Emergent Models: Machine Learning from Cellular Automata

Abstract

Emergent Models (EMs) are a machine learning paradigm based on simple yet open-ended substrates, such as cellular automata, in which modeling is treated not as the learning of a closed-form input-output map but as the emergence of computational behaviors in simple dynamical systems. Such substrates are often based on the iteration of a fixed local rule across a latent space for an adaptive number of steps, with an interface linking the automaton to external input/output signals. In this setting, training proceeds through evolutionary search over the space of initial conditions and sometimes update rules until the desired behavior emerges. We hypothesize that some instances of this framework are biased toward global generalization of functions, offering strong extrapolation capabilities. We present a minimal cellular-automaton realization, EM43, that supports this hypothesis while highlighting some limitations. This work is foundational and theoretical: it advances a different ontology of learning systems and provides a minimal proof of concept in an intentionally narrow setting, with the aim of motivating further studies. Code: https://github.com/EmergentComputing/exp-em43 Demo: https://emergentcomputing.github.io/em43demo/