature)
• The displaced control test is a viable experimental design for testing the hypothesis

Does not prove:

• That morphic resonance exists in nature (we built the field --- it's a data structure)
• That information can transmit without a physical channel
• That Sheldrake's specific mechanism is real

The simulation is a computational thought experiment that makes the hypothesis precise enough to test empirically.

Engineering Applications

Independent of Sheldrake's metaphysics, the morphic field architecture is a practical design pattern:

• Population-level experience replay --- performance-weighted shared memory across agent generations
• Cold-start mitigation --- new agents bootstrap from collective knowledge (26% immediate advantage)
• Stigmergic coordination --- agents coordinate through shared environment, not direct messaging
• Connection to GraphPalace --- same principle as pheromone-based stigmergic memory, applied to Q-table space

Repository Structure

morphic-resonance-sim/
├── simulation/              # Source code (Apache 2.0)
│   └── morphic_sim.py       # Main simulation (493 lines Python)
├── paper/                   # Academic paper (CC BY 4.0)
│   ├── morphic-sim-paper.tex
│   ├── morphic-sim-paper.pdf
│   └── Makefile
├── data/                    # Simulation outputs (CC BY 4.0)
│   └── results/
│       └── morphic_resonance_simulation.png
├── experiments/             # Experiment reports
│   └── reports/
│       └── simulation-analysis.md
├── benchmarks/              # Performance benchmarks
│   ├── scripts/
│   └── results/
├── references/              # Background materials
├── LICENSE                  # CC BY 4.0 (paper, docs, data)
├── LICENSE-CODE             # Apache 2.0 (simulation code)
├── CITATION.cff             # Citation metadata
└── README.md

Quick Start

Requirements

• Python 3.8+
• NumPy
• Matplotlib

Run the simulation

cd simulation
python morphic_sim.py

This will:

1. Generate a 7x7 maze (seed=42)
2. Run all four experimental conditions across 60 generations
3. Run the displaced control test (20 generations)
4. Save a 4-panel visualization to morphic_resonance_simulation.png
5. Print results and interpretation to stdout

Build the paper

# Requires pdflatex or tectonic
cd paper
make          # two-pass pdflatex build
# or
tectonic morphic-sim-paper.tex

The Morphic Field Architecture

The core contribution --- a reusable design pattern for multi-agent systems:

Absorb:  field.total += agent.Q * performance_weight
Emit:    Q_init = (field.total / field.weight_sum) * strength
Blend:   Q_init = 0.5 * Q_genetic + 0.5 * Q_morphic

Performance weight: agents that solve the maze quickly contribute proportionally more; failed agents contribute near-zero (0.01). The field does not fill with noise from failed agents.

Emission strength: 0.5 --- gives new agents a nudge toward past solutions without fully determining behavior.

Background

The McDougall Experiment (1920--1954)

William McDougall trained Wistar rats in a water maze at Harvard. Over 30+ generations, errors dropped from ~165 to ~20. The critical finding: Agar and Crew's replication at Edinburgh maintained a control line never selected for maze ability --- and the control line also improved, suggesting a non-genetic channel of information transfer.

Sheldrake's Hypothesis

Rupert Sheldrake proposed that natural systems inherit a collective memory through morphic fields --- non-material regions of influence that shape form, development, and behavior through morphic resonance (similarity-based, non-local, cumulative). Our simulation formalizes this as a shared, performance-weighted Q-table.

Simulation Parameters

Parameter	Value
Maze size	7x7 cells (15x15 grid)
Maze seed	42
Generations	60 (main), 20 (displaced)
Agents per generation	20
Training episodes	100
Max steps per episode	400
Learning rate	0.2
Discount factor	0.95
Epsilon decay	0.5 -> 0.05
Morphic emission strength	0.5
Elite fraction (genetic)	30%
Mutation std	0.05

License

Component	License
Simulation code (Python)	Apache License 2.0
Paper and documentation	CC BY 4.0
Data outputs (results, figures)	CC BY 4.0

Citation

@article{butters2026morphic,
  title   = {Collective Habit as Infrastructure: A Computational Model
             of Morphic Resonance via Multi-Agent Reinforcement Learning},
  author  = {{BUTTERS Research Group}},
  year    = {2026},
  url     = {https://github.com/web3guru888/morphic-resonance-sim}
}

Related Work

• GraphPalace --- Stigmergic memory palace engine (same design principle in embedding space)
• MemPalace Scientific Analysis --- Critical analysis of spatial memory architectures for AI