Neural cellular automata
Cellular automata in which each cell's update rule is a small neural network trained by gradient descent. The canonical line of work is Mordvintsev et al.'s Growing Neural Cellular Automata, which trains the per-cell update so that a target morphology emerges and persists from a single seed cell.
Ratings across the 9 properties
| Property | Crowdsourced (N) | Expert baseline |
|---|---|---|
| Emergence Macro-level outputs unpredictable from the microscopic rules. | — (0) | high |
| Collective intelligence Distributed knowledge production and problem-solving across many agents. | — (0) | high |
| Non-linear dynamics Small parameter or input changes produce disproportionate output shifts. | — (0) | high |
| Criticality The system operates at a boundary between order and disorder (edge of chaos). | — (0) | medium |
| Multi-scale hierarchy Distinct scales of organisation visible (micro → macro levels). | — (0) | high |
| Phase transitions Qualitative regime shifts at thresholds in a control parameter. | — (0) | medium |
| Attractors Stable configurations in state space that the system converges towards. | — (0) | high |
| Path dependence History and prior states condition future possibilities. | — (0) | medium |
| Open-endedness Unbounded generation of novelty over time. | — (0) | medium |