A ~300-line Python reproduction of the BFF universe from the paper Computational Life: How Well-formed, Self-replicating Programs Emerge from Simple Interaction.

A 240×135 grid of randomly initialized Brainfuck-like programs interact with their neighbors. Every epoch, each program is paired with a random neighbor — their instruction tapes are concatenated, executed for up to 2¹³ steps, and then split back apart.

The instruction set allows programs to read from and write to each other's tapes, creating selection pressure for self-replicating programs. As shown in the paper, self-replicators spontaneously emerge and quickly spread across the entire grid.

Every pixel represents one instruction; each opcode has a unique color, while near-black pixels are raw data. Every 8×8 block is one program.

Run a simulation yourself:

uv run main.py --seed 1

This produces universe_mutation.gif — a self-replicating pattern emerges early and spreads across the grid, until an even more efficient replicator evolves and takes over completely.

• Numba JIT — parallel execution of program pairs via @njit(parallel=True)
• Background mutation — configurable random bit-flip rate
• GIF export — colored visualization of the evolving grid
• Metrics export — opcode density and Shannon entropy over time
• AI researcher — researcher.py autonomously runs experiments and analyzes results using the Claude API

Requires Python ≥ 3.14 and uv.

git clone https://github.com/as311-ops/artificial-life-explorer
cd artificial-life
uv sync

# Basic run (7500 epochs, saves universe_mutation.gif)
uv run main.py --seed 1

# Faster run without GIF rendering
uv run main.py --seed 1 --no-gif --num-epochs 2000

# Export metrics to JSON
uv run main.py --seed 1 --metrics-path metrics.json --metrics-every 50

# Custom grid size
uv run main.py --grid-width 80 --grid-height 45 --num-programs 3600

researcher.py is an autonomous research agent that uses the Claude API to design and analyze experiments:

export ANTHROPIC_API_KEY=sk-ant-...
uv run researcher.py --max-experiments 20 --max-cost 2.0

The agent iteratively modifies simulation parameters, runs experiments, and documents findings in research_journal.md.

Faldor, M., Zhang, J., Bhatt, U., & Cully, A. (2024). Computational Life: How Well-formed, Self-replicating Programs Emerge from Simple Interaction. arXiv:2406.19108.