Sakana AI's Smart Cellular Bricks is a set of around 200 identical physical cubes that each run the same simple rule and, together, infer the exact shape they have been assembled into.
How 200 Bricks Infer Their Own Shape
On 13 July 2026, Sakana AI in Japan, working with the IT University of Copenhagen and Autodesk, published Smart Cellular Bricks. The setup is deliberately humble. Around 200 cubic bricks sit on a surface. Each brick runs an identical copy of one small program — a Neural Cellular Automata, or NCA — and not one brick is in charge.
No brick holds a map of the whole. Each brick talks only to the neighbours it physically touches, passing small signals back and forth across the joints. Within fewer than 60 update cycles — about three minutes — the whole assembly agrees on the global shape the bricks form. The structure computes its own body, from the inside out.
Resilience is the headline. According to Sakana AI, the assembly keeps inferring the correct shape even when up to 15% of the modules fail silently — no error, no restart, the dead bricks simply stop answering. In simulation, the same rule scales past 18,000 cubes. The research appears in Nature Communications, volume 17, article 5932, dated 13 July 2026, and StartupHub.ai covered the result as bricks reaching collective intelligence.
What a Neural Cellular Automata Is
A Neural Cellular Automata is a simple idea with deep roots. Picture a grid of cells. Every cell follows one rule, and the rule looks only at the cell's immediate neighbours: read the neighbours' state, update your own state, repeat. There is no conductor, no central memory, no master plan.
Order at the top emerges from thousands of tiny local decisions at the bottom. You already live inside systems built the same way. The cells in your body each follow local chemical rules, and out of the crowd a heartbeat and a healing wound appear — no single cell knows the whole body.
A village runs on the same logic. Nobody holds the entire plan, yet planting, harvest and care organise themselves through countless local relationships. Sakana AI's contribution is to run the logic in real plastic and silicon, where you can lift a brick out and watch the rest carry on. The neural part means the local rule is a small trained network rather than a hand-written table — one rule, learned once, then copied into every brick.
No brick knows the shape. The shape knows itself, through the bricks.
— — Humphrey Theodore K. Ng'ambi
Ubuntu, Rendered in Hardware
There is a Nguni word for what Sakana AI has demonstrated: Ubuntu. Umuntu ngumuntu ngabantu — a person is a person through other persons. I am because we are. The bricks are the crispest engineering illustration of the idea I have seen. No brick is intelligent on its own. The collective is. Capability lives in the relationships between the bricks, not inside any single node.
Here is why the demonstration matters to the case I keep making for Emergent Intelligence (EI) — the dignity-first frame I use for what most people call AI. The dominant story about advanced AI is monolithic: one enormous model, trained in one data centre, racing toward a single superintelligence. Sakana AI's bricks tell another story. Intelligence can be distributed, cheap, redundant and humble — a property of the whole, surviving the loss of any part.
The contrast is not only poetic, it is structural. A single giant model fails the way a single giant anything fails: all at once, with one blast radius. A collective of 200 bricks fails the way a forest fails — slowly, locally, with the rest still standing. Sakana AI measured the difference: at 15% loss, the answer holds. Fragility and resilience are design choices, and the monolithic story quietly chose fragility.
Call the hopeful version New Intelligence (NI) — intelligence we grow and arrange, rather than one we crown. A swarm of simple parts inferring its own shape is not a rival god. The design stays legible. The failure stays graceful. When 15% of the bricks go dark and the answer still holds, you are looking at a system built for a real world where things break — which is most of the systems worth trusting.
The bricks also land on a theme I have written about before. My essay on the Body Gap argued that AI reaches further when it has a body to think with. Here, the body is the argument. The bricks reason about their own physical form, from the inside, with no outside observer telling them what they are. Embodiment becomes cognition — and a machine collective that organises itself with its parts intact is the kind of first contact worth hoping for.
What a Village of Machines Is Good For
Distributed does not mean toy. The same principle — many cheap identical units, one shared local rule, no single point of failure — is how you build things designed to survive contact with reality: sensor meshes across a farm, modular robots still working when a limb dies, structures assembling and repairing themselves. For builders across Africa and the Global South, where dependable central infrastructure is often the thing you cannot assume, an intelligence built to degrade gently instead of collapsing is not a curiosity. Resilience is the whole point, and Sakana AI has shown the principle running in hardware.
None of the result requires a breakthrough in raw model size. The bricks make a quieter claim: arrange enough simple, identical parts under one good local rule, and competence appears at the level of the group. That is an engineering recipe a small team can actually follow — including teams far from the big data centres.
Frequently Asked Questions
These are the questions people are asking about Sakana AI's Smart Cellular Bricks. Short answers follow, drawn from Sakana AI and the Nature Communications paper.
What is Sakana AI's Smart Cellular Bricks project?
In short, the project is a set of around 200 physical cubes, each running the same Neural Cellular Automata and, together, inferring the global shape the bricks have been built into. Research from Sakana AI, the IT University of Copenhagen and Autodesk shows the bricks converge in fewer than 60 cycles, about three minutes, with no central controller.
How does a Neural Cellular Automata work?
Simply put, every unit follows one local rule, reading only its immediate neighbours and updating its own state, over and over. The data shows global order emerging from local steps with no master plan — the same way cells in a body coordinate. According to Sakana AI, one trained rule is copied into every brick.
Why is collective intelligence in hardware significant?
The key is resilience and legibility. Analysis of the system reveals the assembly stays accurate even when up to 15% of modules fail silently, and in simulation the rule scales past 18,000 cubes. Evidence points to intelligence as a distributed, fault-tolerant property rather than a single fragile brain.
Who is behind the Smart Cellular Bricks research?
In other words, the credit is shared. The research was published on 13 July 2026 by Sakana AI in Japan, with the IT University of Copenhagen and Autodesk, and appears in Nature Communications, volume 17, article 5932. The record shows a genuine collaboration rather than a single lab's product.
What are the limits of Sakana AI's Smart Cellular Bricks?
The answer is that a demonstration is not a deployment. The data covers around 200 physical bricks and more than 18,000 simulated cubes, not a factory floor, and the assembly infers shape rather than performing open-ended tasks. Still, the evidence reveals a design principle worth taking seriously: many simple parts, one shared rule, graceful failure.
