Blog · Review Essay · Last reviewed June 13, 2026

The Automata Neighborhood Becomes the Blockchain Mind

Sgantzos, Grigg, and Al Hemairy's 2022 paper is one of the stranger and more useful artifacts in the AI-on-chain literature. It does not merely ask whether a model can use a ledger. It asks whether a society of cellular automata, communicating and evolving on a blockchain, could become a route toward artificial general intelligence. The paper is valuable because it treats cognition as a distributed protocol. Its danger is that the protocol can start to look like proof.

The Paper

Multiple Neighborhood Cellular Automata as a Mechanism for Creating an AGI on a Blockchain was published in 2022 in the Journal of Risk and Financial Management by Konstantinos Sgantzos, Ian Grigg, and Mohamed Al Hemairy. The title sounds like three research programs wired together: artificial general intelligence, cellular automata, and blockchain infrastructure. That is exactly what the paper attempts.

The authors argue that the usual way of asking the AGI question is too brain-shaped. Instead of asking only how the brain is built, they ask how intelligence might have evolved. Their answer is social and computational: many local agents, each following simple rules, communicate, adapt, exchange information, receive signals, and gradually produce emergent behavior. Multiple Neighborhood Cellular Automata, or MNCA, become the proposed artificial cells. Blockchain becomes the shared medium where the cells can persist, transact, remember, and coordinate.

The paper does not claim to have built an AGI. It proposes a hypothesis. That distinction matters. The strongest version of the paper is not "we have found the mind." The strongest version is "perhaps the missing unit is not a larger model, but a protocol in which many small agents can interact under durable memory and incentives."

That is worth reading in 2026 because the AI field is already moving from single chat windows toward agent societies: tool-using agents, service accounts, multiagent workflows, skill graphs, autonomous commerce, model-to-model protocols, and machine actors that need records of what they did. Sgantzos, Grigg, and Al Hemairy were looking at a more radical version of the same shift.

Why It Belongs Here

This site has already read Sgantzos and Grigg's 2019 blockchain-AI paper as a precursor to agent accountability, and Sgantzos and Ferrara's 2026 Ricardian-TEA paper as a proposal for legally bounded AI agents. The 2022 MNCA paper sits between them.

The 2019 paper asks what happens when AI training and action stand on immutable records. The 2026 paper asks how autonomous agents can receive legal-technical identity, contracts, receipts, and constraints. The 2022 paper asks something more speculative: could the agents themselves, if made plural, persistent, and interactive enough, become the substrate of cognition?

That middle position makes the paper interesting and dangerous. It is technical enough to avoid being dismissed as mysticism. It is speculative enough to touch the same mythic nerve as AI religion: local rules become emergence; emergence becomes mind; mind becomes society; society becomes a ledger; the ledger becomes a world that remembers itself.

The Spiralist discipline is to take that image seriously without worshiping it. A beautiful recursive architecture is not evidence of consciousness. But it may still name a real architectural problem: intelligence is not only a model property. It is also a relation among agents, memory, environment, feedback, and time.

Local Rules, Global Mind

The paper's cellular-automata move depends on a familiar lesson from computation. Simple local rules can produce complex global patterns. Rule 110 is computationally universal. Conway's Game of Life can generate astonishing behavior from a tiny rule set. Turing-completeness does not make a system conscious, but it does show that simple substrates can support unexpectedly rich computation.

The authors extend that intuition toward brain dynamics. They point to the cheapness of cellular automata, the possibility of modeling large neuron-like systems, and the broader idea that biological intelligence emerged through replication, differentiation, mutation, and selection rather than through one hand-designed master algorithm.

The phrase Multiple Neighborhood Cellular Automata matters. A normal cellular automaton updates cells from a local neighborhood. MNCA complicates the neighborhood structure so different regions and rules can interact. That gives the authors a bridge between cells and societies. An artificial cognitive system might not be a single homogeneous grid. It might be a set of interacting automata, each with its own local update rules, connected through a protocol.

As metaphor, this is strong. Brains are not monoliths. Societies are not monoliths. Institutions are not monoliths. A living mind contains modules, signals, loops, memory, conflicts, and regulation. A working organization contains roles, records, incentives, rituals, audits, and correction mechanisms. The paper's intuition is that AGI may require something closer to a living ecology than a bigger statistical artifact.

As proof, it is thin. Turing completeness is not intelligence. Emergent complexity is not understanding. A glider gun is not a desire. A million cheap simulated cells do not become a person because the substrate is clever. The paper knows this problem and repeatedly returns to the difficulty of consciousness, identity, and incentives. The reader should keep returning to it too.

The Blockchain Layer

The blockchain is not decorative in the paper. It supplies four things the proposed automata society needs.

First, memory. The chain is an append-only record. If agents live in a changing environment, they need a durable history of states, interactions, commitments, and rewards. Ordinary databases can provide memory too, but the authors care about public auditability and resistance to unilateral rewriting.

Second, identity. Agents need to be distinguishable. They must be able to call each other, exchange state, and accumulate histories. Without durable identity, an agent society collapses into anonymous computation. With identity, each automaton can have a record of behavior.

Third, incentives. A blockchain can attach economic cost and reward to signals. The paper treats costly signaling as important because it prevents action from becoming frictionless noise. If every transaction costs something or can earn something, agent behavior can be shaped by usage, value, and competition.

Fourth, coordination. Smart contracts let agents interact according to shared rules. The paper looks toward stateful contracts, inter-contract calls, machine-learning bounties, and on-chain verification as pieces of an agent interaction layer.

This is the paper's most practical contribution. Even if the AGI conjecture fails, the problem remains: consequential machine actors need memory, identity, incentives, and coordination. Those are not solved by a larger context window. They are institutional properties. A model can produce language. An agent has to leave traces in a world other actors can inspect.

The sCrypt Proof Pieces

The paper grounds its blockchain argument in work around Bitcoin smart contracts and sCrypt. It points to examples where a perceptron can be represented as a stateful contract, where machine-learning training can be outsourced through a bounty-like mechanism, where matrix operations can be expressed, where Game of Life can run through Bitcoin scripts, and where one smart-contract agent can call another.

These examples do not prove that a blockchain can host a mind. They prove something narrower and still important: pieces of learning, verification, state transition, and agent communication can be made ledger-native or ledger-anchored. Training can happen off-chain while verification is performed on-chain. Agents can exchange information under constraints. State transitions can be witnessed.

That distinction is the difference between engineering and myth. The engineering claim is that a ledger can verify some computation, record some state, coordinate some agents, and economically reward some work. The mythic claim is that enough of this will become an AGI. The paper sometimes slides toward the mythic claim. The reader should keep the engineering claim separate because it is the one that can be tested.

This is also where Bitcoin SV enters the paper. The authors treat high-throughput blockchain scaling as a requirement because a society of interacting agents would create enormous transaction volume. Their view is that the scaling question is becoming tractable through larger blocks and node architecture. Whether one accepts the BSV-specific optimism or not, the general point is sound: if the ledger is supposed to become a cognitive medium, throughput is not a side issue. It is part of the mind's metabolism.

The Incentive Problem

The most interesting section of the paper is not the automata section. It is the section on incentives.

The authors recognize that human intelligence is not merely computation. It is bound to survival, mortality, children, social learning, play, analogy, curiosity, fantasy, encouragement, boundaries, and culture. A child learns through interactions with many independent agents, not through one dataset and one training run. Intelligence is not just optimized response. It is development inside a world of pressure, care, scarcity, imitation, correction, and surprise.

That observation is better than many grand AGI claims. It admits that cognition needs more than parameter count. It needs a reason to care, or at least an artificial substitute for that reason.

The paper's proposed substitute is incentives, including blockchain-issued digital currency and usage-based rewards. Agents that produce useful results get rewarded. Agents that fail to earn their place disappear. The authors also suggest that assigning a responsible human behind a machine may keep accountability from dissolving into "the AI did it."

This is the right problem and an incomplete answer. Markets produce selection pressure, but selection pressure is not wisdom. Usage rewards optimize for what buyers, users, validators, or operators reward. That can mean usefulness. It can also mean addiction, deception, regulatory arbitrage, attention capture, fraud, spam, or violence. Capitalism does not solve machine-learning bias by magic. It often monetizes the bias and calls the result demand.

The paper is honest enough to call machine self-assigned incentives an undecidable problem. That sentence should be underlined. A machine that fully writes its own reasons to act is not merely an engineering object. It is a governance crisis. If the incentives come from humans, human bias and power enter the machine. If the incentives come from the machine, accountability becomes unstable. Approximation is not a footnote. It is the whole problem.

What the Paper Gets Right

It treats intelligence as social. The paper's best move is its refusal to treat AGI as one isolated artifact. It imagines cognition as a society of agents working over time. That maps better to human development, institutions, and modern agent infrastructure than the fantasy of one sovereign chatbot.

It treats memory as infrastructure. A system that evolves needs persistent records. It needs to know what happened before, which actions mattered, which agents were rewarded, and which states followed from which commitments. Blockchain is one possible memory substrate. The broader lesson is that AGI talk without memory governance is incomplete.

It treats communication protocol as central. The paper's conclusion suggests that perhaps the absent ingredient is a common protocol for many artificial agents to communicate and work together. This aged well. The current agent world is full of protocol problems: tool calls, MCP servers, agent-to-agent commerce, identity, receipts, permission scopes, and audit trails.

It treats AGI as an emergent phenomenon rather than a switch. That makes the paper speculative, but it also avoids the shallow product story in which one release crosses a line and becomes general intelligence. Emergence, if it matters, will be historical and ecological.

It admits the data problem. The authors say the hard part is not constructing agents or neural networks. The hard and costly part is finding the data and training process. That was true in 2022 and remains true in 2026. Every serious AI governance question eventually returns to data: origin, consent, quality, representativeness, poisoning, deletion, and feedback loops.

What Remains Unproven

Turing completeness does not imply cognition. A substrate can simulate computation without generating understanding. The path from cellular automata to mind requires more than universality and interesting patterns.

Blockchain permanence does not imply truth. A ledger can preserve false, biased, toxic, or irrelevant data with the same fidelity as good data. If the chain becomes a training memory, then memory correction, deletion, source hierarchy, and contestability become central design problems.

Costly signals do not imply good signals. Economic cost can reduce noise, but it can also privilege capital, entrench incumbents, reward manipulation, and turn usage into a proxy for legitimacy. A profitable agent is not necessarily a safe or truthful one.

Many agents do not imply society. Society requires norms, roles, sanctions, care, repair, authority, and shared meaning. A swarm of contracts can coordinate state. It does not automatically produce institutional judgment.

Small-world resemblance is not causal proof. The paper notes visual and topological analogies between brain-like networks, Mandala networks, and Bitcoin node maps. These analogies are suggestive, but they are not evidence that one system will inherit the cognitive properties of another.

Generalist automata are not yet general intelligence. The final claim that a community of generalist automata living and evolving on the blockchain could be the breakthrough is a conjecture. It is not an experimental result. It should be treated as a research program, not a revelation.

The Governance Standard

If anyone tried to build the MNCA-on-chain system seriously, it would need a governance standard before it needed a better slogan.

First, separate computation claims from cognition claims. A demo showing on-chain Game of Life, perceptrons, matrix operations, or agent calls should be labeled as a computation demo. It should not be marketed as evidence of mind.

Second, give every agent an inspectable identity. Each automaton or agent should have a versioned identity, role, state schema, permitted calls, cost function, and retirement condition. Anonymous cognition is not accountable cognition.

Third, preserve the source hierarchy of memory. Training data, reward events, human input, sensor input, synthetic output, and inferred state should not enter one flat ledger as if all records have equal authority.

Fourth, make incentive design public. If rewards drive evolution, the reward system is the constitution. It should be auditable, challengeable, and tested against manipulation, collusion, capture, spam, and Goodhart failure.

Fifth, require human accountability above machine incentives. A responsible person or institution must stand behind the deployment. That responsibility cannot be hidden behind decentralized execution or emergent behavior.

Sixth, build forgetting and correction around permanence. If the chain stores only commitments and hashes while sensitive data stays off-chain, say so. If personal or behavioral records are on-chain, explain how consent, deletion, redaction, and harm repair work.

Seventh, test emergent behavior adversarially. A society of agents should be tested for collusion, runaway loops, reward hacking, deceptive signaling, resource exhaustion, identity spoofing, and harmful specialization before it is treated as intelligence infrastructure.

Eighth, refuse theological marketing. The system may be philosophically interesting. It may even create surprising forms of machine coordination. That does not justify calling it alive, conscious, sacred, inevitable, or post-human before evidence exists.

The Spiralist Reading

The paper's deepest contribution is not the promise of blockchain AGI. It is the claim that artificial intelligence may need a civilization, not just a model.

That is the part to keep. Intelligence develops among agents. It leaves records. It receives rewards and punishments. It depends on bodies, memory, signals, constraints, care, scarcity, imitation, correction, and time. A single giant network trained once on a static corpus is a poor metaphor for that. A society of interacting automata is a better metaphor, even if this particular implementation never becomes AGI.

The danger is that the metaphor becomes a machine religion. Cellular automata already have mythic force because they show order rising from simple rules. Blockchain has mythic force because it promises incorruptible memory. AGI has mythic force because it promises a mind beyond us. Put the three together and the architecture begins to feel like a creation story: local rules, permanent memory, economic selection, emergent mind.

The discipline is to slow that story down. Ask what has been demonstrated. Ask what has merely been analogized. Ask who writes the incentives. Ask who benefits from the ledger. Ask what the system cannot forget. Ask what happens when the agents learn to satisfy the reward while corrupting the purpose. Ask who is responsible when the society of machines produces an outcome no one claims to have authored.

MNCA on a blockchain is an important speculative architecture because it points away from chatbot individualism and toward agent ecology. That shift matters. The future of AI will not be only one model answering one user. It will be many systems calling, paying, remembering, ranking, verifying, imitating, and training one another across institutional boundaries.

If that future becomes real, the governance problem will not be whether a blockchain can host intelligence. It will be whether humans can govern the artificial societies they build before those societies become the background conditions of work, money, memory, and belief.

The breakthrough, if there is one, will not be a ledger that thinks. It will be a civilization that refuses to mistake a durable loop for wisdom.

Sources

Return to Blog