Linguistic Diversity Slows Consensus in Neural Cellular Automata
A new study from arXiv (2606.21202) investigates how communication heterogeneity affects collective consensus in Neural Cellular Automata. The research introduces 'languages' as sub-populations that read each other's messages through a translation with tunable 'linguistic distance'. Key findings show that linguistic distance slows consensus, produces mild divergence rather than full fragmentation, and that collectives trained under diverse protocols are robust to mismatch, while homogeneously trained ones are not. The results hold on both ring and two-dimensional grid topologies and are interpreted as Ising relaxation. This work challenges the assumption of a single communication protocol in models of collective intelligence and has implications for understanding consensus in distributed systems.
Key facts
- The study is published on arXiv with identifier 2606.21202.
- It uses Neural Cellular Automata to solve the density classification task.
- Languages are introduced as sub-populations with tunable linguistic distance.
- Linguistic distance slows consensus.
- Mild divergence between groups occurs, not full fragmentation.
- Collectives trained under diverse protocols are robust to mismatch.
- Homogeneously trained collectives are not robust.
- Findings hold on both a ring and a two-dimensional grid.
- The results admit a natural reading as Ising relaxation.
Entities
—