DOCUMENTATION / START HERE
Introduction.
How the experiments and this documentation are organized.
Topics
Mind37’s experiments are interactive simulations. Each one is also a small research project with a question, a method and a result. This documentation explains each experiment in more depth than the page itself has room for: what you are looking at, what every control does, the science behind it and how to read what it found.
BACKGROUND
- Cellular automata
Grids of cells following one local rule: elementary rules, Lenia, and neural cellular automata.
- Biobots
Xenobots, anthrobots and neurobots: what they are, how cilia move them, and what isn't known.
- Evolution and search
Evolutionary search, MAP-Elites, POET, and evolution inside a simulation.
- Fluids and learned simulators
How a fluid is simulated, and neural networks trained to stand in for the solver.
COMPUTATIONAL WORLDS
- Rule Observatory
All 256 elementary cellular automata, and how eight local decisions make a world.
- A Single Difference
Flip one cell and measure how far the change spreads: damage spreading.
- Worlds at the Boundary
Walls of a second rule that pass one signal and block most others.
ARTIFICIAL LIFE
- Replication Arena
Bots that build new bots from loose cells, as xenobots do.
- Petri Ecology
A Particle Life community of biobot species, food and debris.
- Morphospace Atlas
A MAP-Elites map of evolved soft bodies, tested on tasks the search never saw.
- Anthrobot Lenia
Continuous creatures with cilia, and whether their motility falls into classes.
- Growing Anthrobots
A neural cellular automaton that grows, heals and rolls, and its hidden prepattern.
- Rejuvenation Field
A speculative model of why anthrobots test younger than their source cells.
- Neurobot Motion
A behavior score for neurobots, what it counts, and which explanations a study of its size can tell apart.
- Biobot Tierra
Four invented ways for replicating bots to inherit, and which comes closest to evolving.
- Chip Archipelago
Bodies and microfluidic chips evolving together (POET), against direct optimization.
PHYSICAL WORLDS
- Learned Simulators: Early Warnings
A neural network standing in for a flow solver: which warnings come before it fails, and does it pay?
- Steering Active Matter
A self-driven turbulent fluid, and fixed light patterns found by optimization that make it pump.
- Steering a Swarm
A swarm of microrobots pushed by one shared field through branching channels, and sequences that deliver it.
Computational Worlds are mathematical: one-dimensional cellular automata, where every detail of the rule is exact and every pattern can be traced to its cause. Artificial Life experiments are modeled on real biological robots (xenobots, built from frog cells; anthrobots, built from adult human airway cells; and neurobots, xenobots with neurons inside), simplified so they run in a browser. Physical Worlds run physics solvers: a fluid, to test how far faster stand-ins for a solver can be trusted, and a self-driven fluid steered with patterns of light.
New to these subjects? The Background topics explain them from the start: cellular automata, biobots (xenobots, anthrobots and neurobots), evolution and search, and fluids and learned simulators.
How each topic is laid out
- What the page shows
- The parts of the experiment page and what each picture or chart represents.
- How to use it
- Every control, preset and gesture, and what to try first.
- Background and theory
- The science and the method: where the idea comes from and how the model works.
- Results and what they mean
- What the research runs found, how strong the evidence is and how to interpret it.
- Limits and open questions
- What the experiment does not show, and what would test it further.
The statistics and terms the result sections share (seeds, standard errors, controls, heritability and so on) are explained once, in Reading the results.
Things every experiment shares
- The simulation runs on your device. Most experiments use WebGPU, the browser’s interface to your graphics card. Where WebGPU isn’t available (or you prefer reduced motion), the page shows a recording or a still image instead, and the rest of the page still works. The Computational Worlds and Biobot Tierra don’t need WebGPU.
- Presets are good starting points. Each preset is a setting chosen to show one behavior, with a sentence saying what to look for.
- Share links copy a URL that reopens the experiment with your exact settings (and, where it applies, the same random seed), so someone else sees the same thing.
- Simulations pause when they scroll out of view or the tab is hidden, to save battery.
- The same physics runs in the research code. The Artificial Life results come from longer runs of the same models in Python (JAX), and the browser and research versions are checked against each other. Small floating-point differences mean a live run can drift from a recorded one over time.
How to read the claims
Each Artificial Life page has a What’s real / What’s simplified box. The real column lists findings from published lab work that the experiment is based on. The simplified column lists what the model invents or leaves out. A result about the model is a result about the model: it can suggest what to test in a lab, but it isn’t evidence about real cells until it has been tested there.