How Octopuses Solve Mazes: The Strange Cognition of a Decentralized Animal

The octopus lineage diverged from the vertebrate lineage roughly 600 million years ago, well before the appearance of brains in the form we recognize. What evolved in the cephalopod lineage is the closest thing to an alien intelligence we have direct access to on this planet: a complex nervous system, problem-solving behavior, and apparent learning, all in a body that has no skeleton, no central blood pressure regulation, and a neuroanatomy in which most of the neurons are not in the head.

The numerical anatomy

An adult Octopus vulgaris has roughly 500 million neurons. By way of comparison, a rat has about 200 million; a dog about 2.2 billion; a human about 86 billion. The octopus sits between rat and dog by neuron count, which is interesting on its own but becomes more interesting when you look at the distribution: only about 180 million of those neurons are in the central brain. The remaining 320 million are distributed across the eight arms, with each arm containing its own ganglion and a peripheral nervous system that has been characterized as semi-autonomous.

The architecture is the surprising part. The central brain coordinates and directs but does not micromanage. An arm reaching for a food item is making local decisions about how to wrap around an obstacle, how to grip texture, when to use suckers as chemoreceptors. The 2011 work by Sumbre, Hochner, and colleagues on octopus arm motor control showed that an isolated octopus arm continues to execute reaching behaviors after its connection to the central brain is severed — the motor program is locally encoded.

Problem-solving behavior

The literature on octopus problem-solving is enormous. They solve maze puzzles, open jars from the inside, escape sealed aquaria through impossibly small gaps, and exhibit the kind of trial-and-error learning we recognize as cognition. The 2007 paper by Mather and Anderson documented individual octopuses exhibiting consistent personality traits across encounters: some bold, some shy, some apparently playful. The 2016 escape of Inky from the National Aquarium of New Zealand — through a small gap at the top of the tank, across the floor, and down a 50-meter drainpipe to the sea — is the most-publicized of many similar cases.

The tool-use evidence has accumulated more recently. The veined octopus (Amphioctopus marginatus) carries coconut-shell halves and assembles them into mobile shelters, a behavior documented in detail by Finn, Tregenza, and Norman in a 2009 Current Biology paper. This is among the clearest examples of tool use outside the primate and bird lineages. The shells are not just used in place; they are transported across distances at considerable energetic cost, which is the marker that distinguishes tool use from incidental object interaction.

Pattern recognition and the human-face question

Roland Anderson at the Seattle Aquarium ran a series of experiments in 2010 in which giant Pacific octopuses learned to distinguish between two human caretakers, one who fed them and one who poked them with a brush. After repeated exposures, the octopuses approached the friendly caretaker and either avoided or sprayed water at the unfriendly one. Critical experimental control: the two caretakers wore identical uniforms during the discrimination tests, so the octopus was distinguishing on facial or body features rather than clothing.

The pattern-recognition capacity is more impressive given the visual system. Octopus eyes are excellent at resolution and motion detection but are colorblind in the conventional sense — they have a single photopigment. The 2015 paper by Stubbs and Stubbs proposed that octopuses use chromatic aberration in their eyes to extract color information from the timing differences in focus across wavelengths, but the behavioral evidence is still incomplete. Whatever mechanism they use, it produces visual discrimination that supports complex behavioral learning.

Memory and learning

Octopuses have both short-term and long-term memory, with apparent consolidation between them. Work by Boycott and Young in the 1950s on the vertical lobe of the octopus brain established that this structure is necessary for visual and tactile learning; lesions in the vertical lobe abolish the ability to learn novel discriminations. The vertical lobe is structurally similar to the vertebrate hippocampus despite the 600-million-year divergence — one of the more striking cases of convergent evolution in neuroanatomy.

The learning is fast. A trained octopus can acquire a new visual discrimination in 10-20 trials and retain it for weeks. The 2017 work by Hvorecny et al. on observational learning in Octopus vulgaris showed that naïve octopuses could acquire a discrimination task by watching trained conspecifics, which is the kind of social learning that was thought to be limited to vertebrates with explicit social structure. The octopus lineage is mostly solitary, so the capacity for observational learning is presumably general rather than socially adapted.

The decentralization question

The deepest puzzle is what kind of cognition is happening when most of the nervous system is distributed across the arms. The Sumbre work showed that arms can execute complex motor programs independently. The Godfrey-Smith book Other Minds (2016) argues that the architecture suggests an experience of the world in which "perception and action are not as cleanly separated as in vertebrates" — the arms are not just effectors but participate in the cognition.

This is hard to evaluate empirically. Octopuses do not have a way to report their experience to us, and the conventional behavioral tests (mazes, discriminations, problem-solving) measure outcomes rather than process. What we can say is that the architecture is sufficient to produce flexible, novel behavior in a wide range of circumstances, and that the architecture is sufficiently different from vertebrate architecture that the underlying cognitive process is probably also different in ways we do not yet have a vocabulary to describe.

The conservation question

Octopus populations face significant pressure from overfishing, climate change, and ocean acidification. The 2022 ban on commercial octopus farming (a proposed industry in Spain and Mexico) was driven in part by recognition that conventional farming practices for octopuses produce welfare outcomes that are widely regarded as unacceptable for animals at this level of cognitive complexity. The UK's 2021 Animal Welfare (Sentience) Act explicitly included octopuses on the same legal footing as vertebrates.

The deeper observation

The vertebrate lineage and the cephalopod lineage have arrived at sophisticated cognition through independent evolutionary trajectories separated by hundreds of millions of years. The molecular biology is different (octopuses have unusual RNA editing rates and a relatively small genome with the gene expansions that produced their nervous system tracing to specific lineage-restricted families). The neuroanatomy is different (distributed rather than centralized, with arms participating in cognition rather than just executing motor commands). The behavioral repertoire is recognizably analogous to vertebrate cognition (problem-solving, learning, memory, pattern recognition, tool use) without obviously sharing the underlying mechanism. The species is the best evidence we have that the universe of possible minds is much larger than the small region currently occupied by vertebrates, and the closest opportunity we will ever have to study an intelligence that is not a variation on our own.