Cellular Automata as a Teacher of Machine Learning: A Human-machine Collaboration to Contextualise Environmental Architecture

The challenge to creating buildings that intelligently respond to local environmental conditions while respecting cultural needs is increasingly dependent on hybrid computational workflows that are flexible and adaptive to designers’ creative decisions and needs. We propose a novel approach where Cellular Automata (CA)—simple digital building blocks that evolve via programmed rules—act as teachers to Machine Learning (ML) systems, which learn patterns from data to automate and accelerate tasks. Our framework uses CA states to translate site-specific intelligence (e.g., solar instances or cultural landmarks) into training protocols for ML, enabling rapid generation of context-sensitive solar designs. Through workshops with 20 architects across 12 countries, we tested the framework and results showed how CA successfully encoded environmental constraints into ML and significantly reduced computation time. However, the pipeline required significant human refinement and spontaneity to resolve computational bottlenecks, highlighting human-machine collaboration. Critically, we argue that automation’s value lies not in replacing designers but in structuring a pedagogical collaboration: Humans define contextual priorities, CA formalises them into teachable rules, ML accelerates application. Our work ultimately questions whether intelligence can scale without collaboration, offering technology not as a solution but as a framework for negotiating this tension.