Schema Redescription in Cellular Automata: Revisiting Emergence in Complex Systems

We present a method to eliminate redundancy in the transition tables of Boolean automata: schema redescription with two symbols. One symbol is used to capture redundancy of individual input variables, and another to capture permutability in sets of input variables: fully characterizing the canalization present in Boolean functions. Two-symbol schemata explain aspects of the behaviour of automata networks that the characterization of their emergent patterns does not capture. We use our method to compare two well-known cellular automata for the density classification task: the human engineered CA GKL, and another obtained via genetic programming (GP). We show that despite having very different collective behaviour, these rules are very similar. Indeed, GKL is a special case of GP. Therefore, we demonstrate that it is more feasible to compare cellular automata via schema redescriptions of their rules, than by looking at their emergent behaviour, leading us to question the tendency in complexity research to pay much more attention to emergent patterns than to local interactions.

💡 Research Summary

The paper introduces a novel method for removing redundancy from the transition tables of Boolean cellular automata (CAs) by means of a “two‑symbol schema” redescription. Traditional schema redescription uses a single wildcard symbol (“#”) to indicate inputs that are irrelevant for a given state transition, effectively compressing the lookup table (LUT) into a set of prime implicants that capture the essential inputs (enputs) governing the transition. The authors extend this approach by adding a second symbol, the position‑free marker (“○”), which denotes groups of inputs that can be permuted without affecting the outcome. This captures symmetry or group‑invariance among inputs, further reducing the number of schemata needed to represent the rule.

The methodology is applied to the well‑studied density classification task (DCT), where a binary CA must converge to a homogeneous state reflecting the majority of cells in the initial random configuration. Two high‑performing CA rules for DCT are examined: the human‑designed GKL rule and a rule discovered via genetic programming (GP). Although previous analyses based on emergent space‑time patterns (using computational mechanics, particle catalogs, etc.) reported markedly different collective behaviours for these rules, the two‑symbol schema analysis reveals that their underlying transition functions are almost identical. In fact, the GKL rule is a special case of the GP rule; the GP rule simply adds extra symmetric input groups that do not change the essential decision logic. Both rules achieve comparable classification accuracies (≈81.5 % for GKL and ≈82.2 % for GP on 10⁵ random initial conditions).

The authors argue that focusing on emergent patterns—regular domains, particles, and their interactions—provides an “emergentist” view that can obscure the fundamental control structures embedded in the local rule. By contrast, schema redescription makes explicit the canalizing inputs and the symmetry properties of the rule, offering a compact, mathematically precise description that is more amenable to automated search, comparison, and manipulation. The paper demonstrates that the space of schemata is far smaller than the space of raw LUTs, enabling more efficient evolutionary or heuristic searches for high‑performing CA rules.

Furthermore, the work positions schema redescription as a bridge between local interaction specifications and global collective dynamics, addressing a gap in complex‑systems research where the link between micro‑level rules and macro‑level behaviour is often underexplored. The authors suggest that this approach could be extended to other automata‑based models of biological and engineered networks, where canalization and symmetry play crucial roles in stability and function.

In summary, the paper provides (1) a rigorous extension of schema redescription with two symbols, (2) evidence that this representation captures essential rule properties better than emergent‑pattern analyses, and (3) a concrete case study showing that two seemingly different DCT solutions are fundamentally the same at the rule level. The findings challenge the prevailing emphasis on emergent phenomena in complexity science and advocate for a more balanced focus that includes detailed examination of local transition functions.