Reasoning with Neural Cellular Automata
Abstract
Modern AI architectures used to solve visual reasoning tasks typically rely heavily on global connectivity and synchronization. As biological systems demonstrate, though, sophisticated computation can be performed in a more decentralized fashion. In this work, we test the reasoning capabilities of Neural Cellular Automata (NCAs), networks of recurrent cells that use strictly local connectivity and asynchronous updates. NCAs have been extensively studied in artificial life experiments, but it is unclear whether they can perform complex multi-step reasoning. We show that NCAs produce spatio-temporal dynamics capable of solving challenging visual reasoning tasks, including large mazes, Sudoku, and ARC-AGI-1. Furthermore, we provide evidence that NCAs generalize out-of-distribution when running with larger grids, longer rollouts, or parallel trials; and that the latter can be made more efficient via pruning of redundant trajectories. We find that these generalization capabilities depend on training with sample replay and stochastic perturbations, and that stochasticity remains beneficial at test time. Finally, we show that NCAs are robust reasoners capable of dynamically modulating compute to recover efficiently from damage, and that they can scale to solve reasoning in raw pixel space.
Related
BibTeX
@misc{etcheverry2026reasoning,
title = {{Reasoning with Neural Cellular Automata}},
author = {Mayalen Etcheverry and Pietro Miotti and Aidan Sirbu and Konstantin Schürholt and Mariia Drozdova and Arna Ghosh and Blaise Agüera y Arcas and James Manyika and Blake Richards and Eyvind Niklasson},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.36126},
eprint = {2609.36126},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.36126},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Reasoning with Neural Cellular Automata
AU - Mayalen Etcheverry
AU - Pietro Miotti
AU - Aidan Sirbu
AU - Konstantin Schürholt
AU - Mariia Drozdova
AU - Arna Ghosh
AU - Blaise Agüera y Arcas
AU - James Manyika
AU - Blake Richards
AU - Eyvind Niklasson
PY - 2026
DA - 2026/09/28
PB - arXiv
DO - 10.48550/arxiv.2609.36126
UR - https://arxiv.org/abs/2609.36126
AB - Modern AI architectures used to solve visual reasoning tasks typically rely heavily on global connectivity and synchronization. As biological systems demonstrate, though, sophisticated computation can be performed in a more decentralized fashion. In this work, we test the reasoning capabilities of Neural Cellular Automata (NCAs), networks of recurrent cells that use strictly local connectivity and asynchronous updates. NCAs have been extensively studied in artificial life experiments, but it is unclear whether they can perform complex multi-step reasoning. We show that NCAs produce spatio-temporal dynamics capable of solving challenging visual reasoning tasks, including large mazes, Sudoku, and ARC-AGI-1. Furthermore, we provide evidence that NCAs generalize out-of-distribution when running with larger grids, longer rollouts, or parallel trials; and that the latter can be made more efficient via pruning of redundant trajectories. We find that these generalization capabilities depend on training with sample replay and stochastic perturbations, and that stochasticity remains beneficial at test time. Finally, we show that NCAs are robust reasoners capable of dynamically modulating compute to recover efficiently from damage, and that they can scale to solve reasoning in raw pixel space.
ER -