An intelligent system cannot process everything in the same place.
Place too much decision-making at the centre and the centre may become slow, overloaded and detached from conditions at the point of action. Distribute too much authority to the edge and the system may react quickly but lose coherence, memory and control over its wider objective.
This is not merely a question about nervous systems.
It appears whenever an organisation decides how much authority to give local teams, an artificial-intelligence system divides work among agents, a robot distributes control across its limbs, or a government decides which problems should be resolved locally and which require central coordination.
The octopus and the raven illuminate two substantially different solutions.
The octopus places remarkable sensory and motor-processing capacity inside its arms, close to the suckers and surfaces encountering the environment. Its central brain remains important, but it does not need to calculate every muscular adjustment of eight highly flexible limbs.
The raven presents a more centrally integrated architecture. Information gathered through vision, movement, memory and social observation can be combined before the bird selects, preserves or sequences an action. Its intelligence is still embodied, but more of the unresolved problem appears to be brought into an integrated planning system before commitment.
The strategic problem is therefore not whether centralisation or distribution is universally superior.
It is where unresolved intelligence should be placed.
The Strategic Question
Should intelligence be distributed close to the point of contact, where local systems can sense and respond rapidly, or concentrated inside a more unified planning centre capable of integrating information across time, components and objectives?
Executive Thesis
The octopus suggests that intelligence should move towards the point of contact when the environment produces large volumes of rapidly changing sensory information, the body has many degrees of freedom, and local adjustments must occur faster than a central controller could specify them.
The raven suggests that intelligence should become more centrally integrated when a problem requires delayed action, comparison among alternatives, inhibition of immediate impulses, social interpretation or coordination across several stages.
Neither animal demonstrates a pure architecture. Octopus arms possess considerable local processing capacity but remain connected to central direction. Ravens integrate information through the brain but still depend on specialised sensory and motor systems distributed through the body.
The deeper mechanism is Decision-Locality Matching:
Place decision authority as close as possible to the freshest relevant information, but no closer than the full consequences of the decision permit.
When the consequences remain local, local intelligence may be sufficient.
When the consequences cross limbs, teams, resources, time periods or ethical boundaries, the decision must be integrated at a wider level.
The most adaptable architecture is therefore likely to combine:
central intent, local resolution and controlled escalation.
This is a StrategizeOS synthesis drawn from the contrasting biological evidence, not a claim that either animal operates according to a formal organisational doctrine. (PMC)
Why These Cases Matter
Both octopuses and ravens can manipulate external objects in ways that have been described as tool use. Yet they approach physical problems with radically different bodies, sensory systems and neural organisations.
The octopus must control eight soft arms without rigid joints. Each arm continuously changes shape while hundreds of suckers encounter chemical and mechanical information along its surface. A command system attempting to calculate every possible arm configuration centrally would face an enormous control problem. Research instead points towards a layered system in which arm-level circuits perform substantial sensorimotor processing while the central brain provides broader organisation. (PMC)
The common raven has a compact vertebrate body, highly developed vision and a densely packed avian forebrain. Ravens are not habitual tool manufacturers in the way New Caledonian crows are, but experimental studies and field observations show that they can solve physical problems, manipulate objects, use tools under some conditions and preserve objects for later actions. (OUP Academic)
This makes the comparison useful, but only within a controlled boundary.
It is not a competition to determine which animal is “more intelligent”. It is an investigation into where each system appears to place the work of adaptation.
Comparison Boundary
Source cases
The octopus side draws principally from:
- motor-control research on the common octopus, Octopus vulgaris;
- chemotactile research on octopus suckers and arm nervous systems;
- and observations of coconut-shell transport by the veined octopus, Amphioctopus marginatus.
The raven side draws principally from:
- behavioural research on the common raven, Corvus corax;
- experimental studies of tool selection, bartering, caching and delayed action;
- and comparative corvid neuroscience, including work conducted on carrion crows.
Evidence from carrion-crow brains is used only as supporting evidence about corvid neural organisation. It is not treated as direct raven neurophysiology.
Unit of analysis
The unit of analysis is the individual animal as an adaptive problem-solving system.
Environmental boundary
The octopus is considered primarily in a submerged, three-dimensional and contact-rich environment in which chemical and tactile information may be detected directly by the arms.
The raven is considered primarily in terrestrial and aerial environments in which visual information, spatial memory, object selection and social observation can influence later action.
In scope
The comparison examines:
- placement of sensory processing;
- motor control;
- local autonomy;
- central integration;
- delayed action;
- tool handling;
- information flow;
- and coordination between immediate contact and wider objectives.
Out of scope
The comparison does not attempt to rank:
- consciousness;
- subjective experience;
- total intelligence;
- moral value;
- evolutionary advancement;
- or the complete behavioural repertoire of either animal.
Outcome boundary
The relevant outcome is adaptive problem-solving: the capacity to convert available information, body structure and environmental opportunity into a useful response.
What the Evidence Shows
The Octopus: Intelligence Near the Point of Contact
An octopus arm is not comparable to a conventional hinged limb.
It is a muscular hydrostat capable of bending, elongating, shortening, twisting and stiffening at many locations. This creates an extremely large movement space. Researchers studying octopus motor control describe neural participation extending from arm nerve cords to the central brain rather than a controller issuing detailed instructions to every muscle from one location. (PMC)
Experiments on arm extension found that important parts of the movement can be generated by peripheral motor programmes. Rather than centrally specifying the position of every point along the arm, the system can activate a more manageable movement pattern and allow local neural and mechanical processes to produce the detailed shape. (Science)
This is an example of embodied intelligence.
The body is not merely equipment receiving instructions from the nervous system. The arm’s material properties, muscular arrangement, sensory surfaces and interaction with water help constrain and produce the movement. Part of the apparent computational problem is resolved through physical structure rather than explicit central calculation.
The suckers deepen this distribution. They do not merely grip. They contain sensory cells involved in touch and chemical detection. Research on octopus chemotactile receptors found specialised peripheral filtering capable of supporting local responses to surfaces encountered by the arm. (ScienceDirect)
This does not mean that every arm is a separate brain.
The popular description of an octopus as possessing “nine brains” obscures the relationship between its central and peripheral systems. The arm nerve cords contain substantial processing capacity, but the arms remain parts of one connected animal.
A 2020 study demonstrated that the central nervous system can use peripheral information about arm movement and touch during learned, directed tasks. The authors concluded that octopus arms have considerable capacity for independent action while remaining subject to central control that supports organised behaviour by the animal as a whole. (PubMed)
The octopus architecture is therefore better described as hierarchically distributed than fully decentralised.
The centre does not disappear. It avoids doing work that can be performed more effectively near the arm.
Octopus Tool Use: Carrying Capability Forward
The veined octopus provides a striking example of object use separated from immediate protection.
Individuals have been observed collecting coconut-shell halves, transporting them across exposed seabed and later arranging them into shelters. The object imposes a temporary movement cost but can create future defensive value in a location where natural cover is limited. (ScienceDirect)
Strategically, the important feature is not the coconut.
It is the separation between:
- acquiring an external capability;
- carrying that capability while it is not immediately useful;
- reaching a later operating position;
- and activating the capability when conditions require it.
However, the behaviour does not by itself prove human-like foresight. It could arise from a specialised behavioural programme, individual learning, environmental affordances or some combination of these.
The permitted conclusion is narrower:
At least some octopuses can transport an object through one phase of activity and convert it into protection during a later phase.
The evidence does not establish where every part of that sequence is represented neurally or whether the octopus experiences the future in a human-like way.
The Raven: Integrating Information Before Commitment
The raven’s adaptive architecture appears more concentrated around integrated perception, memory and action selection.
This does not mean that every choice is made in one anatomical location. It means that raven problem-solving often requires information to be combined across a broader context before the decisive action is selected.
Common ravens can solve multi-stage physical tasks such as suspended-string problems. Early experiments reported that some birds completed the necessary sequence without gradually learning every individual movement through direct reward. Interpretations involving “insight” remain contestable, but the performances demonstrate flexible coordination of perception and action. (OUP Academic)
Ravens are also capable of learning about other individuals as competitors. In caching experiments, they used prior interactions with unfamiliar birds to alter later cache-protection behaviour. This requires more than reacting to the object presently touching the beak. Information about another actor, previous experience and the current resource must be combined. (ScienceDirect)
A broader cognitive test battery found substantial physical and social cognitive abilities in young common ravens, including performance involving quantities, causality, social learning and communication. Such test batteries have methodological limits, but they support the view that raven intelligence is not confined to one narrow behaviour such as caching. (Nature)
Raven Tool Use: Selection Before Opportunity
Common ravens are opportunistic rather than habitual tool users.
A review of raven tool behaviour found a diverse but relatively uncommon collection of object-use cases. One documented captive raven modified and used a feather to access another raven’s food cache. These observations demonstrate capability, but not a species-wide dependence on manufactured tools. (Wiley Online Library)
The strongest planning claim comes from experiments in which ravens learned that a particular tool could operate an apparatus. They were later offered several objects while the apparatus was unavailable. Ravens selected the functional tool and retained it for later use. Related experiments involved exchanging tokens for rewards after delays extending to several hours. (Science)
The strategic sequence differs from the octopus shell case.
The raven must:
- recognise a future task class;
- distinguish a useful object from distractors;
- inhibit immediate alternatives;
- preserve the selected object;
- and apply it when the relevant apparatus becomes available.
This appears to place more of the unresolved problem before contact with the final task.
The octopus carries a physical structure that later becomes shelter.
The raven selects an object according to a learned functional relationship with an apparatus that is not presently available.
The Raven Is Not Simply a Central Computer
The raven side also requires correction.
Birds do not possess a mammalian neocortex, yet parrots and songbirds can contain very high neuronal densities. Research has found that large-brained parrots and corvids may possess forebrain neuron counts comparable to or greater than those of some primates with substantially larger brains. (PubMed)
Neurophysiological work in carrion crows has identified neurons in the nidopallium caudolaterale that represent abstract behavioural rules during flexible decision tasks. The activity was associated with correct rule-guided choices, suggesting an executive integration function analogous in some respects to mammalian prefrontal systems. This is evidence about corvid architecture, but it should not be converted into the claim that one small region is “the raven’s intelligence centre”. (Nature)
The raven still solves problems through an embodied system.
Its eyes determine what can be inspected. Its beak constrains how an object can be grasped. Its feet stabilise or reposition material. Its wings change the spatial field over which information and resources can be acquired.
Central integration does not eliminate the body.
It changes the division of labour between body and brain.
The Central Strategic Contrast
The octopus and raven do not represent reflex versus thought.
Both integrate information. Both act through specialised bodies. Both can adjust behaviour. Both contain central and peripheral neural processes.
The more useful contrast is between two locations of unresolved complexity.
The octopus resolves complexity near contact
The octopus allows substantial sensory filtering, movement production and mechanical adjustment to occur inside or near the arm interacting with the environment.
The central system can specify a broader objective without encoding every local configuration.
Its architecture can be summarised as:
Send broad direction outward and allow the point of contact to resolve the physical details.
The raven resolves more complexity before contact
The raven can compare objects, retain task-relevant information, account for absent opportunities and organise compact motor actions around a selected plan.
Its architecture can be summarised as:
Integrate the wider situation before selecting what should be carried into contact.
The difference is one of control resolution.
The octopus often moves a relatively low-resolution command towards the arm, where local systems convert it into detailed action.
The raven can construct a relatively high-resolution action selection before the beak or feet complete the final physical move.
This contrast is conditional rather than absolute. The octopus also learns and coordinates centrally. The raven also makes continuous local sensorimotor adjustments. (PubMed)
The Mechanism Beneath the Comparison
Decision-Locality Matching
The proposed StrategizeOS mechanism is Decision-Locality Matching.
A decision should be placed at the narrowest operating level that possesses:
- the necessary information;
- sufficient capability;
- a view of the relevant consequences;
- and a safe route for escalation.
This mechanism contains three layers.
Layer One: Local Resolution
Local systems receive authority to handle:
- high-frequency sensory changes;
- continuous physical adjustment;
- minor reversible errors;
- and problems whose consequences remain largely local.
The octopus arm provides the biological source image.
Local resolution reduces the volume of detail that must travel to the centre. It can also shorten the interval between sensing and response.
Layer Two: Central Integration
A wider system retains authority over decisions involving:
- several components;
- competing objectives;
- delayed consequences;
- shared resources;
- strategic sequencing;
- identity and memory;
- or irreversible commitment.
The raven provides the source image for integrating an absent opportunity, a retained object and a later action.
Central integration prevents each local component from acting as though its immediate problem were the entire system’s objective.
Layer Three: Escalation
Local and central intelligence require a bridge.
A local system must know when its operating conditions have exceeded its authority. The centre must know when additional instruction would delay or degrade an otherwise manageable local response.
Escalation should occur when:
- local signals conflict;
- consequences spread beyond the local unit;
- the situation becomes unfamiliar;
- protected resources are threatened;
- a decision becomes difficult to reverse;
- or the local unit no longer understands the wider objective.
The resulting architecture is:
Central intent → local sensing and resolution → exception escalation → central revision.
This is not an executable algorithm. It is a strategic decision procedure requiring context-specific judgement and calibration.
Rival Explanations and Countercases
The comparison remains vulnerable to several rival explanations.
Rival Explanation One: Body Design, Not Intelligence
The octopus may distribute control because its soft body makes central specification impractical.
The raven may centralise more integration because its skeleton, joints and compact effectors reduce the motor-control burden.
Under this explanation, neural placement follows morphology rather than a general strategy of intelligence.
This rival explanation is strong.
It does not destroy the mechanism, because designed systems also possess different bodies, communication networks and processing constraints. It does mean that the architecture cannot be transferred without examining the receiving system’s physical and informational structure. (PMC)
Rival Explanation Two: Ecology Produces the Difference
The octopus and raven occupy different sensory worlds.
A benthic octopus may encounter hidden objects through direct chemical and tactile exploration. A raven can inspect distant objects visually, move across larger areas and observe the actions of competitors.
What appears to be a difference in planning architecture may partly reflect which information is available before physical contact.
The mechanism must therefore include information geometry: where the relevant information becomes visible, reliable and actionable.
Rival Explanation Three: Tool Use Does Not Reveal the Whole Architecture
Carrying a shell or selecting a stick is an observable endpoint.
Many different internal processes could produce that endpoint.
A behaviour classified as tool use does not automatically reveal:
- conscious planning;
- causal understanding;
- where the decision was represented;
- or whether the behaviour was learned, inherited or assembled from simpler associations.
Tool use is evidence of object-mediated problem-solving. It is not a complete map of intelligence.
Rival Explanation Four: Associative Learning May Explain Raven Performance
The raven planning experiments included substantial familiarisation and training. Critics have argued that learned associations, task structure and reinforcement history may explain more of the performance than stronger interpretations involving human-like future planning.
The experiments show that ravens can select and preserve functionally relevant objects under delayed conditions. They do not settle every theoretical question about episodic foresight or subjective representation of the future. (PMC)
Rival Explanation Five: The Octopus Shell Behaviour May Be Specialised
The coconut-shell behaviour may be an evolved or learned solution to exposed soft-substrate environments rather than a general-purpose capacity for planning.
Its strategic importance remains, but its transfer should be based on the observable sequence—acquire, transport and later activate an external capability—not on an unsupported claim that the octopus reasons about the future exactly as humans do.
Countercase: Central Control Inside the Octopus
If octopus arms were genuinely independent decision-makers, central information should be unnecessary for coordinated learned tasks.
Research instead shows communication between peripheral and central systems. This weakens any pure decentralisation model. (PubMed)
Countercase: Embodied Adjustment Inside the Raven
If raven intelligence operated as a complete plan issued from a single centre, detailed interaction with tools should require little ongoing sensory correction.
In reality, object orientation, grip, balance and insertion remain sensorimotor activities. This weakens any pure central-computation model.
The evidence therefore narrows the conclusion:
Octopuses and ravens differ in relative control distribution, not in possessing exclusively distributed or exclusively centralised intelligence.
The Conditional Decision Rule
The strategic rule is:
Place decision authority as close as possible to the freshest relevant information, but no closer than the full consequences of the decision permit.
The following factors help determine placement.
| Operating condition | Favour greater local intelligence | Favour greater central integration |
|---|---|---|
| Information location | Information first appears at the edge | Information must be combined across units |
| Response interval | Immediate adjustment is necessary | Delay is acceptable or useful |
| Problem structure | Continuous and sensorimotor | Discrete, comparative or sequential |
| Consequence range | Mostly local | System-wide or cross-domain |
| Reversibility | Errors are cheap and recoverable | Errors are difficult to reverse |
| Resource use | Resources are locally contained | Units compete for shared resources |
| Environmental stability | Conditions change during contact | Patterns can be represented before action |
| Accountability | Local discretion is acceptable | Legal, ethical or strategic responsibility must remain central |
| Communication | Communication is slow or bandwidth-limited | Communication is reliable and sufficiently fast |
| Objective clarity | Intent can be translated locally | Objectives are ambiguous or contested |
This table is a conceptual decision aid. It is not empirically calibrated and should not be treated as a predictive equation.
When Distributed Intelligence Works
Greater local autonomy becomes useful when the system faces:
- high sensory volume;
- rapidly changing conditions;
- communication delay;
- physically complex action;
- locally observable outcomes;
- and recoverable mistakes.
The centre should define intent and boundaries without attempting to prescribe every movement.
A distributed system also requires competent local units. Delegating decisions to components that lack sensing, capability or judgement does not create octopus-like intelligence. It merely relocates failure.
When Central Integration Works
Greater centralisation becomes useful when:
- several units affect one another;
- decisions compete for scarce resources;
- actions must be sequenced across time;
- the immediate opportunity may be misleading;
- long-term memory changes the correct choice;
- or responsibility cannot safely be fragmented.
The centre must receive sufficiently accurate information. Central authority without sensory access becomes coherent in form but blind in operation.
When the Hybrid Works
A hybrid architecture works when:
- the centre communicates a comprehensible objective;
- local units can interpret that objective;
- authority boundaries are explicit;
- local action generates useful feedback;
- exceptions can be escalated;
- and the centre can revise intent without interrupting every routine action.
The protected relationship is not central command over local obedience.
It is global coherence without local paralysis.
When the Architecture Fails
Distributed Failure
Distributed intelligence can fail through:
- incompatible local decisions;
- duplication of effort;
- hidden drift;
- local optimisation against the wider objective;
- inconsistent standards;
- inability to allocate shared resources;
- and cascading errors that no component can see in full.
A system may become highly responsive at every point while moving nowhere coherent as a whole.
Centralised Failure
Central integration can fail through:
- information overload;
- communication bottlenecks;
- delayed response;
- excessive instruction density;
- loss of local initiative;
- inaccurate abstraction;
- and dependence on one vulnerable decision centre.
A centre can become increasingly confident while receiving increasingly simplified information from the environment.
Hybrid Failure
The hybrid fails when authority is ambiguous.
If local units are punished for acting independently, they will escalate everything. If they are punished for escalating, they will conceal uncertainty. If the centre intervenes unpredictably, local learning becomes impossible.
Central intent and local resolution require a stable contract:
- what may be decided locally;
- what must be reported;
- what must be escalated;
- and what conditions suspend normal autonomy.
Relevant Cross-Domain Transfer
The following applications are StrategizeOS transfers. They are structurally inspired by the animal evidence but are not biologically validated operating laws.
Robotics
A soft robot with many deformable components may benefit from local controllers handling pressure, grip, balance and obstacle contact while a central planner specifies destinations and task priorities.
Sending every sensor reading to one processor may create latency and computational burden. Allowing every limb to choose its own objective may produce conflict.
The structurally relevant transfer is central task selection combined with contact-proximal motor resolution.
Artificial Intelligence
An AI system may distribute narrow decisions to edge agents while retaining central control over:
- shared objectives;
- permissions;
- resource allocation;
- model updates;
- safety constraints;
- and irreversible actions.
Local agents are useful when they possess fresh domain information and can act reversibly. Central review becomes necessary when outputs interact, stakes rise or the system encounters conditions outside local competence.
The raven-like planner should not micromanage every token of activity.
The octopus-like edge should not redefine the mission.
Organisational Design
Customer-facing teams often possess information unavailable to senior management. They may need authority to resolve ordinary cases without waiting for approval.
Central leadership should retain decisions involving:
- institutional identity;
- major capital;
- legal exposure;
- cross-team conflict;
- and long-term positioning.
The decision is not “centralise or decentralise the company”.
It is to identify which classes of uncertainty belong at which level.
Emergency and Field Operations
Field units may need immediate discretion when communication is delayed and local conditions change quickly.
A coordinating centre remains necessary for shared situational awareness, logistics, prioritisation and the allocation of limited resources across locations.
Local speed without global allocation can send every unit towards the most visible problem. Global allocation without local discretion can make units wait while conditions deteriorate.
Product and System Design
A product can resolve routine complexity at the interface while escalating unusual or consequential cases.
This produces an octopus-like contact layer connected to a raven-like integration layer.
The interface handles what it can sense and reverse.
The wider system handles what it must remember, compare and protect.
Limits, Safety and Ethics
The comparison transfers mechanisms, not animal behaviour or biological morality.
It does not imply that:
- decentralisation is naturally superior;
- centralisation is evidence of higher intelligence;
- animal tool use creates a hierarchy of moral worth;
- local actors should be denied wider information;
- or central authorities should retain power merely because they claim broader vision.
Distribution can conceal harm by dividing responsibility. Centralisation can conceal harm by separating decision-makers from the point of impact.
Consequential decisions affecting rights, safety or welfare require identifiable human accountability. They should not be dispersed across agents until no one remains responsible for the outcome.
The biological cases also deserve restraint. Experimental performance does not provide a complete account of animal consciousness, and uncertainty should not be used as permission to disregard animal welfare.
Human judgement remains necessary wherever a system must decide:
- which objectives are legitimate;
- which risks are acceptable;
- who bears the cost of failure;
- and which decisions must never be delegated.
Strategic Summary
Source lesson:
The octopus distributes substantial sensory and motor processing into its arms while retaining central coordination. The raven more visibly integrates information across objects, delays, memories and social situations before selecting certain actions.
Mechanism lesson:
The relevant mechanism is Decision-Locality Matching. Decision authority should follow the location, speed and scope of the uncertainty being resolved.
Decision lesson:
Distribute intelligence when information is local, response must be rapid and errors remain reversible. Centralise integration when decisions cross components, time horizons, shared resources or protected boundaries.
Architecture lesson:
The strongest general design is neither a completely independent edge nor an overloaded centre. It is central intent, local resolution and controlled escalation.
Boundary lesson:
The evidence supports a conditional comparison of control architectures. It does not establish that all octopus intelligence is peripheral, all raven intelligence is central, or either species is universally more intelligent.
Compact Research Basis
- Sumbre, Gutfreund, Fiorito, Flash and Hochner, “Control of Octopus Arm Extension by a Peripheral Motor Program,” Science (2001). (Science)
- Zullo, Sumbre, Agnisola, Flash and Hochner, “Nonsomatotopic Organization of the Higher Motor Centers in Octopus,” Current Biology (2009). (ScienceDirect)
- Gutnick, Zullo, Hochner and Kuba, “Use of Peripheral Sensory Information for Central Nervous Control of Arm Movement by Octopus vulgaris,” Current Biology (2020). (PubMed)
- van Giesen and colleagues, “Molecular Basis of Chemotactile Sensation in Octopus,” Cell (2020). (ScienceDirect)
- Finn, Tregenza and Norman, “Defensive Tool Use in a Coconut-Carrying Octopus,” Current Biology (2009). (ScienceDirect)
- Kabadayi and Osvath, “Ravens Parallel Great Apes in Flexible Planning for Tool-Use and Bartering,” Science(2017). (Science)
- Jacobs and Osvath, “Tool Use and Tooling in Ravens: A Review and Novel Observations,” Ethology (2023). (Wiley Online Library)
- Veit and Nieder, “Abstract Rule Neurons in the Endbrain Support Intelligent Behaviour in Corvid Songbirds,” Nature Communications (2013). (Nature)
- Olkowicz and colleagues, “Birds Have Primate-Like Numbers of Neurons in the Forebrain,” PNAS (2016). (PubMed)
- Pika and colleagues, “Ravens Parallel Great Apes in Physical and Social Cognitive Skills,” Scientific Reports(2020). (Nature)
- Hampton, “Parallel Overinterpretation of Behavior of Apes and Corvids” (2019), and Lind, “What Can Associative Learning Do for Planning?” (2018), as critical examinations of stronger planning interpretations. (PMC)