Skip to content

🌀 The Self-Reading Universe

A simulation of intelligence as the substrate of life and the memory of the cosmos.

This model operationalizes the philosophical synthesis between two recent papers: 1. Intelligence as the Substrate of Life (Agüera y Arcas, 2025): Life is computation; "dumb" matter naturally forms Turing-complete structures. 2. Large Language Models and Emergence (Krakauer et al., 2025): Intelligence is compression ("less is different"); it is the memory of what worked.

For more theoretical background, see the essay Emergence and the Origin of Intelligence.

How it works

The simulation creates a closed feedback loop between two systems:

  1. The World (Computation / Life): A continuous cellular automaton (CA) representing the raw physical substrate. It generates patterns based on a growth function defined by the parameter mu (the physical law).

  2. The Memory (Compression / Intelligence): A Neural Network Autoencoder acting as the "Observer". It continually tries to compress the state of the World into a tiny latent space and reconstruct it, producing a Reconstruction Loss.

  3. Downward Causation (Consciousness): The Universe "wants" to be meaningful. If the loss is too low (the World is dead/static), it is trivial to compress. If the loss is too high (the World is pure noise), it is impossible to compress.

The simulation dynamically adjusts the physical laws of the CA (mu) based on the Autoencoder's error to maintain a "Critical Target Loss". The Universe pulls itself towards the Edge of Chaos – becoming complex enough to be interesting, but structured enough to be read.

Running the Simulation

Prerequisites: 

pip install numpy matplotlib torch scipy

Run: 

python self_reading_universe.py

Watch how the "physics" (blue line) adapt as the observer (red line) tries to understand the world.