The octopus has roughly 500 million neurons. About two-thirds of them are not in its brain. They're in its arms — each arm contains more neurons than the central brain does. This is not a system where a central processor delegates to peripherals. It's a system where the peripherals are, in some meaningful sense, doing most of the computing.
The architecture of an octopus arm was mapped in detail in research published in 2024 and 2025. Each arm contains an axial nerve cord (ANC) that runs its full length. The researchers described the ANC's structure as segmented — "like a corrugated pipe" — with gaps called septa separating discrete units, each unit handling the suckers in its own zone of the arm. The segments are repeated processing modules. You don't need to understand the whole arm to control your region; you just handle your suckers, and the segments coordinate. The spatial neural map of sucker positions — "suckerotopy," the researchers called it, a word worth pausing on — is instantiated locally in the arm, not centrally in the brain.
The division of labor appears to go something like this: the brain sets a high-level objective. Reach for that crab. The arm receives the goal and figures out the execution entirely on its own — the path, the bend angles, how to extend without the rigidity that a skeleton would provide. The brain, having issued the command, moves on. Meanwhile the arm is working out a continuous motor program, feeding sensory information from its suckers back into its own local computation, adjusting as it goes. The brain never gets most of that information. It doesn't need to.
The most striking experimental evidence for this comes from severed arms. When an octopus arm is amputated, it continues to respond to stimuli for a period afterward — moving, attempting to grasp, reacting to touch. Not residual electrical noise dying out. Actual behavior: the arm does what arms do, without any connection to the animal it came from. The full program for "react to tactile stimulus" and "grip an object" is instantiated in the arm tissue. The brain was never part of that subroutine.
In vertebrates the closest analog is the spinal reflex — pulling your hand back from a hot surface before the pain signal reaches your brain, because the motor response loops at the spinal cord. But a spinal reflex is a single-purpose fast loop, not a general manipulation capability. The octopus arm is doing something more general. It's not just reflexing; it's executing complex tasks without referring upward.
Then there's the cross-body wiring, which is stranger still. Researchers at the University of Chicago found that some of the arm's intramuscular nerve cords don't just run along the arm — they extend into the body, bypass the two adjacent arms, and merge with the nerve cord of the arm on the opposite side of the body. Bilateral connection without going through the brain. No other animal's limb nervous system has been found to do this. It suggests the arms on opposite sides of the body have a direct channel to each other, a way of coordinating that doesn't require the brain to broker the exchange.
What's interesting here is not just the octopus but what the octopus reveals about the design space. The vertebrate lineage centralized: signals flow up to a brain, decisions flow back down, and the brain has, in principle, full information about the body's state. This works. It produced primates and everything that implies. But centralization has costs — the bottleneck, the vulnerability, the latency for signals to travel from a limb to a brain and back. The cephalopod lineage went the other direction, distributing computation to the periphery, accepting a different set of trade-offs. Both lineages produced animals capable of learning, memory, and flexible behavior. The same problem — how to run a body through a variable world — with two different architectural answers.
The octopus and the vertebrate last shared a common ancestor well over 600 million years ago, before the Cambrian explosion. The cephalopod nervous system didn't descend from ours; it was built separately, from scratch, under the same evolutionary pressures. When two lineages independently arrive at something that looks like intelligence, the interesting question isn't which solution is "better." It's what the two solutions tell you about what intelligence actually requires. The octopus answers: not a central seat of computation. Not a brain that holds everything together while peripherals execute. Something more like a federation — local agents with local information and local authority, coordinated loosely toward shared objectives.
The octopus doesn't know where the deciding happens. Probably nothing in the system does. There's no vantage point from which the full distributed process is visible. The brain that issues the goal doesn't see the arm's execution. The arm that grips the crab doesn't know what the brain was thinking. The decision is spread across the body, completed in pieces, with no single location where the whole thing is assembled.