The world no single sense receives

The phrase “Islamic cognitive science” can mislead if it suggests one settled doctrine shared across a civilisation. The useful object is narrower: a sustained research tradition in Arabic and Persian philosophy, medicine and theology that argued about the powers of the soul. Avicenna offered its most influential architecture. Al-Ghazālī adopted much of the descriptive psychology while challenging what it could prove. Averroes questioned the need for a separate estimative faculty. Abū al-Barakāt and Fakhr al-Dīn al-Rāzī pressed a deeper objection: perhaps one soul makes hybrid judgements, so multiplying faculties only redescribes the work rather than explaining it. Black 1993 Adamson 2023

That disagreement gives us the right starting point. A faculty should not be posited merely because a label is available. Avicenna distinguishes capacities by their proper objects and operations. A power that receives a form differs from one that retains it. A power that preserves a seen shape differs from one that retains the unsensed significance attached to it. A power that combines particulars differs from intellect, whose characteristic objects are universal intelligibles. This is a functional decomposition joined, in Avicenna, to a physiological localisation in the brain’s ventricles. The localisation did not survive. The decomposition remains testable as a way of asking where cognitive work occurs. Ivry 2008 Kaukua 2014

Common sense is a binding operation

Avicenna’s al-ḥiss al-mushtarak, usually translated as common sense, is not prudence and not public consensus. It is the receptive power in which inputs from different external senses can belong to one sensible object. A pomegranate is red to sight, cool to touch, faintly fragrant to smell and tart to taste. No external sense receives the pomegranate as red, cool, fragrant and tart. Yet the animal encounters one thing, not four unrelated streams. Common sense names the operation that makes cross-modal comparison possible. Avicenna 1952 Ivry 2008

Retention is not another act of sensing

Close your eyes after seeing the pomegranate. Its sensible form can remain available even though sight no longer receives it. Avicenna therefore separates reception from retention. Common sense receives the unified sensible form; the retentive imagination preserves it. The distinction seems modest until a task depends on what happened a moment ago. Following an occluded object, comparing two tones, recognising a partly hidden face or finishing a movement all require a present operation over something no longer present to the external sense.

Modern vocabulary can blur this. “Memory” may refer to model weights, a cache, a document store, an episodic record, a recurrent state or an authoritative world model. Avicenna’s distinction is narrower. The retained item is a sensible particular, not a universal proposition and not the disposition created by learning. A useful comparison must preserve that object and its temporal role.

The wolf is dangerous before it becomes a proposition

Avicenna’s most distinctive internal sense is wahm, commonly translated as estimation. His canonical animal example is a sheep encountering a wolf. Colour, shape, movement and sound are sensible forms. Hostility is not red, angular, loud or fast. Yet the sheep does not need to form the universal proposition “wolves are predators” before fleeing. It apprehends this particular as to-be-avoided. Estimation names the capacity that receives such unsensed, particular meanings. Memory, in the paired sense, retains them so a later particular can be encountered under the significance of the earlier event. Black 1993 Black 2000

The word “meaning” can make the view sound linguistic. That is the wrong starting point. The sheep need not possess a sentence. What changes is the action field around the object. A neutral patch of terrain becomes escape route; a shape becomes pursuer; proximity becomes urgent. The meaning is particular because it belongs to this encountered object, not to a universal definition. It is unsensed because no external sense receives hostility in the way sight receives colour. It is agent-relative because the same wolf can mean danger to the sheep, opportunity to a hunter and kin to another wolf.

Why estimation was disputed

Avicenna’s decomposition invites a clean story: external senses receive forms; common sense unifies them; estimation receives intentions or meanings; memory stores those meanings. Yet the most interesting judgements are mixed. “This coloured moving thing is dangerous” binds sensed properties and unsensed significance in one act. Why assign them to different faculties? Deborah Black shows that later philosophers challenged both the logical status of estimative meanings and the need for a separate faculty. Averroes redistributed much of the work. Al-Ghazālī used the psychology while resisting some Avicennian conclusions. Fakhr al-Dīn al-Rāzī often preferred the activity of a unified soul over a proliferation of powers. Black 1993 Adamson 2023

This disagreement supplies a modern break test. If form recognition and appraisal can be altered independently, a functional separation earns support. If every intervention changes both together, the distinction may describe two aspects of one process rather than two causal components. The test does not settle Avicenna’s metaphysics. It tells us whether the decomposition improves explanation in the target system.

Imagination is controlled transformation of particulars

Retention alone cannot explain a winged horse. The system must preserve forms and then combine parts that were not received together. Avicenna therefore distinguishes the store of sensible forms from a compositive power that joins and separates them. In animals, estimation can direct this operation towards action. In humans, the cogitative use of imagination can prepare particulars for intellectual thought. Mehmet Zahit Tiryaki’s reconstruction highlights this active role: perception is not finished when an image arrives; internal activity reorganises the sensible material. Tiryaki 2025

Calling every generative output “imagination” is too weak. Randomly splicing fragments demonstrates combination but not cognitively useful composition. The mechanism matters when a retained scene can be changed along a controlled dimension and the change informs prediction or action. Nadia imagines the flame extinguished after turning the valve. The useful contrast is not real versus fictional. It is received particular versus deliberately transformed particular.

Ablate operations, not historical labels

A modern experiment should not ask whether a system “has common sense” or “has wahm”. Those questions invite impressionistic scoring. Give the system matched tasks, then remove one state or transformation. A binding ablation should impair cross-modal object identity more than single-cue recognition. A retention ablation should spare immediate judgement but damage delayed comparison. A composition ablation should spare recall while damaging controlled scene changes. An appraisal-state swap should change action priority while leaving object description stable.

flowchart LR
  A["The world no single sense receives"] --> B["The wolf is dangerous before<br/>it becomes a proposition"]
  B --> C["The person with no world"]
  C --> D["A disciplined bridge to modern<br/>cognitive systems"]
  D -. "re-test" .-> A
  classDef first fill:#0a1424,stroke:#087A98,stroke-width:2px,color:#e6f3ff
  classDef step fill:#0a1424,stroke:#3f7da0,stroke-width:1px,color:#e6f3ff
  class A first
  class B,C,D step

The person with no world

Avicenna then performs a more radical boundary experiment. Imagine a person created all at once, mature and sound, suspended in empty air. The person’s limbs are separated so no skin touches another part. There is no sight, sound or contact with an external object, and no acquired sensory history. Would this person affirm that they exist?

Avicenna asks the reader to say yes. The person would not affirm a hand, face, heart or brain, because none is sensed. Yet the person would not be absent to themselves. The experiment appears in several forms across Avicenna’s works and has generated a large interpretive dispute. Michael Marmura reads it within Avicenna’s argument for the soul’s incorporeality. Jari Kaukua emphasises its appeal to familiar first-person self-awareness. Contemporary work by Shaun Gallagher treats it as a probe of minimal, pre-reflective self-consciousness rather than reflective self-knowledge. Marmura 1986 Kaukua 2015 Gallagher 2023

Self-awareness is not another internal image

The flying person matters to the internal-sense architecture because it prevents a simple upward pipeline ending in a self-model. Common sense unifies forms. Imagination retains and recombines them. Estimation can mediate an animal’s awareness of its body and activities. Yet Avicenna holds that human self-awareness is direct: no representation stands between the self and its awareness of itself. Ahmed Alwishah’s reconstruction contrasts this with animal self-awareness, which Avicenna treats as bodily, intermittent and mediated by estimation. Alwishah 2016

Whether that contrast is correct is a separate question. Contemporary theories often make minimal selfhood depend on interoception, proprioception, agency or multisensory body models. The flying person deliberately subtracts those contents. It therefore creates a useful disagreement rather than a result. If minimal self-awareness vanishes when embodied regulation is genuinely removed, Avicenna’s directness claim weakens. If a first-person structure remains while body representation is absent or distorted, the claim gains support. Neither outcome alone establishes an immaterial substance.

A disciplined bridge to modern cognitive systems

A historical architecture becomes useful to engineering only after two temptations are refused. The first is dismissal: the ventricles were wrong, therefore the cognitive distinctions are irrelevant. The second is annexation: a familiar modern component is renamed “common sense” or wahm, therefore Avicenna anticipated it. Both moves avoid the harder question. A defensible comparison must preserve a named relation, survive an intervention and state where the mapping fails.

Consider multisensory binding. Modern causal-inference models ask whether two cues arose from one source and predict partial integration when common-cause evidence is uncertain. That supplies a mathematical account Avicenna did not possess. His common sense supplies a different explanatory demand: qualities received by distinct senses must become jointly available as qualities of one sensible particular. The relation preserved across the comparison is not a shared algorithm. It is the need for an operation whose success and failure determine whether one object can be encountered through several modalities. Körding et al. 2007

The same discipline applies to estimation. Contemporary research shows that affect and relevance can alter perception, attention, memory and choice, but this does not turn wahm into an “emotion module”. Avicenna’s claim concerns an unsensed particular meaning such as threatening, friendly or to-be-avoided. A modern appraisal process may preserve the agent-relative relation while implementing it through learned value, bodily regulation, task policy or social history. The mapping breaks if the modern target is merely a scalar reward, a text label or a post-hoc explanation. Brosch et al. 2013

Worked example: an inspection system that must not hide appraisal

The rover first forms a candidate scene: the hot patch, hiss, vapour region and gas reading probably concern one battery cabinet rather than four unrelated causes. It retains the scene across several sensor cycles because any single frame is weak. It generates two controlled variants: the hiss ceases after valve isolation, or the thermal patch continues rising. It appraises the current scene as hazardous to the worker now present, rather than only anomalous in the abstract. A planner proposes withdrawal, zone isolation and human escalation. A deterministic authority kernel checks identity, permitted actions, current world state and the reversibility of each step before execution.

This configuration is not Avicenna’s psychology implemented in silicon. It is a modern system forced to expose distinctions his model makes hard to ignore. The fused scene belongs in authoritative world state. Recent sensor frames belong in an event buffer. Counterfactual variants belong in a scene model. The particular hazard judgement belongs in a typed appraisal record with its inputs and policy basis. The action remains a proposal until authority is checked. When appraisal is buried inside the planner, the system can act for reasons its final explanation never reveals.

The intervention fingerprint matters more than the vocabulary

Suppose two architectures produce the same safe action. In the first, a fused scene feeds a separate appraisal process whose state can be swapped while the scene remains fixed. In the second, a single network maps sensor history directly to action. Calling the first process wahm adds no evidence. What matters is the fingerprint. Hold the scene constant, change only agent condition or remembered significance, and see whether appraisal and action change in the predicted direction. Then hold appraisal constant and alter the scene. Distinct sensitivities support a functional separation. Their absence supports the objection that one integrated process may be doing all the work.

Fakhr al-Dīn al-Rāzī’s challenge therefore becomes an engineering test. Hybrid judgements such as “this coloured shape is harmful to me” may be acts of one unified subject rather than outputs handed between specialised faculties. A recurrent model may encode form, retention and significance in one evolving state. If intervention cannot separate them, we should not force the state into medieval compartments. The Avicennian decomposition earns modern use only where it predicts a dissociation that a unified alternative does not.

The failure mode: salience laundering

This is salience laundering: an unexamined relevance assignment enters upstream, while downstream reasoning supplies a respectable narrative. The recovery is not “ask for a better explanation”. It is to preserve the appraisal record, expose its provenance and expiry, and run a counterfactual removal. The concept is inspired by the distinction between form and particular meaning, but the failure mode is a new practitioner synthesis rather than a claim found in Avicenna.

How to read the audit

A positive result at the strongest level permits one limited move: treat the comparison as a structural-homology candidate worthy of replication. It does not establish a shared substrate, ontology, phenomenology or historical anticipation. A negative result means the mapping should remain a heuristic or be abandoned. It may also reveal that the target architecture entangles operations too tightly for the proposed distinction to be useful.

The audit encodes six assumptions. Source and target objects must be specified independently. A relation must be named rather than inferred from similar language. An intervention must target the proposed operation. A matched rival must be able to produce the same output by another route. The break condition must be explicit. Finally, the direct self-awareness claim is capped as unresolved because no behavioural checkbox converts first-person presence into a third-person engineering property.

from dataclasses import asdict, dataclass
from enum import Enum
from json import dumps
from typing import FrozenSet


class Status(str, Enum):
    LEXICAL = "lexical analogy"
    FUNCTIONAL = "functional correspondence"
    INCOMPLETE = "structural claim incomplete"
    HOMOLOGY = "structural homology candidate"
    UNRESOLVED = "unresolved comparison"


EXPECTED_OPERATION = {
    "common_sense": "unify",
    "retentive_imagination": "retain_form",
    "compositive_imagination": "compose",
    "estimation": "appraise",
    "meaning_memory": "retain_meaning",
    "self_awareness": None,
}


@dataclass(frozen=True)
class ConcordanceClaim:
    source_claim: str
    target_operations: FrozenSet[str]
    source_defined: bool = False
    target_defined: bool = False
    relation_named: bool = False
    intervention: bool = False
    rival_tested: bool = False
    break_stated: bool = False

    def classify(self) -> Status:
        if self.source_claim not in EXPECTED_OPERATION:
            raise ValueError("Unknown historical source structure")
        if self.source_claim == "self_awareness":
            return Status.UNRESOLVED
        expected = EXPECTED_OPERATION[self.source_claim]
        operation_matches = expected in self.target_operations
        definitions = self.source_defined and self.target_defined
        if not operation_matches or not definitions or not self.relation_named:
            return Status.LEXICAL
        if not self.intervention:
            return Status.FUNCTIONAL
        if not (self.rival_tested and self.break_stated):
            return Status.INCOMPLETE
        return Status.HOMOLOGY

    def receipt(self) -> str:
        result = asdict(self)
        result["target_operations"] = sorted(self.target_operations)
        result["status"] = self.classify().value
        result["equivalence_permitted"] = False
        return dumps(result, indent=2)


if __name__ == "__main__":
    example = ConcordanceClaim(
        source_claim="estimation",
        target_operations=frozenset({"appraise"}),
        source_defined=True,
        target_defined=True,
        relation_named=True,
        intervention=True,
        rival_tested=True,
        break_stated=True,
    )
    print(example.receipt())

What the code tests: whether a comparison has progressed from a shared label to defined operations and a discriminating intervention. A homology result licenses replication, not equivalence. An unresolved result for self-awareness records the measurement gap rather than turning self-report into proof.

stateDiagram-v2
  [*] --> Claim
  state "Common sense is a binding<br/>operation" as Claim
  state "Retention is not another act<br/>of sensing" as Test
  state "Why estimation was disputed" as Revision
  Claim --> Test: expose mechanism
  Test --> Decision: evidence survives
  Test --> Revision: rival survives
  Revision --> Claim: narrow or rebuild
  Decision --> [*]

Separate binding, retention, composition and appraisal

The familiar architecture says that perception supplies neutral facts and reasoning decides what they mean. Avicenna’s internal senses make that split difficult to sustain. A creature must already have one sensible object rather than unrelated signals. The object must remain available after it vanishes. Its parts may need to be separated or recombined. Most importantly, it must acquire a particular significance for this creature in this condition. Explicit reasoning inherits a world already organised for possible action.

The decision is therefore not to add five medievally named modules to a modern stack. It is to treat binding, retention, composition and appraisal as separate causal questions before evaluating reasoning. For a cognitive experiment, intervene on each operation and look for selective failures. For an AI system, give each consequential appraisal a typed record, provenance, expiry and counterfactual-removal test. For a philosophical comparison, stop at the level the evidence supports.

The flying person adds a final discipline. It prevents the architecture from quietly identifying every mental fact with an internal representation. Avicenna claims that human self-awareness is not another image passed along the chain. A modern system may model itself, report confidence and describe its own processing. Those are important functions. They remain third-person mechanisms until a measurement bridge connects them to first-person presence. Do not infer a subject from fluent self-reference, and do not infer the absence of a subject from an unfamiliar substrate.

The strongest alternative remains live. Perhaps a unified recurrent process performs all these operations without stable faculty boundaries. Perhaps appraisal is inseparable from perception rather than added after it. Perhaps bodily regulation constitutes the minimal self that the flying person asks us to imagine away. The correct response is not to average the positions. It is to build interventions that make them disagree.

When Nadia crosses the corridor towards the gas valve, her action does not begin with a neutral picture followed by a detached syllogism. A scene has cohered, persisted, mattered and opened a path. That middle is where many intelligent systems succeed, fail and conceal their causes. The practical inheritance of the internal senses is a better place to look.