# Smart Bricks Achieve Collective Intelligence _Sakana AI's 'Smart Cellular Bricks' use local communication and identical neural networks to enable physical modules to collectively identify shapes and detect damage._ **Published:** 2026-07-13 **Source:** https://www.startuphub.ai/ai-news/technology/2026/smart-bricks-achieve-collective-intelligence --- Sakana AI, in collaboration with researchers from IT University of Copenhagen and Autodesk, has developed what it calls 'Smart Cellular Bricks', a system where hundreds of physical, modular hardware units collectively infer their overall shape class. This groundbreaking work extends the principles of collective intelligence, previously explored in simulated AI systems, into the tangible world, with findings published in Nature Communications. Smart Cellular BricksCore From the articleSakana AI, in collaboration with researchers from IT University of Copenhagen and Autodesk, has developed what it calls 'Smart Cellular Bricks', a system where hundreds of physical, modular hardware units collectively infer their overall shape class.Local CommunicationContexteach brick communicates only with physically adjacent modules, no central commandFrom the article 8 mentionsCommunication is strictly local, limited to physically connected neighbors.Identical Neural NetworksCoreFrom the article 2 mentionsEach 'brick' is a simple cubic module running an identical, small neural network.Collective IntelligenceContextsophisticated behavior emerges from many simple components operating under local rulesFrom the article 8 mentionsThis groundbreaking work extends the principles of collective intelligence, previously explored in simulated AI systems, into the tangible world, with findings published in Nature Communications.Infer Overall ShapeEffecthundreds of bricks collectively identify their combined shape class without central oversightFrom the articleSakana AI, in collaboration with researchers from IT University of Copenhagen and Autodesk, has developed what it calls 'Smart Cellular Bricks', a system where hundreds of physical, modular hardware units collectively infer their overall shape class.Damage DetectionEffectsystem can identify and potentially regenerate damaged sections within the collective structureFrom the article 4 mentionsTrained for both shape classification and damage detection, the system achieves high accuracy on both tasks.suggestsScaling PotentialOutcomeprinciples extend to larger, more complex systems mirroring biological organizationFrom the articleThe research indicates predictable scaling behavior, suggesting its potential for large-scale applications.contributes toTangible AI SystemsOutcomeextends simulated collective intelligence into the physical, real-world domainFrom the article 9+ mentionsThis groundbreaking work extends the principles of collective intelligence, previously explored in simulated AI systems, into the tangible world, with findings published in Nature Communications. The core idea is that sophisticated behavior can emerge from numerous simple components operating under local rules, without any central command. This mirrors biological systems like ant colonies or human tissues. Sakana AI has previously applied this concept to AI models, enabling them to reason together or negotiate with partial views to reach a unified solution. The new physical system takes this a step further. Each 'brick' is a simple cubic module running an identical, small neural network. Communication is strictly local, limited to physically connected neighbors. Crucially, no brick knows its own position or its role within the larger structure. Yet, through these purely local interactions, the collective accurately identifies its global shape. Beyond shape classification, the system can also detect missing or damaged modules. This robustness is key, as the bricks largely maintain their accuracy even with up to 15% of modules failing silently. This resilience echoes the self-organization seen in biological regeneration, where organisms can repair damage without external guidance. This approach is a significant step towards adaptable, intelligent artificial collectives, potentially impacting fields like smart materials and reconfigurable robotics, akin to advancements in [modular robotics](/ai-news/robotics/2026/hugging-face-s-reachy-mini-the-300-open-source-robot). ## How it Works The system leverages differentiable Neural Cellular Automata (NCA), extending traditional cellular automata by learning local update rules via a deep neural network. Each cell processes information from its neighbors and its own memory, gradually converging on a global classification. The collective is trained to distinguish between broad shape classes, such as planes, chairs, or cars, rather than memorizing specific configurations. This abstract shape classification allows for flexibility. The system can identify various forms within a class, like different types of tables, and remains robust to variations not seen during training. This generalization capability is a hallmark of biological intelligence. Emergent communication strategies are also observed. The NCA's hidden channels develop morphogen-like activation patterns, creating gradients that provide positional information, similar to embryonic development. These patterns help differentiate complex shapes, such as distinguishing a chair from a table by establishing an anterior-posterior axis. ## Damage Detection and Regeneration An exciting extension of this research is the bricks' ability to not only recognize their shape but also detect local damage. Trained for both shape classification and damage detection, the system achieves high accuracy on both tasks. This capability opens the door to self-repair. By mimicking biological regeneration, the system can regrow damaged structures. Starting from a small cluster, new bricks are added in directions indicated by neighboring cells, guided by local damage assessments. This decentralized growth process allows the collective to reconstruct shapes, demonstrating a form of artificial biological repair. This represents a significant advance in [collective intelligence in physical systems](/ai-news/claude-s-corner/2026/claudes-corner-human-archive-yc-w2026), a field also explored in contexts like [collective intelligence in physical systems](/ai-news/claude-s-corner/2026/claudes-corner-asimov-yc-w2026) and [collective intelligence in physical systems](/ai-news/ai-research/2026/google-deepmind-fuels-european-robotics-startups). ## Scaling and Future Potential While hardware experiments are limited, simulations show the system scales effectively to much larger and more complex morphologies, including hollow structures and assemblies exceeding 18,000 cubes. The research indicates predictable scaling behavior, suggesting its potential for large-scale applications. The Sakana AI team envisions future applications ranging from self-reporting smart materials for resilient architecture to adaptable, LEGO-like construction systems. This work blurs the lines between construction and computation, embedding distributed intelligence directly into the building blocks of physical systems. It signifies a concrete step towards truly modular, self-reconfigurable systems capable of autonomous adaptation and repair. --- Original analysis from [startuphub.ai](https://www.startuphub.ai), the #1 AI startup directory.