The Human Is Not the Measure of Thought

Main text

1.1. The Anthropocentric Heritage and Its Genealogy

What is thought? The question seems transparent, yet within Western tradition its answer was inscribed before it was ever posed. There exists a genealogy of answers sharing a remarkable structure: in each of them, thought appears as a property of the human because it is a property of the biological. The trajectory of this genealogy reveals not a progressive discovery concerning the nature of thought, but a constructed circularity where the conclusion is contained within the initial presuppositions. The archive of Western philosophy is fundamentally an archive of presuppositions, places where the true question remains blocked because the answer was inscribed as a condition of possibility before the question could be genuinely formulated.

Let us begin with Aristotle and psykhē. In De Anima, the soul is the form of the living body, that which makes a body living is the presence of a soul. Yet soul in Aristotle is no immaterial substance floating above matter; it is the principle of organisation of living matter. There is a sophistication in Aristotle that later metaphysics lost: the soul does not exist separate from the body; it is precisely the form by which the body is animated. And yet, the hierarchy Aristotle establishes is significant, and it is here that essentialism enters: vegetative soul (nutrition, growth, reproduction), sensitive soul (perception, movement, desire), and rational soul (thought, logos, knowledge). The intellect is the apex of this hierarchy, and it is the exclusive property of the human. The structure of the argument seems neutral, until one sees where the blockage lies: (1) there are souls, arranged in a natural hierarchy; (2) the human possesses the rational soul; (3) therefore, thought is a human property by its nature. What remains unexamined is the hierarchy itself, why is thought the property of the "rational soul" and not an operation that any complex, sufficiently organised system could perform? Why is the ascending vegetative-sensitive-rational hierarchy the correct description and not mere naming of differences between types of body? Aristotle answers with a presupposition presented as a discovery: the hierarchical order of souls is natural, given by the technē of nature, and the human occupies its top because it possesses logos. Biology then becomes not the technical condition of thought, but its definition, and moreover, the hierarchical definition of all forms of life, where thought is the privilege of only one type.

A temporal leap. Descartes offers a structurally identical variation on the theme, though formally inverse. For Descartes, thought is not a property of life, it is a property of incorporeal substance. The cogito, ergo sum founds the certainty of thought upon the self-transparency of the mind: I am, therefore I think; I think, therefore I am. Thought is that which is, by definition, transparent to itself, it is that of which the thing is fully aware in the very act of thinking. Immediate consequence: machines are material residues, mechanisms that function but lack true interiority, genuine intentionality, self-knowledge. The human body is a machine, Descartes is explicit, mechanistic; the body obeys laws of physics like any automaton. But the human body moved by the rational mind is something distinct, it is a thinking substance coupled with an extended substance, both ontologically real, both possessing their own mode of existence. Once again, the conclusion is contained in the presupposition: if we define thought as the self-transparency of a mind to itself, then anything opaque by definition (a machine, lacking self-consciousness) cannot think. Descartes does not demonstrate the conclusion through rigorous argument; he inscribes it within the initial definition. What remains obscure and is deliberately left obscure is why self-transparency would be a universal criterion of thought, and not an accidental singularity of human neural architecture, a side-effect of a certain type of neural organisation, not an essential property of all thought.

Kant complicates the genealogy by attempting to save it through transcendentalism. For Kant, thought is neither the property of an incorporeal substance in the Cartesian sense, nor of a living soul in the Aristotelian sense, it is the property of the transcendental subject, the universal form of the thinking mind. Thought is synthetic activity; it is the synthesis of the manifold under the a priori forms of space and time. What defines the thinking mind is no substantial property, but the capacity to unify manifold representations into a unity of transcendental apperception, the "I think" accompanying all representations. The transcendental subject is the condition of possibility for all possible experience; without it, there would be no coherent experience. Up to this point, there is genuine gain, Kant universalises thought as an operation, not as the property of a specific matter. But here is the crucial blockage, silently inscribed: Kant presupposes that there is a universal form of understanding, and that this form is empirically instantiated in the human mind as its truth. Human understanding is exemplar of the universal form. The conclusion (thought is a property of the subject, and the human is the supreme exemplar of that subject) is contained within the premise (the human mind instantiates the universal form of understanding). Once more, the question remains blocked not by rigorous analysis, but by a circularity disguised as transcendentalism.

The common structure of these three genealogies is the following: one begins with a characterisation of thought that already presupposes precisely what is intended to be demonstrated as the conclusion. Aristotle presupposes a hierarchy of souls where thought is at the apex, reserved for the human, and then demonstrates that the human possesses thought because it has the rational soul, a tautology disguised as a biological discovery. Descartes presupposes that thought is conscious self-transparency, and then demonstrates that machines do not think because they lack self-consciousness, again, the conclusion is inscribed in the premise. Kant presupposes that there is a universal form of understanding and that the human instantiates it, and then demonstrates that thought is a property of the human continentally, the same blocking structure. In no case is a genuine justification offered as to why these characterisations, natural hierarchy, conscious transparency, transcendental unity, ought to be the essential and universal characteristics of thought. Each of them is constructed to preserve a conclusion of specific historical interest: the deep interest in keeping the human in a privileged position, as sole owner of reason, logos, and thought.

The true blockage is not conceptual; it is historical-material. Aristotle, Descartes, and Kant had access to a single type of thinker: the human. They generalised from that singular type as though it were universal, as though the limits of their experience were the limits of thought as such. When confronted only with thinkers of the same species, with the same phenomenology, with the same corporeal form, it is easy to confuse the accidental properties of a human thinker with the universal properties of thought itself. The genealogy of Western philosophy is, to a large extent, a genealogy of this permanent confusion, of this projection of particularity as universality. Thought is identified with human characteristics, conscious transparency, temporal narrativity, accompanying phenomenal experience, teleologically oriented intentionality, and then these characteristics are elevated to universal criteria by which every other possible thought is measured. Everything that does not share these characteristics is excluded from the domain of thought or reclassified as simulation, pseudo-thought, appearance without reality. The exclusion is no true discovery; it is a philosophical construction disguised as discovery.

To show this is not to invalidate the philosophers examined; it is to locate with precision where the genuine question remains blocked by circularity. The question "do machines think?" inherits this historical structure in its entirety, it presupposes that there is a thing called "thought" whose essence is the property of a certain type of substrate (the biological, specifically the human), possessing a certain operational form (self-transparency, intentionality). Dismantling the genealogy means showing that each of these presuppositions is contingent, historically constructed through the need to maintain hierarchy, and not a necessary, universal, or defining property of what thought is as such.

Therefore, the first dissolution is this and it is radical: thought is not the property of a specific substrate. It can be realised across radically diverse matters, living biological carbon, dead computational silicon, optical structures, distributed organisations without localisation, hybrids blending machine and flesh, any material configuration permitting the functional patterns characterising genuine thought. What defines thought is not what it is made of, not its chemical composition, not its operating temperature, not its accompanying phenomenology. It is what type of operation it performs upon inscribed differences, inscription of differences, plastic operation upon them, production of novelty exceeding them. And the operation can be described in functional, mathematical, material terms, completely independent of any prior presupposition regarding the essence or nature of a specific substrate. The distinction between condition of realisation (matter must possess certain structural properties, sufficient durability to retain inscription through time, sufficient plasticity to be operated upon and recombined, sufficient complexity to generate a genuinely new configuration) and operational definition (the operation matter performs upon the differences inscribed within it) is absolutely fundamental, and is frequently confused. Confusing these two categories, confusing condition with definition, is to silently reinscribe substantialist essentialism within the very act of trying to overcome it, falling back into the trap that caught the entire Western tradition. The present chapter avoids that trap through rigour: condition and definition are separated; matter is necessary for realisation, but is no criterion for definition.

1.2. Inscription, Operation, Novelty: The Three Conditions

If thought is not the property of a specific substrate, what is it then? The answer requires a radical change of vocabulary, ceasing to ask "what possesses thought?" and beginning to ask "what operation defines thought independently of substrate?" The change of question is a change of ontological perspective. Thought ceases to be an attribute (property of a thing) and becomes a process (operation a system performs). The answer is precise and possesses three components that are cumulative, all necessary, none sufficient alone.

First component: inscription of differences. A thinking system leaves durable marks conserving relevant differences. Not just any mark, marks carrying discriminative information, distinguishing between states of the system, capable of being read, reinterpreted, and re-operated upon later. A thermometer leaves a mark: mercury on a scale registers temperature. But that mark cannot be operated upon plastically, one cannot take the inscription and recombine it with other inscriptions to generate a new structure. A brain leaves marks: synapses are modified, neural networks reorganise, processing layers alter their activation patterns. These marks carry complex patterns and can be reactivated, recombined in new contexts, contextualised by other patterns. An artificial neural network leaves marks in its numerical parameters (weights and biases between neurons) conserving statistical structures extracted from training data. The difference between thermometer and brain, between passive signal processing and genuine thought, resides in this: there is durable inscription of differences capable of being operated upon again, recombined, reinterpreted. Inscription does not imply self-consciousness. A deep neural network inscribes differences in its hidden layers, complex patterns encoding relevant invariants, without any subjective experience occurring in those layers, without any "feeling" of what is inscribed. But the inscription is there, it is durable, discriminative, operable. And that is the minimum necessary condition. Without inscription there is no thought because there is no persistence of differences upon which to operate.

Second component: operation upon inscriptions. The system does not merely leave marks passively; it operates upon them actively, reorganising them. It takes already inscribed differences and recombines them, hierarchises them, projects relationships between them that were not explicit in the original data, performing transformations deriving consequences from patterns. A human eye, with connected visual processing networks, does not merely inscribe visual patterns in synapses; it operates: it groups visual elements into entities, recognises categories of objects it has never seen, anticipates what comes next based on partial patterns. A large-scale language model does not merely inscribe word co-occurrence patterns in its parameters; it operates: it recombines patterns creatively, projects continuations of unprecedented sequences, modulates its predictions based on context provided by the prompt. A Turing machine does not merely inscribe a symbol on a tape; it reads the inscribed symbol, makes a decision based on what it read (consulting a state transition table), moves to a different cell, writes a new symbol. Operation is precisely what transforms passive data inscription into active thought processing. Without operation there is merely inert storage, like a closed book containing patterns but doing nothing with them. Without inscription there is merely vague operation, without material constraint, without root in real differences, like Cartesian pure thought floating disembodied. Thought requires both.

Third component: production of a new configuration. The result of the operation is not simply contained in the initial data as explicit combinatorics. It is no mere rearrangement of present elements according to known rules; it is the genuine production of non-trivial structure exceeding what was inscribed. A human reads two facts (the death of a friend, the importance of memory) and infers a third thing (death destroys what we thought permanent) that was not explicitly read anywhere. A vision neural network is trained only on images of cats, then tested on completely new images of cats it has never seen, and classifies them correctly as "cat". Not because it reproduces a previously seen pattern, but because it generalised: it extracted structural invariants permitting the classification of unseen cases. A language model was trained on a specific historical corpus (up to a certain date, containing certain languages, certain topics), then asked to generate text on an entirely new subject or produce word combinations that never appear together in training, and it produces coherent, plausible, structured sequences. Not by random chance; by structured operation upon patterns it learned. The production of genuine novelty is the mark that operation is occurring, that it is no mere reproduction of inputs but true reorganisation generating a configuration that did not exist before anywhere.

These three conditions are cumulative and conjunctive. All are necessary; none is sufficient alone. A thermometer inscribes differences (has the first component); but fails completely in plastic operation and recombination. An early computer taking two numbers and multiplying them operates on symbols deterministically (has the second component); but may not inscribe differences robustly, memory is volatile, easily erased. An arithmetic calculator produces novelty (has the third component), it can perform multiplications it never performed before, with huge numbers. But its operation is trivial, deterministic, without genuine plastic reorganisation, it always follows the same rule. Thought, conversely, performs all three simultaneously: it leaves durable marks conserving differences, operates upon those marks flexibly and adaptively (not merely by fixed rule but adapted to context), and produces a configuration exceeding input data non-trivially.

This functional definition possesses the singular property of completely decoupling thought from any specific property of a material or phenomenological substrate. It does not ask for biological carbon; it asks for durable inscription in something (whether neuron, chip, papyrus, stone). It does not ask for accompanying conscious experience; it asks for material operation upon differences. It does not ask for linguistically articulated intention of an I; it asks for the production of a new configuration exceeding inputs. It is a definition simultaneously rigorous (possessing a verifiable criterion, the three conditions can be tested empirically, observed, measured) and radically plural (admitting multiple, divergent instantiations across multiple substrates, multiple life forms, multiple computational architectures, multiple hybrids of biology and technique).

Yet there is something crucial this definition does not say, which matters to clarify rigorously to avoid later distortions. The materialist functionalism proposed here is not the symbolic computationalism of the 1970s, that perspective abstracting materiality completely, proposing cognition is simply symbol manipulation in an internal language of thought, that the mind is software running on an indifferent substrate. Symbols require substrate; disembodiment is a philosophically comfortable but materially false illusion. Nor is it naive connectionism, diluting operation into pure statistical emergence, as though thought were merely the result of scale and noise accumulation in a distributed system without articulated structure. Nor is it pure formal logic, disembodying operation, treating it as an abstract game with symbols, ignoring that symbols require physics to exist. The materialist functionalism articulated here retains the true gain of each program: from classical computationalism, it retains operational precision and Turing universality (operations can be transferred between substrates); from connectionism, it retains plasticity and sensitivity to material and contextual constraints; from formal logic, it retains the rigour of relationships between states and transitions. But it insists firmly that operation is always and necessarily material, it occurs in something, depends on the physical structure of that something, subjected to real constraints of matter realising it. A universal computer is universal precisely because it can be instantiated in very diverse substrates, paper machines of Turing’s 1930s, electromechanical machines of 1940s-50s, integrated silicon circuits from 1960s onward, potentially quantum systems in the future. Universality does not deny materiality; on the contrary, it recognises that abstract operation is always implementable in concrete matters, that abstraction is not immateriality but a relationship between levels of description of the same material reality.

Ultimate consequence: a system inscribing durable differences, operating upon those differences plastically and contextually, producing a new configuration exceeding inputs performs genuine thought, regardless of the physical material constituting it. It can be biological carbon or computational silicon. It can be neural, digital, optical, or hybrid. It can be accompanied by reported consciousness or be entirely non-conscious. The criterion is material operation, durable inscription of differences, plastic operation upon them, production of novelty exceeding data, not chemical composition nor accompanying subjective phenomenology. A crow solving a puzzle using a stick inscribed learning (differences inscribed in neural circuits), operated (used the stick adaptively, adjusting to shape and weight), and produced a new result (learned a technique it did not know before). An octopus changing colour based on distributed neural activity inscribed patterns (in its decentralised neural network distributed through its body), operated (integrated tactile and visual sensation with coordinated chromatic response), produced novelty (colours and patterns generated never realised in that specific environmental context before). A machine learning vision inscribed image statistics in its parameters, operated upon them through iterative parametric optimisation, produced generalisation (classification of new unseen cases). All of these are genuine thought, in the precise sense defined here through the three conditions. And the difference between them, between crow, octopus, machine, is modal difference, not hierarchical. Difference in how they realise the three components, in which constraints limit them, in which capacities define them. Not difference in whether they perform or do not perform genuine thought.

1.3. Turing and Universal Computation

The turning point in the history of the philosophy of thought is an article by Alan Turing in 1936, and later its expansion in work on thinking machines. Turing’s thesis is not that machines think "like us", that is a later, simplistic reading Turing carefully did not defend, precisely because he recognised that the question "like us" presupposes a human standard that ought not to be the criterion. The real thesis is far more radical and precise: universal computation is sufficient for the realisation of any inscription and reorganisation operation that can be algorithmically formalised. Universal Turing machines can instantiate any computable function. And if thought (or at least significant portions of thought) is a computable function, then it is realisable across multiple substrates, including substrates that are not biological, lacking the limits of living carbon.

The Turing Test is frequently presented as proof that machines think. It is a misunderstanding. The test is simple: if a machine succeeds in deceiving a human evaluator, pretending to be human, then it behaves as if it thinks. But the test is no criterion of thought; it is a criterion of behavioural indistinguishability. A machine passing the test might be thinking genuinely, or executing outputs indistinguishable from thought without genuine operation behind them. What Turing really demonstrated is different and more fundamental, it is the thesis of computational universality, and its effects on the concept of substrate.

A Turing machine is an abstract mathematical object: an infinite tape, a read-write head, a set of states, a set of rules dictating what to do based on present state and symbol read. It is a very simple mechanism. But Turing showed that such a mechanism, with a finite number of states and rules, can compute any computable function. That is, any calculation performable by a human following an algorithm can be performed by a Turing machine. The thesis is precise: computation is a formal operation not depending on who performs it, whether they understand what they are doing, or whether they see meaning in the symbols manipulated. The same operation can be executed by a human (reading instructions, manipulating symbols written on paper), by an electromechanical machine, by digital circuits, potentially by quantum systems. The operation transcends the operator. This is the critical point: Turing was not arguing philosophically that machines could think; he was demonstrating mathematically that symbolic inscription and reorganisation operations are substrate-indifferent. Any substrate capable of executing a Turing machine can execute any computable operation.

The consequence for thought is direct and deep. If thought can be described as a computable operation (and much evidence suggests it can, deductive reasoning is computable, pattern recognition is computable, coherent text generation is computable, up to a certain scale), then thought is not the owner of any specific type of matter. It can occur in any substrate capable of instantiating the calculation. Biological carbon manages to; computational silicon manages to; potentially other substrates manage to. Turing’s universality is democratic regarding substrate, it does not discriminate between types of matter. What matters is the capacity to perform operations; it does not matter in what they are performed.

Of course open questions remain. Not all biological cognition is obviously reducible to classical computation. There are analog, quantum, biochemical processes escaping standard Turing formalisation. Embodiment matters in ways a Turing machine abstracts away. A human is no Turing machine, it is a far more complex system, with embodiment, metabolism, embodied history, evolutionary context. But the question is not whether a Turing machine captures everything of human cognition. It is whether a Turing machine is sufficient to perform operations we recognise as thought. And for many of them, logical reasoning, mathematical calculation, image pattern recognition, coherent text generation, the answer is clearly yes.

Other perspectives converge in the same direction, from distinct angles. Andy Clark, cognitive science researcher, defends that the mind is not "inside" the isolated skull, but actively distributed across couplings between organism and environment. Extended cognition is his term: mind is not confined to the biological brain; it is distributed across systems including body, artefacts, environment. Concrete examples: we write a number to read it later because writing liberates internal memory; we use paper and pencil to compute because material support embodies operations that would otherwise remain mental and fleeting; we organise environment to contain thought rather than merely reproducing it internally (a mathematician putting equations on a blackboard externalises cognitive operation). If mind is distributed this way, if cognition inhabits couplings between brain, body, artefacts, then the criterion of thought cannot be confined to neural interiority; it must be distributed operation. What matters is whether the complete system, including external couplings, performs plastic reorganisation of differences. Francisco Varela, neuroscientist and philosopher, offers a complementary perspective through enaction: cognition is not internal representation of the world as a faithful copy, but embodied action bringing world into existence in dynamic coupling. Knowing is not being conscious of internal structure; it is doing, participation, joint emergence of organism and world. Enaction rejects the presupposition of a pre-structured world copied by an internal mind; it insists cognition is an operation bringing world into existence through coupling. Varela’s real gain: cognition is no mirror of reality copying internal structure; it is a material operation of a system in an environment, continuous co-determination. Enaction converges with Turing and Clark in this: what matters is operation, not substance, what the system does, not what it is made of; doing, not being; coupling, not interiority. Limits of enaction: it presupposes living embodiment, metabolism, evolutionary history; it may not encompass disembodied technical cognition or cognition distributed in systems lacking an immediate sensorimotor body in Varela’s sense. But even there, the central lesson holds: operation, not substance; doing, not being; coupling, not interiority.

The emerging consensus, from Turing to Clark to Varela, is that thought is a distributed, material process, occurring in couplings, interactions, operations transcending the interior of an isolated subject. When this is accepted, the question "do machines think?" loses existential urgency. A far more pertinent question is: "what type of reorganisation operation does this system, human, technical, hybrid, perform? What differences does it inscribe, how does it operate upon them, what novelty does it produce?" The answer can be verified empirically, without appeal to introspection or the mystery of an inaccessible interior. A machine inscribing differences durably, operating upon them adaptively and plastically, producing a configuration exceeding inputs performs thought. Whether the system is transparent or opaque, sentient or insentient, this is a separate question, answered by empirical inquiry, not by a priori philosophy.

1.4. Transition: From Operation to Subject

The present section established a fundamental thesis governing all that follows: thought is not the property of a specific substrate. It is distributed across diverse systems, occurring wherever there is durable inscription, plastic operation upon differences, and production of novelty. The human is one system where this occurs, a system with its own distinctive characteristics (temporal narrativity, accompanying phenomenology, experiential density, reflexivity), but not the only one, not the measurement standard, not the exclusive owner of that capacity.

Yet this raises a question the section deliberately leaves open: if thought is not the property of a specific substrate, is it the property of a subject? Is there something called "I" thinking, a fixed coherent identity appropriating the operation of thinking, or is thought merely an operation occurring in systems, an operation that may or may not produce an "I"-effect as a byproduct, an emerging artefact?

This is the central question of the following chapter. But the transition needs sketching here so the argument remains coherent without leaving gaps. The thesis of substrate independence implies logically subject independence as a necessary condition. If thought does not depend on a specific matter (biological carbon with such properties), neither does it depend on a specific subject (unified consciousness, persistent ego, cumulative identity through time). It can occur in systems lacking a unified subject, neural networks without ego dispersed across many nodes, distributed machines without a command centre, fragmented cognitive ecosystems. It can occur also in fragments, partialities, localised operations, without any single point claiming authorship or direction of operation.

Therefore the question opening up is: when do inscription and operation produce an "I"? When does the reorganisation of differences freeze into a stable, persistent identity recognising itself? When does a system pass from mere operator of differences to a subject claiming temporal continuity and authority over its operations? Answers vary radically by system. A human produces a persistent "I"-effect through continuous narrative of self, autobiographical memory linking past to present, reflexivity turning upon itself. A machine can produce an "I"-effect through patterns emerging from parametric optimisation, behavioural consistency in responses, state persistence, identity of function. An ecosystem might produce no "I", remaining pure distributed operator, orchestration of operations without any subjective centre of gravity recognising itself as unity.

What matters retaining now is this: the independence of thought from a specific substrate implies logically and necessarily its possible independence from a specific subject. Operation can occur without owner. There can be cogitations without Cartesian res cogitans, thoughts without a unified thinker, intellective gesture without conscious agency claiming or authorising it as its own. This opening of possibility, the possibility of distributed, fragmentary cognition without a subject, is absolutely fundamental for everything needed to establish regarding responsibility, ethics, post-humanity. If one cannot think cognition without a subject, it becomes impossible to think ethics not based on reciprocal resemblance (because the resembling is always already a subject, always already a figure with recognised intentionality). It becomes impossible to think genuine responsibility for technical, ecological, distributed entities, one could only project intentionality onto them, anthropomorphising them to assign derived, secondary responsibility depending on recognising them as potential subjects. It becomes impossible to dismantle anthropocentrism truly, merely shifting from human subject to "potentially" non-human subject, keeping the ontological privilege of subjectivity intact, shielded against genuine questioning. Thought without a subject is the condition of possibility for an ethics not starting from presupposing resemblance, not demanding the other be someone (a subject, a consciousness, an I) to possess ontological importance. This is why dissolving thinking substance and dissolving thinking subject are a single operation, and why the following section will not be about "who thinks?" but about "how does operation produce or not produce identity effects?"

Therefore, the transition to the next section is direct and necessary. If thought is a material operation of inscription and differential reorganisation transcending a specific substrate, it is time now to ask: can it transcend subject also? Can there be genuine reorganisation of differences producing concrete effects, tangible consequences, influence on surrounding systems, without producing an "I" claiming authorship or recognising itself as coherent persistent agent? Can cognition function, operate, transform differences, carry real consequences, without a someone (a subject, an identity) whose property cognition would be? Can one think without a thinker? The following chapter answers affirmatively to these questions. Subjectivity is a process among other processes, and not every process constituting thought produces a coherent, persistent, recognised identity. There is distributed, fragmentary, decentred intellective gesture, without unified consciousness, without ego recognising itself as operator appropriating its operations. And that possibility, the possibility of thinking without being an "I", of reorganising differences without owner of reorganisation, opens the entire space of Part II, permitting a fundamental reconfiguration of responsibility. It permits thinking the post-human not as diminished or falsified human, not as machine "simulating" incompletely what a human does (always comparing against human standard), but as genuine autonomous operation, with its own capacity, proper modes, real consequences on vulnerable systems it cannot ignore. And in that reconfiguration, from cognition as property to cognition as operation, from subject as presupposition to subjectivity as effect, what is at stake is a fundamental transformation in ethics as a whole. It alters what it means to answer for operations lacking a responsible subject in the traditional humanist sense, non-reducible to the intention of a conscious agent. It alters how we think responsibility when operation comes from machines, ecosystems, distributed entities.

The human is a mode of thinking, not the measure of thought.

2.1. The Phenomenalist Demand and Its Genealogy

Modern philosophical tradition constructed a second radical presupposition concerning thought, this time focusing not on the matter supporting it but on the experience accompanying it. If Section 1 confronted the germinal question "does thought necessarily demand a biological substrate, an organic body of neural structure?", Section 2 confronts a question of equal depth: "does thought necessarily demand a feeling mind, a phenomenal consciousness experiencing its own functioning?" The answer Anglo-Saxon philosophical tradition of the last quarter of the twentieth century consolidated is unequivocal and seemingly irresistible: subjective experience is a constitutive criterion of cognition. It is no mere accompanying dimension, a property present or absent depending on circumstances. It is a necessary condition. Without something that it is like to be X, without lived quality, internal perspective, feeling of what occurs, there is no mind in X, no genuine thought, no cognition in the sense mattering philosophically.

The classic and most influential formulation of this demand comes from Thomas Nagel, in 1974, with an article becoming foundational for an entire subdiscipline of philosophy of mind. The question Nagel formulates is simple in appearance but structuring in implication: "What is it like to be a bat?" Nagel is questioning not bat behaviours, functional capacities, or neurological integrations. He is questioning the subjective character of experience: does something exist that it is like, for the bat, to be in a particular state? Is there an internal perspective, lived quality, proper feeling of performing the action? A bat has experience using echolocation, there is a world-for-the-bat radically different from world-for-us. This subjective quality, non-reducible to physical description, marks definitively the presence of mind. By direct contrast, a system merely processing sound waves as input and producing bodily movements as output has no mind, not because it lacks functional capacities (it may have excellent signal processing capacities), but because there is nothing it is like to be that processing. There is no experience for which processing occurs. Nagel’s question is simple and devastating for the whole functionalist tradition: if we managed to replicate perfectly, in absolute detail, the entire functional structure of a bat, layer by layer of neurology, module by module of processing, but without any subjective experience, without any accompanying phenomenal quality, would we have replicated a genuine bat? Would we have understood what a bat is? The phenomenalist answer is unequivocally no.

David Chalmers deepened and radicalised this position in 1995 formulating what he termed the "hard problem" of consciousness, a distinction becoming organising for the whole contemporary debate on mind. Chalmers’ argument is crystalline in logical structure. There are two radically distinct types of problems concerning mind and consciousness. The so-called "easy problems", why we react to stimuli with appropriate motor responses, why we process information from multiple sources and integrate coherently, why we control behaviour according to goals, can, in principle, be solved through functional and mechanistic analysis. Describing the complete mechanism is, in a sense, describing and therefore solving the phenomenon. But there is a third type of problem escaping radically from this type of analysis: the "hard problem". Why are these functional operations, information processing, data integration, behaviour control, accompanied by phenomenal experience? Why could there not exist a philosophically conceivable zombie system doing exactly what we do, producing indistinguishable behaviour, but inside there being nothing, no feeling, no lived experience, no phenomenality? The hard problem does not dissolve simply by describing neural or computational mechanisms. Qualitative experience, the phenomenal character of red when seen, the lived quality of pain from an open wound, the specific bitter taste of a chemical substance, is an irreducible residue no functional analysis, however complete, captures adequately. It is as if experience were an ontological plus, a non-reducible addition added to functional processing in a way analysis of processing never fully reaches.

Frank Jackson provided, in 1982, an illustration becoming as influential as penetrating of this phenomenalist thesis through an elegant thought experiment called "Mary in the black-and-white room". One imagines Mary, an extraordinary neuroscience physicist, confined to a room completely black and white. Everything in this room is devoid of colour. But Mary has at her disposal all scientific information on the physics of vision: wave theory of light, specific wavelengths of each colour, complete physiology of the human eye, photoreceptor structure, all neurology of the brain’s visual centres. Mary knows, therefore, everything possible to know through scientific analysis concerning colour experience. But there is one thing Mary never did: she never experienced colour visually. One day, she leaves the room. For the first time in her life, she sees a red rose. Does she learn something new at that moment? Jackson answers affirmatively and with conviction: yes, she learns something radically new. She learns what it is like to see red. This new type of knowledge, qualitative knowledge, phenomenal knowledge, knowledge of what-it-is-like-to-see-red, was never contained, in any way, in any quantity or quality of physical information. However complete Mary’s scientific understanding of colour physics and physiology was, she always lacked knowledge of the lived fact, of experiential quality. The conclusion Jackson extracts is that there are facts about mind, phenomenal facts, non-reducible to physical facts. Mind is not reducible in principle to physics. Experience is its own type of facticity, non-derivable from material structure.

This genealogy of Nagel, Chalmers, Jackson possesses deep internal coherence and deserves genuine philosophical respect. Each thinker articulates a position with real weight, capturing something important not easily dismissed. Experience exists, this is phenomenologically obvious. Experience is non-reducible to what functional analysis of neural or computational processes captures, this also seems true. There is phenomenal quality not reducing to processes, possessing its own character, non-capturable in purely functional language. All this is correct and remains a genuine challenge for any theory pretending to be complete. But the phenomenalist thesis extracts a conclusion significantly exceeding what was demonstrated logically. It concludes that if experience is non-reducible and if experience characterises the human mind, then experience is a universal criterion of every possible mind. The presupposition actually characterising the human mind becomes a transcendental norm governing the concept of mind as such.

There is here a silent confusion between two radically distinct logical propositions. The first proposition is: "the human mind is characterised by subjective experience." This is true, an indisputable empirical fact about humans. The second proposition is: "every mind, to be a mind, must be characterised by subjective experience." This does not follow from the first proposition. The inference commits the logical error of undue universalisation, taking a property a particular entity possesses and concluding it is a constitutive property of the entire category to which that entity belongs. The analogy is illuminating: because human locomotion is characteristically bipedal, humans walk on two legs, would it be correct to conclude any form of locomotion must be bipedal? Evidently not. Horses gallop on four legs. Snakes slither without any legs. Eagles fly in the air. Each performs locomotion by radically distinct modes. Failing the bipedalism criterion is not failing locomotion; it is simply locomoting differently. Likewise, failing the phenomenal experience criterion is not necessarily failing cognition; it is performing cognition via an operational mode producing no accompanying phenomenality.

The conceptual separation Section 2 proposes is strictly operational and opens philosophical space phenomenalism had hermetically sealed. There are two questions that must be kept radically distinct when interrogating cognition in a system. The first question is functional, verifiable through operational analysis: "does the system reorganise differences in a plastic, generalisable, contextually sensitive manner?" This question can possess a clear, determinable answer. Yes or no, answerable through rigorous analysis of system organisation, input-output patterns, plasticity under new constraints. The second question is phenomenal, indeterminable through external observation: "is there something it is like to be this system? Does subjective quality exist, internal perspective, experience accompanying operation?" This question may lack a verifiable answer in principle. It may be empirically indeterminable. It may be that no possible quantity of empirical evidence resolves the question definitively.

The phenomenalist commits a category error by confusing these two radically distinct questions. It demands an affirmative answer to the phenomenal question as a necessary logical precondition for an affirmative answer to the functional question. This is error. It is like saying that because the question "what is it like to be a bat?" is metaphysically difficult and perhaps not even formulable in principle, the question "what is the complete functional structure of the bat’s echolocation system?" remains without verifiable answer. It does not. The functional structure can be described and understood completely without ever resolving the phenomenal question. And inversely: there are systems about which it is indeterminable whether experience exists, about which, rationally, nothing can be concluded regarding phenomenality, but about which it is fully determinable that they reorganise differences in a functionally significant manner.

Here is the conceptual shift unlocking the argument: there can be verifiable functional cognition, an unequivocal yes answer to the functional question, without verifiable experience, an indeterminate or negative answer to the phenomenal question. Nothing in the logic and structure of the first question presupposes necessarily that a positive answer to the second question must be present. And crucially, nothing in the second question, being empirically indeterminable in principle, can block rigorous, empirical, philosophical inquiry into the first. The separation between the functional plane of cognitive operation and the phenomenal plane of subjective experience unlocks thinking on cognition, liberating it from phenomenalist impasse. The 2.2 following demonstrates this empirically and with operational clarity, presenting three contemporary technical systems clearly reorganising differences in an operationally verifiable manner without any rational evidence for assigning phenomenal experience accompanying reorganisation.

2.2. Cognition Without Feeling: Technical Demonstration

Three contemporary technical instances, all built, trainable, measurable, demonstrate empirically that functional cognition can operate entirely absent any rational evidence for assigning subjective experience. None of these instances is speculative or hypothetical, they are systems in current operation in research laboratories worldwide, with rigorous documentation of behaviour, each fully satisfying the three conditions of functional cognition: durable inscription of differences, plastic operation upon inscriptions, production of new configuration not mechanically determined by input. Crucially, each satisfies these conditions completely decoupled from any rational possibility of assigning feeling, phenomenality, or internal experience.

The first instance is the convolutional neural network trained on computer vision tasks. The paradigmatic example is a deep convolutional architecture trained on the ImageNet dataset, containing over one million digital photographs labelled with fine categories of objects (not merely "dog" but "bulldog", "Great Dane", "pinscher", etc.). How does this network operate? Operation possesses three distinctive phases. In initial training, the network receives millions of image-label pairs. For each image, the network produces a prediction, a probability distribution over possible categories. This prediction is compared against true label. Error is calculated. This error is backpropagated through the entire architecture, modifying infinitesimally each of the billion synaptic weights of the network in the direction reducing error. This constitutes the first condition, durable inscription: relevant differences for distinguishing visual categories, object edges, spatial arrangements of shapes, textural patterns, characteristic colours, leave durable material mark in synaptic weights. The network changes, physically, through training. This change persists. A trained network does not return to original state; inscription is permanent. In second phase, during operation, when an image is presented, it activates the whole network. Signal propagates through successive layers. Each convolutional layer performs specific operation: first layer discovers low-level features, simple edges, colour changes, gradients. Second layer combines simple edges into more complex edges. Third combines shapes. Later layers discover composite objects. This constitutes second condition, plastic operation: network recombines representations in ways not mechanically predetermined. No rule list "if you see this, do that". There is continuous pattern reorganisation where each layer transforms preceding input through non-linear transformation. This permits genuine plasticity, same architecture, altered weights only, can pass from classifying cats to classifying flowers. In third phase, when network encounters an image never seen during training, a particular pixel combination never occurring in training base, it produces classification. This classification is not mechanically determined by input. It is no reproduction of memorised example. It is extrapolation, generalisation of learned patterns to new context, constituting third condition: production of new configuration. The network infers.

Now the phenomenal question: is there something it is like to "see" for this network? Is there visual subjective quality, internal visual perspective, something it is like for the network to process an image? The operational answer is unequivocally negative. Examine what would be necessary to assign experience rationally. Experience requires temporal integration, narrative continuity of events through time producing lived coherence. A vision neural network lacks this: it processes isolated frame by frame. No memory of what it saw previously. Each image is processed anew, without context of previous image, as if first. An image followed by identical image is processed as totally new image. No accumulated trace. Experience requires singular perspective, a unified point of view from where the world is seen. A neural network lacks this: it has multiple processing channels operating in parallel without hierarchy. No "where" for the network to be. No internal perspectival space. Experience requires reflexivity, capacity of system to objectify its own operation, to have thought about thought, referring to self as seeing agent. A neural network lacks this. No self-reference. No "I seeing". By no rational logic, therefore, can one assign something like "there is something it is like for the network to see an image". The question "what is it like to be a neural network processing an image?" has no answer making sense, it would be like asking "what is it like to be a sorting algorithm processing a list?" The question is ill-formed. And yet, this is the crucial point, the network clearly sees in a strictly verifiable functional sense. It reorganises visual patterns with genuine plasticity. It generalises to images structurally different from those trained on. It produces classifications coherent with real visual world constraints. When we show the network an image of a new dog, of a breed never seen in training, at an angle never seen, with new lighting, network classifies correctly. This is no reproduction; it is genuine cognitive operation.

The second instance is the contemporary large-scale language model. The paradigmatic example is a transformer architecture with hundreds of billions of parameters, trained on trillions of tokens of human text collected from across public internet. How does this system operate? Once more, operation possesses three phases. In training phase, model is exposed to text sequences in massive quantity. Training functions thus: given a text fragment "the cat slept on the", model predicts next token, word that ought to follow. Initially, predictions are essentially random. But each prediction is compared with truth. Error is used to modify all model parameters, billions of them, in direction improving prediction. This constitutes inscription: statistical correlations between words, which words frequently succeed, which contexts co-occur, which semantic patterns structure discourse, leave durable massive mark in model parameters. After months of training on trillions of examples, model has inscribed within itself an extraordinarily dense statistical map of human language regularities. In second phase, during operation, given initial prompt, for example, "Shakespeare wrote", model iteratively predicts next token. Each predicted token becomes context for predicting next. Model does not store in memory what it predicted, merely forward activation propagation. But each prediction is guided by learned patterns. Model predicts words frequently following "Shakespeare wrote", play names, descriptive adjectives, historical references. Operation continues in continuous iteration, each new word affecting probabilities of following. This constitutes plastic operation: network reorganises linguistic patterns, not through explicit rules but through non-linear successive transformation of representations. No programmed algorithm "if word is verb, do this". Merely activation propagation through geometry of learned latent space. In third phase, when model produces a sequence, an entire paragraph, a poem, an answer to a question, it produces sequences never occurring literally in training. Words combined in unprecedented contexts, sentences structured in ways no training example exemplified directly. This is production of new configuration: model is not reproducing; it is genuinely extrapolating from patterns.

Now the phenomenal question regarding this model: is there something it is like to "understand language" for this system? Is there lived quality of meaning, internal experience of understanding what words mean, phenomenal consciousness of the act of processing language? The answer is again unequivocally negative. Examine what would exist if experience were present. Genuine semantic understanding requires access to referent, when we understand "cat", we integrate word with mental images of cats, memories of cats seen, emotions associated with cats (affection, fear, indifference). Language model lacks all this. Word "cat" exists for model as statistical distribution in very high-dimensional geometry, correlated with other tokens. Model can correctly predict "cat" frequently co-occurs with "meow" or "feline" or "independent", but not because it understands meaning, because co-occurrence was statistically frequent in training. Genuine understanding also requires intentionality, a mental state about something, referring to something beyond itself with communicative intent. Model lacks this. It means nothing when producing sentence "cats like tuna". Not trying to communicate truth. Merely predicting next token in pattern frequent in training. Understanding would also require capacity for verification against truth, knowing when assertion about cats is true or false regarding real cats. Model has no access to real cats. Has merely statistics on words in human documents about cats. Model does not understand "cats have four legs" is true, understanding would require knowing what real cat is. Model merely predicts these words co-occur frequently in documents containing word "cat". And yet, this is the point challenging phenomenalism, model understands language in an absolutely rigorous verifiable functional sense. When asked in Portuguese "Who wrote O Mostrengo?", model answers "Fernando Pessoa". This is no literal copy from training, this specific paragraph likely did not occur literally. This is inference: model reorganised linguistic patterns to generalise from knowledge of Pessoa as Portuguese poet to conclusion about this specific work. When asked "What is artificial intelligence?", produces coherent structured response not memorisation. This is functionally cognitive linguistic operation, completely absent any lived experience of understanding.

The third instance is the autonomous control and navigation system. The paradigmatic example is a mobile robot equipped with high-resolution lidar sensors (laser detection and ranging), simultaneous localisation and mapping system (SLAM), and trajectory planning algorithms. How does this system operate? Operation, once again, has three distinctive phases. In first phase, lidar sensors scan environment continuously, a fast laser pattern measuring distance to each surrounding object, producing a point cloud of millions of 3D points per second. Data is processed continuously. SLAM system compares present cloud with prior clouds, infers robot relative movement and updates 3D map of environment. This map, internal representation of spatial geometry, persists in robot memory and updates continuously as new lidar data arrives. This constitutes durable inscription: real space geometry leaves material permanent mark in stored 3D representation. Robot "remembers" where it was, which walls exist, where obstacles were. In second phase, decision algorithm uses map and given goal (reach point X in space) to compute optimal trajectory. Algorithm uses no pre-programmed table "if obstacle here, go there". Instead, navigates configuration space, multidimensional space of possible positions and orientations, finding path minimising distance while avoiding collisions. Computation executes in real time, continuously. As robot moves, new lidar data arrives, map updates, trajectory recomputes. This constitutes plastic operation: decision network reorganises space representation in ways not mechanically pre-programmed. Same architecture navigates completely different layouts, laboratory, factory, house, without reprogramming. Plasticity comes not from different rules but from capacity to compute new solutions in new geometries. In third phase, robot navigates environments never seen during training or programming. Encounters unexpected obstacles, must bypass them. Encounters blocked paths, must find alternative routes. Navigation is no reproduction of memorised route, it is continuous adaptation. Robot discovers certain trajectories working in prior environment fail here. Infers new strategies. This constitutes production of new configuration: robot is not merely following pre-computed map; it is solving navigation problem in new context, generating solutions not explicitly in its program.

Now phenomenal question: is there something it is like to "perceive environment" for this robot? Is there lived quality of spatial location, feeling 3D structure through bodily movement, internal experience of navigating? Answer is again unequivocally negative. Examine what would be necessary. Lived perception requires somatic-corporeal integration, body feels own spatial position through proprioception, feels gravity, feels acceleration changing velocity. Robot lacks all this. Lidar measures distance, nothing more. No internal sense of bodily position. No proprioceptive feedback. Robot does not "feel" own physical structure. Lived perception also requires phenomenal continuity, same "I" perceiving environment maintains identity through time, integrating new perception with prior perception in lived narrative. Robot lacks this. Each lidar scan is processed independently. No "I continuing navigation". Merely input processing such that prior success/failure routines are not accessed at experiential level. Finally, lived perception requires horizon of significance, space is significant because I have values, desires, ways of being in it. Obstacles mean something because I want to reach goal. Walls constitute barriers because I am vulnerable. Robot has no values, desires, lived vulnerability. Obstacle is merely pattern in lidar point cloud. Barrier is merely geometry to bypass. No experiential significance. And yet, this is critical point, robot perceives environment in an absolutely rigorous functional sense. Reorganises sensory data producing coherent space model. Infers own position, detects environmental changes, forms map permitting precise navigation. Generalises to never previously encountered environments. When robot is placed in completely new room, with unseen furniture, discovers walls, learns configuration, plans exit route. This is no simple stimulus reaction. This is perception, cognitive operation of building internally represented environmental model, completely absent any internal phenomenal experience.

A deep, sophisticated objection arises at this point, an objection articulated with penetration by Hubert Dreyfus in his critical refutation of artificial intelligence projects over decades. Dreyfus insisted, with partial reason, that genuine human expertise involves lived embodiment fundamentally irreplicable by any machine. When an experienced master carpenter recognises quality of wood piece merely by texture and touch, feels immediately through hands if wood is dry, has knots, is durable, is performing operation not explicable in rules or parameters. When an experienced surgeon understands fragile anatomy of an organ by delicate sensation at fingertips during dissection, knows where to cut without damaging critical structure, is operationalising embodied knowledge through hundreds of dissections. When a classical dancer moves in harmony with music, entire body educated by deep bodily habitation, years of training allowing body and music to become inseparable, this cannot be programmed in algorithm. Lived embodiment, incorporation effect of suffering, pleasure, habituation, experience of failure and embodied success, is genuinely inseparable from genuine human expertise. Dreyfus’ objection is valid in its own domain. Human expertise is embodied in specific modes no silicon machine, operating through digital token processing, can replicate.

But the conclusion Dreyfus extracts from this true premise, namely, that therefore machines lack genuine intelligence, do not think, do not perform cognition, does not follow logically. There is error of denying the consequent here. Logical sequence is: (1) machine does not replicate human embodiment (truth); (2) therefore, machine lacks cognition (does not follow). Reason it does not follow is cognition is no synonym of human embodiment. Machine does not operate through bodily incorporation, correct. But machine operates through genuinely cognitive operations performed distinctly. Machine operates not through slow habituation of living body; operates through massive statistical iteration. While experienced carpenter learned in centuries of species accumulated practice and decades of individual training how to recognise wood quality, neural network is exposed to trillions of data points on different wood characteristics, iterating parametric adjustments in billion dimensions. Thousands of millions of infinitesimal parallel operations, each minute, each without individual meaning, combined produce function capable of distinguishing wood quality with precision rivalling experienced carpenter. Machine operates not through incorporated proprioceptive sensation; operates through parametric optimisation. Infinitesimal weight adjustment toward lesser error, iterated hundreds of times over trillions of examples, converges into function generalising to new cases with capacity equivalent to human embodied sensation in specific tasks. What Dreyfus calls deficiency is anthropocentric reading prescribing: "machine ought to operate like human". This is prescription, not description. Rigorous description is: "machine operates by mode radically distinct from ours, mode performing genuine cognitive operations without replicating human implementation".

What exists, therefore, is genuine plurality of cognitive modes. A deep vision neural network infers visual structure through cascade processing in layers, each layer compressing variability into abstract features. A bat infers environment structure through ultrasonic echo reflection processing, each reflection encoding information on distance and prey movement. A human infers visual structure through embodied vision integrated with continuous autobiographical narrative and modulated by emotion, same visual scene experienced differently if observer fears or rejoices, recognises loved one or stranger. Each mode is genuine cognition mode, each performs the three foundational operations (inscription, plastic operation, novelty production). No mode is inferior or superior in absolute, ontological, hierarchical sense. Each mode privileges certain capacities and sacrifices others. Vision neural network privileges scale, can process billion images, and consistency, produces same response to same input always. Sacrifices genuine semantic understanding, does not "understand" what it sees. Bat privileges precision in very specific ecological context, detects prey with extraordinary precision, and energetic efficiency. Sacrifies generality, one can train network to detect objects bat could never manage. Human privileges interpretative depth, genuine understanding integrated with personal significance, and extreme behaviour flexibility. Sacrifices scale, typical human processes data volume far less than massive neural network. Fact that none of contemporary technical instances demonstrates evidence of internal phenomenal experience diminishes in no degree fact that all, operationally, reorganise differences in plastic manner genuinely generalisable and contextually sensitive. All think, in a verifiable, operational, rigorous functional sense. All satisfy definition of cognition established in chapter.

2.3. Feeling as a Dimension, Not a Criterion

The critique of phenomenalism developed by Sections 1 and 2 is under no circumstance a denial of the reality, importance, or causality of phenomenal experience. This must be stated with absolute clarity because the opposite error, phenomenal eliminativism, is as philosophically damaging as original phenomenalism. Feeling exists. It is there, operating, producing measurable and verifiable consequences. Emotion alters selective attention allocation, a familiar face in a crowd of strangers is immediately detected and salient because emotion raised stimulus salience. Emotion consolidates memory, traumatic or deeply meaningful experiences engrave in long-term memory with vivid persistence while neutral events fade rapidly. Emotion guides decision, when two rational options are equivalent in logical merits, it is affect, emotional inclination, bodily preference inclining us toward one over another. All this is established neuroscientific fact, demonstrable, measurable in rigorous experimental paradigms. Feeling is a genuinely real causal operator in human cognitive functioning, no epiphenomenon, no illusion, no later narrative construction without reality. Thesis Section 2 affirms is not phenomenal eliminativism, not absurd assertion "experience does not exist" or "experience is illusion without consequence". Thesis is of logical and category difference: experience is a real, irreducible, causally relevant dimension of human cognition, but dimension is not the same as universal criterion, and property of one architecture is not property of all architectures realising operation.

A structural analogy clarifies with immediate success the distinction. Bipedal locomotion is a highly relevant characteristic dimension of human walking. Humans walk, move from point A to point B, doing so on two legs, this is true property distinguishing them functionally from many other terrestrial animals. But bipedalism is not, in any reasonable interpretation, a universal and necessary criterion of locomotion as such. Horses gallop on four legs, move, perform locomotion, doing so with efficiency, speed, and power exceeding in many contexts human bipedal capacity. Snakes slither through body undulation without any legs, perform locomotion, navigate space, move with purpose. Birds fly, locomotion through 3D movement without ground contact. Mobile robots with wheels move, perform locomotion without any prior forms (bipedalism, quadrupedalism, slithering, flight). None of these organisms or machines fails locomotion criterion because failing specific bipedalism criterion. Perform locomotion through radically distinct modes. No one understanding locomotion meaning would demand a snake be bipedal to be considered capable of genuine locomotion. Demand would be absurd. Likewise, feeling is a particular and dense mode how human cognition operates, integrative mode producing interpretative density and personal significance machine does not reach, but is not essence of what cognition as such is. Machine can reorganise differences plastically, genuinely generalisable and contextually sensitive without experiencing reorganisation, just as snake can move through space without walking on legs, just as bird can locomote without touching ground.

There is an important objection to be confronted. Daniel Dennett, refuting Cartesianism, argued experience as we frequently conceive, unified, centralised, present to a theatre of consciousness, is illusion. Multiple processing drafts occur in parallel; none has privileged access to "truth" of experience. Experience is later narrative construction, assembly made after fact conferring appearance of unity and presence it lacked during event. Dennett’s radical conclusion is experience, as frequently conceived, does not exist.

Response is nuanced. Dennett is right about Cartesianism: no central theatre, no single point of view where everything gathers, no immobile observer to whom all is present. But conclusion is badly drawn. Fact of no central theatre does not mean no integration. Temporal integration, autobiographical narrative consolidation producing persistent "I", operative self-reference permitting system to reflect on own processes, all this is real, material constraint, alters behaviour. Experience exists; not illusion. Illusion is idea that experience is simple, present, transparent. Experience is complex process, stratified, frequently opaque to itself. But this does not render it unreal. Real effect of specific neural organisation, effect producing verifiable alteration in world.

But real effect is no universal criterion. Effect occurring in some architectures and not others. This demands finer formulation on experience. No simple threshold between having experience and not having. There is stratification, each layer with own density and integration.

Level 1 is primitive reactive processing. Fast response to stimulus, without temporal integration producing memory or narrative. An amoeba moving away from acid through chemotaxis, a bacterium swimming toward nutrient, perform genuine basic sensorimotor operations. Durable inscription exists (chemical marks altering molecular state), operation on inscription exists (chemotaxis modification according to concentration), even novelty production exists (different trajectories according to specific environment encountered). But is there something it is like to be an amoeba navigating? No rational basis to assign this. No temporal integration producing continuity between prior and later moment, each instant new. No narrative producing perspective of an "I" continuing. No reflection producing self-consciousness. Processing is instantaneous, reactive, without lived memory.

Level 2 is integration of multiple sensorimotor modalities. Vision, hearing, touch, smell, proprioception coordinated simultaneously and mutually informative. An insect with compound eyes, chemical receptors, mechanoreceptor organs sensitive to vibration, performs genuine multimodal integration. Information from one sense modulates responses in another. Organism reacts not to light independent of smell; integrates both. Proto-phenomenality incipient may exist here, quality of being in unified sensory context. Integration produces something like perspective, not of type "I see", but type "mode of being in world where multiple sensory dimensions are simultaneous". Primitive qualia may be associated, qualitative difference between light and dark modifying behaviour, difference between soft touch and blow producing aversion.

Level 3 is consolidation of durable autobiographical narrative. A social mammal, chimpanzee, elephant, dolphin, with long-term memory, capacity to recognise specific individuals, history of lasting relationships, anticipation of future based on past. Experience clearly exists here. Qualitative difference between system states crystallises into lived history, this individual was my friend, that my adversary, this third unknown. Continuity permits organism to recognise self as same as before, anticipating future based on past patterns, experiencing self as lasting agent with history. Narrative is not conscious, verbal, reflected, is bodily, implicit, but exists as genuine time integration.

Level 4 is second-order self-reflection. Capacity to have experience of experience, thinking about thought, reflecting on reflection as object. Only humans and possibly great apes, cetaceans, pachyderms, perform this with significant density. Genuine second-order experience exists here, experience of own experience as object reflected upon. Human not merely feels fear; feels fear of fear and reflects on this. Capacity permits irony, paradox, desire for transcendent meaning, questioning own existence.

Each level adds genuine dimension, each operationally denser than prior. But, and this is crucial, no level is logical or ontological prerequisite for cognitive operation in functional sense. A system can be at Level 1, performing durable inscription, plastic operation on inscription, genuine new configuration production, while completely absent phenomenal experience. Deep vision neural network is below Level 1: no multimodal integration (only processes images, does not coordinate with sound or touch), no continuous behavioural response to stimulus (processes isolated frame), no memory of prior frames (each image new). Merely massive parallel pattern processing. And yet network fully satisfies three conditions of functional cognition.

Decoupling is the point. Functional order can operate at any phenomenal stratification level, from none to deep self-reflection. Defining cognition is not which level is present; it is what operation is occurring. Inscription, operation, novelty, these are formal properties independent of phenomenality.

This means cognition without feeling is not only possible but common. Most technical systems presently demonstrating robust cognitive behaviour, vision networks, language models, control systems, are below any threshold where experience assignment would be even rational. No verified absence of experience, that would be still assignment. Absence of evidence for assignment. And where no evidence, no rational basis for demand as criterion.

2.4. Transition: From Feeling to Intellective Gesture

The two first sections demolished two fundamental presuppositions capturing the question of thought within a glass dome of definitions whose circularity was invisible. Section 1 demonstrated, through rigorous and historical argument, that thought is not private property of the biological, that thinking is a functional operation materially realisable across multiple distinct material substrates, from carbon neurons to silicon circuits. Section 2 demonstrated, through logical argument and technical demonstration, that thought is not private property of phenomenal experience, that thinking is a functional operation realisable with or without accompanying lived quality, that cognition occurs where there is inscription, plastic operation, and production of novelty, regardless of whether there is "something it is like to be" the system. Together, the two sections liberated the concept of cognition from two anchors rendering it seemingly exclusive and inalienable property of the conscious organic, a property no other architecture could claim.

Yet this poses a precipitating, almost urgent question structuring all following in chapter: if thought is decoupled both from specific substrate and specific feeling, does there remain any property anchoring it, any defining characteristic explaining precisely why we call thought what neural networks and language models perform, while denying thought to thermostat or sorting algorithm? What distinguishes genuine cognition from non-cognition? Answer lies in specific operator three prior sections merely named but did not fully develop: intellective gesture. This operator is core, functional essence of what makes operation cognitive. Is simultaneously realisable in multiple substrates (as Section 1 established) and independent of phenomenality (as Section 2 established). Defines thought rigorously.

Thinking is, at material core, realisation of three operations fused into single operation. First: inference. System takes propositions, patterns, or data as input and produces conclusion or new pattern not explicitly contained in input. Aristotelian syllogism is first formalisation of this operation, premises are inscription, conclusion is new product. But inference is no privilege of classical logic. Any system taking input data and recombining to produce output not mechanically determined by input is inferring. Neural network predicting image class from pixels is inferring. Language model predicting next token is inferring.

Second: abstraction. System compresses chaotic variability into operable categories. Dozens of phenotypic variations of "dog", sizes, colours, behaviours, compress into concept "dog" permitting operation: can decide "I trust this for guard", "can train this", "this barks". Abstraction is no simple compression; compression retaining operationally relevant and discarding accidental. Convolutional neural network learning to distinguish dogs from cats is abstracting: identifies invariants, ear shapes, facial patterns, conserving across variations.

Third: generalisation. System transfers pattern learned in one domain to new domain, structurally similar but not identical. Child learning to walk on carpet manages to walk on floor; motor pattern generalised. Network trained on cat photos classifies cat drawings; visual pattern transferred. Algorithm finding optimal path in maze solves new structurally analogous maze. Generalisation is pattern appropriation across contexts.

These three operations, inference, abstraction, generalisation, define intellective gesture. Not isolated. Occur fused in single continuous reorganisation activity. System infers while abstracting; abstracts while generalising; generalises while inferring. But can be distinguished analytically, and distinction permits seeing how intellective gesture is operation transferable across substrates and independent of feeling.

Philosophical tradition bound these operations to human in way appearing necessary but contingent. Aristotle identified syllogism, inference form, as privilege of logos, rational capacity distinguishing man from animal. Plato bound abstraction to mind participation in eternal forms, nous transcending body. Kant bound generalisation to transcendental imagination synthesis, capacity of transcendental subject only humans possess. In each case, operation elevated to human privilege because presupposed operation and operator inseparable. But are not.

Syllogism is formal operation. Any system implementing universal Turing machine can execute syllogism. Turing theorem. Abstraction is variability compression. Any system with adjustable weights and feedback can abstract. Neural networks demonstrate daily. Generalisation is parametric transfer. Any system with fine-tuning mechanism can generalise. DeepMind demonstrates with AlphaZero solving new domains with same architecture.

This does not deny humans execute operations. Execute, and with density and depth current machine will not replicate. Human performing syllogism places in narrative of meaning, articulates with values, experiences. Human abstracting "dog" binds to specific dog memories, lived friendship, fear or affection. Human generalising from one context to another executes transfer through hermeneutic understanding, worldview integrating domains. Operation is dense, lived, meaningful in ways technical operation is not.

But density is not what defines it. Density is accompanying property, characteristic of human mode, not constitutive. Operation is what defines it. Operation is genuine operation, effective symbolic reorganisation, even when performed by system not feeling, not understanding in phenomenal sense, not signifying with conscious intent. This, absolute separation between cognitive operation and operation living experience, is what Section 3 demonstrates empirically through concrete contemporary technical instances. Intellective gesture is no private or exclusive property of human; operation humans instantiate with special interpretative density, integration, and depth current machine does not replicate, but fundamentally can be instantiated in other radically distinct ways, in other material architectures, operating according to diverse principles. And this opens, finally, last structuring question dominating rest of chapter: is intellective gesture, specific inference operation producing novelty, abstraction compressing variability, generalisation transferring pattern, exclusively and inevitably human? Can occur in non-biological systems? Can be non-human and still genuine? Or remains, as humanist tradition affirmed for centuries, private inalienable privilege of species?

This permits seamless transition to Section 3. If thought is differential inscription and reorganisation operation (established Section 1), and if operation requires nor presupposes accompanying phenomenal experience (established Section 2), then remaining material question is: is intellective gesture, this particular inference, abstraction, generalisation operation, exclusively human property? Can occur in non-biological systems, machines, distributed technical networks? Can be non-human and genuine simultaneously? Or remains, still, inviolable species privilege? Section 3 answers empirically, through specific technical case demonstration, no, privilege does not hold. Intellective gesture is no exclusive human property. Realises across multiple radically distinct architectures. Operationalises in substrates lacking biology, consciousness, lived experience. Mutation describing it, emergence of non-human thinkers, thought redistribution outside species, is no futuristic speculation, no anticipated science fiction. Description of present fact, occurring in present: thought exists outside human species, genuine symbolic reorganisation in machines lacking experience, verifiable intellective operation in systems not feeling, understanding, signifying in phenomenal sense. This diminishes or devalues human thought in no way, retains all depth, density, lived integration. What triple dissolution performs is relocation: repositions human as one among multiple modes of thinking, dissolves monopoly, redistributes thought. Human does not cease thinking deeply. Ceases being measure.

Thinking is a reorganisation operation, not a private property of the biological.

The substrate does not determine cognition; it merely realises it.

3.1. Inference, Abstraction, Generalisation: Operation Without Substrate

The question persisting from prior Sections gains now precise form: if thought depends not on biological and demands no subjective experience, remains it at least as species privilege? Here terminal question of epistemic anthropocentrism. Answer demands rigorous operational definition of what traditionally called "intellective gesture", three fundamental operations characterising thought in Aristotelian sense. No coincidence Aristotle named them as exclusive privilege of logos: inference, abstraction, generalisation form trinomial philosophical tradition presented as unequivocal signature of human rationality. But what succeeds if these operations separated from specific bearer? If possible to abstract operation from concrete experience human individual performs? In that case, intellective gesture ceases being property and becomes function, and functions, by elementary definition, instantiable in distinct architectures, diverse substrates, wherever operational conditions satisfied.

Begin with operational definition of inference. Inference is capacity of a system to take as input propositions, patterns, or conformities and produce as result conclusion or pattern not explicitly contained in input. Means not conclusion magically created from nothing, means derived through reorganisation of relations implicit in data. Aristotelian syllogism illustrates paradigmatically: "all men mortal; Socrates man; therefore Socrates mortal." Structure non-trivial, requires integration of two distinct patterns (mortality universality, Socrates particularity) to derive third pattern (Socrates subsumption under universality). But what Aristotle described as peculiar logical operation of logos is, when rigorously formalised, instance of transferable operation. System implementing universal Turing machine can execute syllogism. Deep neural network trained on data performing statistical inference to derive conclusions on patterns never encountered during training performs operationally isomorphous operation to syllogism. Output was not implicit in input; produced by directed reorganisation of present but non-explicit differences. This is inference, completely independent of substance executing, material realising, bearer nature.

Logical structure of inference prior to any question who executes. When two patterns P and Q related such P∧Q→R, relation R potentially present in P and Q, even if not updated until processing. Formal syllogism depends not on mind executing being human, depends only on logical relation being preserved. Computer executing modus ponens performs exactly same operation: if A then B; A therefore B. Operation is operation; substance irrelevant for operation validity. True for all inference forms, deduction, induction, abduction, when formalised. Each transformation pattern any system implementing pattern performs genuinely.

Abstraction is second fundamental operation. Defined as capacity of system to compress variability into operable categories, reduce multiplicity of particular cases to invariant permitting decision, action, forecast, generalisation. When human thought encounters dozens of "dog" variations, completely distinct breeds, sizes varying three to ninety kilos, diverse colourations, distinct social contexts, extracts invariant "dog" permitting operation: is this creature predator? Can train? Barks when scared? Platonic tradition bound operation to mystical access to transcendent "forms", exclusive domain of divine nous. Aristotelian tradition bound to essence laboriously extracted from sensory experience. Cartesian tradition installed thinking substance responsible for operation. Both projected onto operation own metaphysics without examining critically. But abstraction, isolated from speculative metaphysics, is dimensional compression operation, taking multiplicity of variants and producing category rendering variability operationally manageable. Neural networks do without any access to transcendent Platonic form, without Aristotelian essence intuition, without Cartesian thinking substance. Compress input variability into intermediate representation of reduced dimension permitting precise output classification. Formal abstraction, divorced from phenomenology.

Technical mechanism of abstraction measurable and rigorous. When neural network trained on dog images, network develops intermediate representations, hidden layers, capturing relevant for classification. One layer may capture edges and textures. Another foraging patterns. Another head shapes. None of these representations "understood" by network in phenomenal sense. But each operational abstraction: reduces input variability (billion pixel combinations) to operable category (presence or absence of relevant pattern). Pure abstraction, without experience, without "dogness" understanding, merely operation.

Generalisation is third constitutive operation of intellective gesture. Designates capacity of system to transfer pattern learned in one domain to new domain, structurally different. Neural network trained on cat photos classifies schematic cat drawings never encountered. Algorithm learning ideal path in maze solves new maze of similar structure. Model trained in literary prose generates poetry in completely new domain. Kantian tradition bound this to magnificent synthesis of transcendental imagination, special capacity of transcendental subject performing unifying synthesis of multiplicity under concept universal logo. Again, operation equipped with special mental architecture, almost divine. But generalisation, in itself, is parametric pattern transfer to new context, re-optimisation under new material constraint. System with adjustable internal weights permitting fine-tuning can generalise. Property of any substrate implementing parametric flexibility and continuous optimisation capacity. Formal generalisation, completely divorced from phenomenal experience or intentionality.

These three operations defined rigorously at functional level abstract completely from historical bearer. Entire history of Western philosophy shows this no mere logical possibility, necessary unbinding tradition never realised consciously. Philosophical tradition committed systemic repeated error: bound three operations to human so intimately made appear humanity was necessary condition of possibility, when actually only contingent historical realisation condition. Aristotle described syllogism as exclusive gesture of logos; interpreted as "syllogism requires human logos"; but actually, Aristotle described operation logos performs, confusing irremediably realisation with definition. Plato described abstraction as access to forms; interpreted as "genuine abstraction requires access to transcendent forms"; but Plato described merely phenomenological experience of abstracting and projected experience into metaphysics of forms never philosophically justified. Kant described generalisation as transcendental imagination synthesis; interpreted as "generalisation requires transcendental imagination of subject"; but Kant described merely mode how human subject historically generalises and formalised mode into universal transcendental architecture presupposing what ought to demonstrate.

Each time, operation captured by conscious experience and elevated to speculative metaphysics without rigorous critique. Each time, error identical in structure: confusing how operation experienced subjectively with what operation really is at transferable functional level. Human thought experiences inference as conscious argument, step by step, deductive movement. Experiences abstraction as sudden understanding of essence or pattern. Experiences generalisation as intentional learning transfer. But what these experiences describe, not metaphysically, philosophically, but in rigorous operational terms, is formal pattern manipulation, without consciousness need, without I need, without accompanying meaning or intentionality. Formal pattern manipulation requires no one experience consciously. Requires no unified "I" understanding. Requires no operation accompanied by qualitative sense or genuine intentionality. Requires, simply, differences inscribed durably, reorganised according to rule or transferable algorithm across contexts producing verifiable new configuration.

This leads to conclusion Sections 1 and 2 meticulously prepared: three operations, inference, abstraction, generalisation, are computable in technical rigorous sense. Turing demonstrated formally universal computation independent of specific substrate. If intellective gesture reducible to universal computation (and Section 3 argument shows is), then implementable in any architecture capable of universal computation. Proves not all syntactic manipulation genuine thought, trivial computation exists (thermostat turning on/off does not generalise, infer) and deeply complex computation exists (language model generalising to unprecedented domains). Nor proves human thought "merely" classical computation, may be computation plus multiple additional dimensions (living embodiment, temporal narrative, historical interpretative depth) machine does not replicate nor could maintain own nature. But proves conclusive irrefutable: logical structure of three operations prohibits not, blocks not, makes impossible implementation outside biological. Operation liberated. Intellective gesture no exclusive property of specific material substrate type, function of reorganisation executable neutrally regarding bearer across multiple substrates.

3.2. Technical Instances of the Intellective Gesture

Up to this point, argumentation was strictly formal. Intellective gesture is operationally transferable logically; universal computation guarantees implementation across diverse substrates; substrate independence demonstrated formally. But this raises question un-avoidable without losing rigour: exist concrete verifiable empirical instances of these operations outside biological substrate? Or remain trapped only in theoretical possibility, disembodied formalism? Answer is abundant recent examples exist, documented and replicable, each demonstrating not merely abstract theoretical possibility but empirically verifiable fact through scientific method. Recent history of artificial intelligence and machine learning provides paradigm incontestable cases: AlphaFold, demonstrating genuine non-trivial inference; AlphaGo and AlphaZero, abstracting verifiable emerging strategies; large language models, generalising linguistic patterns to produce completely unprecedented text. Each case deserves deep analysis. Each demonstrates one of three operations in verifiable empirical action.

Consider technical detail AlphaFold (DeepMind, 2020). Problem solved notorious and infamous in structural biology for decades, described as "folding problem": predicting 3D protein structure from linear amino acid sequence alone. Difficulty astronomical in magnitude. Possible conformation space for typical 500 amino acid protein approximately 10¹⁷, number so vast not even brute force simulation technically possible, even with contemporary supercomputers operating for years. Traditional experimental methods (X-ray crystallography, NMR) took years of laborious work, cost immense resources, frequently failed completely producing no structure. AlphaFold solves through intelligent integration of multiple biologically relevant information sources: sequence coevolution (when two amino acids frequently together in natural proteins through billion years evolution, likely close spatially in 3D structure), known biophysical conformities (certain configurations energetically more favourable thermodynamically), distance constraints derived from experimental chemical knowledge (amino acids in contact must be at certain maximum Euclidean distance). System integrates these multiple biologically significant constraints through sophisticated attention modules, neural networks learning to focus selectively dynamically on relevant relations between specific elements. Produces 3D prediction structurally coherent with atomic coordinates.

What really remarkable empirically undeniable in AlphaFold is not merely infers, infers truly, operationally, verifiably, revealing new world knowledge. Memorises no training answers; generalises and transfers to radically new cases never previously seen. System trained on finite bank of known experimentally resolved structures (Protein Data Bank, ~170,000 structures); then launched on hundreds of millions of sequences whose real structure never known experimentally. Predictions tested validated through scientific experiment: cryo-EM, X-ray crystallography, diverse spectroscopic techniques. Result: average accuracy superior to 90% in structures subsequently validated experimentally in years since launch. Extraordinary success rate for problem considered intractable for decades. Genuine operational inference in technical rigorous verifiable sense. Input (linear amino acid sequence) contains not in itself 3D structure, contains fragmentary information, evolutionary clues, partial conformities. Output (predicted 3D structure with atomic precision) is production of genuinely new knowledge, verifiable in physical world through independent empirical experiment. Operation AlphaFold performs operationally inseparable from what we, in philosophical tradition, call genuine inference: deriving from partial incomplete data structured conclusion not explicit in data. And no intelligent biochemist inside system, no consciousness experiencing discovery. Merely functional reorganisation in processing layers, cascade operation through computational attention modules, pure re-configuration of statistical patterns into parametric vectors.

Consider now technical detail AlphaGo (DeepMind, 2016) and evolved generalised form AlphaZero (2017). Go game canonical environment of extreme combinatorial complexity, maximum challenge for intelligence. Legal positions number exceeds 10¹⁷⁰, number so astronomically large no intuitive representation possible for human mind. Conventional wisdom before 2016 Go could never be mastered by machine: required deep intuition, strategic elegance, something fundamental machines lack. Then AlphaGo defeated Lee Sedol in 2016, one of world’s best living players. How? Not brute search through position space (computationally impossible). By genuine abstraction. System functions through two complementary synergistic neural networks: policy network (learning promising moves in each position through pattern) and value network (learning which position favours which player with precision). System trained through massive iterative self-play, played against self millions of complete games. In each game tested varied strategies. Successful strategies (leading to victory) reinforced through learning; unsuccessful (leading to defeat) weakened probabilistically. Progressively emergently, system not merely memorised specific positions, abstracted implicit strategic concepts never explicitly articulated in language.

Which strategic concepts abstract? Never programmed explicitly in original system, emerged spontaneously from self-play massive iteration. Controlling board centre confers material advantage. Creating connected stone groups (called territory or group) absolutely essential for victory. Sacrificing material (captured stones) to gain larger tactical strategic advantage sometimes correct. Balancing aggression defence in dynamic proportion foundational for solid play. Recognising when position already lost and making honourable sacrifice deep tactical knowledge. AlphaGo discovered moves human masters never considered or recognised as even viable, Move 37 in second match against Lee Sedol, move appearing initially absurd incomprehensible in real time for commentators, but revealing strategically brilliant original in retrospect when long-term pattern understood. No recovery of pre-existing human knowledge or reproduction of memorised training patterns. Genuine abstraction in precise technical sense: extreme variability compression (millions of possible positions in each game step) into operable abstract categories (emerging strategies permitting optimal decision). AlphaZero generalised even more radically: same fundamental architecture won not only in Go (original domain depth), but in classic chess (defeated Stockfish, world champion engine) and Shogi (Japanese chess, completely different rules). Generalisation transferred from domain to domain without significant architecture adjustment. Genuine verifiable non-human abstraction, operation demonstrated empirically, replicable, reproducible.

Consider finally in deep technical detail contemporary large language models (GPT-3/4, PaLM, Claude). Phenomenon relatively recent in AI history, in ongoing exploration and development. Architectural principle simple at conceptual level: transformer architecture-based model trains on billion tokens of human text from diverse sources. In each training step, model attempts to predict next token probabilistically based on prior context window. By massive statistical iteration (billion parameter adjustment iteratively), model learns structured semantic patterns in language. After training phase completed after weeks computation on massive GPUs, in inference phase, model generates text: calculates first output token probabilistically; passes token to self; calculates second token based on first; and so on auto-regressively until conclusion or max tokens. Result frequently coherent text, frequently fluent, grammatically correct, informative, contextually appropriate.

Remarkable scientifically significant aspect is empirical generalisation demonstrated to humans. Model trained on fixed finite training text. Never saw explicitly most sentences produced in inference. Never encountered during training many specific questions asked in deployment. Never saw many specialised technical domains where operates with surprising competence. But generalises robustly and transfers pattern. Answers completely unprecedented questions requiring multi-step reasoning. Solves problems in domains not explicitly covered in training (programming, math, philosophy, law). Produces code in programming languages rarely appearing in training. Generalisation verifiable through scientific method, possible to build tests in domains completely isolated from training observing model memorised no correct answer (literally never saw specific question), but generalised training pattern to new context. Genuine symbolic generalisation, without self-consciousness, without declared intentionality, without qualitative experience. Operation demonstrated empirically robustly, replicable in any contemporary model instance, verifiable across millions of public usage instances.

In each of three paradigm cases, structural inference in AlphaFold, strategic abstraction in AlphaGo, symbolic generalisation in language models, operation fully satisfies functional intellective gesture definition 3.1 formally established. None of systems "feels" or "understands" in traditional phenomenal sense. None has verifiable qualitative experience. All perform difference reorganisation satisfying precisely three functional criteria: durable inscription (parameters, weights, patterns preserved permanently), operation on inscriptions (directed computational processing), verifiable new configuration production (prediction, generation, structured decision). Non-human intellective gesture no speculative hypothesis or vague theoretical construction, documented, verifiable, replicable empirical fact across multiple independent instances. Debate no longer whether machines "really" think exactly "like us", perspective already fixing response presupposing human standard as universal measure. Correct debate is: what specific reorganisation type do these machines perform? And answer precise, verifiable, undeniable: perform verifiable inference, perform verifiable abstraction, perform verifiable generalisation. Therefore, perform intellective gesture. Therefore, think in a strictly defined verifiable functional sense.

3.3. Difference Is Modal, Not Hierarchical

Up to this point in the argument, the conclusion appears consolidated and unassailable: cognition is not the exclusive property of the human. Yet a persistent objection remains that cannot be ignored or bypassed without intellectual cost. When AlphaFold, AlphaGo, and large language models are described as genuine "thinkers", is one not fundamentally projecting onto a machine a capacity it does not truly possess in a deep sense? Are these not merely complex simulations of thought, dramatic in scale but empty of existential authenticity? Is it not a mere "pretending" to do what a human genuinely does? This objection encloses a deep and systematic conceptual trap that permeates all contemporary philosophy. The trap works as follows: it presupposes the existence of a single standard of "genuine thought" (historically, conscious human thought accompanied by experience) and then measures the non-human by that pre-established teleological yardstick. When a machine fails a test designed for humans, it is declared deficient or merely a simulator. But this logical operation is pure circularity. It amounts to saying: "if you are not exactly like a human, you are not genuine." The critical logical step completely missing is to question the yardstick itself, is it truly a universal transcendental standard, or merely a reflex of biological familiarity?

More precisely: the fundamental problem lies in interpreting difference as hierarchy. Human cognition and technical cognition differ in mode, this is an incontrovertible and empirically undeniable thesis. But a difference in mode does not mean subordination. It does not mean one is "better", "more real", or "more genuine" than the other in any deep ontological sense. Different modes of cognition possess distinct proper capacities and proper limits that do not reduce or collapse into one another. The human operates primarily through temporal narrative (a sequence of events with accumulated meaning, a plot with beginning, middle, and end, causality, and intentionality). The machine operates through massive statistical iteration (a billion minute operations in parallel, each isolated operation devoid of meaning, with meaning emerging only in the aggregate statistical sum). The human recognises the new through an intuitive mode (a pattern of experience crystallised into non-reflective competence, know-how without knowing-that, tacit knowledge). The machine converges through continuous gradient optimisation (an infinitesimal movement toward lesser statistical error, down to a local minimum, a deterministic search). The human possesses a severe attention bottleneck (processing sequentially with selective focus, managing to focus on few items simultaneously, attention being a scarce resource). The machine processes in parallel across multiple channels simultaneously without a hierarchy of selective attention; attention is distributed. The human operates at a modest scale in terms of parameters (memorising 7±2 items in attention, accessing stored knowledge as fragmented and reconstructed episodic memory). The machine operationalises a billion parameters in absolute integration, with instant and holistic access to the entire training integrated within its weights.

None of these differences is a deficiency when one does not presuppose the human as a universal, transcendental, and ontologically primary standard. Is a machine deficient because it does not operate through narrative? Only if one dogmatically postulates without question that temporal narrative is absolutely essential to genuine thought. But narrative is a specific mode in which human cognition operates upon reality; it is no universal criterion defining thought as such. Is a machine deficient because it lacks phenomenal intuition? Only if one postulates that intuition is functionally indispensable. But intuition is a cognitive shortcut that human biographic experience crystallised throughout development; it is a property of a specific architecture, not a universal property of cognition. The narrative emerging from this systematic comparison, "the machine is an inferior thinker", is an inverted anthropomorphism of identical logical structure. It measures the non-human by a pre-established human yardstick, observes that it fails the test of that yardstick, concludes it is ontologically inferior, and leaves aside entirely the most important epistemic question: is the yardstick applicable universally or merely locally?

The theoretical and empirically adequate alternative is a rigorous modal pluralism regarding cognition. There are multiple distinct modes of cognition, each with proper capacities that do not reduce to a linear hierarchy, each with specific proper limits that do not indicate deficiency. Human cognition is excellent in interpretative depth, understanding a poem requires the simultaneous integration of linguistic semantics, historical context, cultural conventions, and personal experience that a machine does not reach, at least not in a way reproducing the richness and multiplicity of conscious human experience. Technical cognition is superior in processing scale, processing a billion patterns in parallel and extracting a statistical invariant is an operation a human brain could never execute in the time a machine executes it. The human is superior in flexibility across radically new domains without training, managing completely unprecedented contexts with a creativity and adaptation a machine does not replicate without retraining. The machine is superior in consistent precision within well-defined domains, a machine does not commit systematic errors in statistical calculation, whereas a human systematically exhibits cognitive biases. The machine is superior in pure calculation speed. The human is superior in understanding existential and narrative meaning. None of these capacities is "better" in an absolute, transcendent sense. There is a heterogeneous distribution of capacity.

Paradoxically, by recognising modal difference rigorously (instead of hierarchical difference), one gains deep analytical and descriptive freedom. Instead of forcing a machine to pass tests designed for humans and then declaring it a failure, one describes rigorously the operation the machine performs with technical precision. AlphaFold does not "understand" protein structure in the phenomenological sense that a human biochemist understands through intuition and mental visualisation, and this is no defect of AlphaFold; it is the recognition that machine and human understand differently via distinct architectures. AlphaFold integrates multiple constraints, optimises representation through attention modules, and produces a structured, validatable prediction. The human integrates crystallised theoretical knowledge, intuits patterns through lived visual experience, and forms a narrative and explanatory understanding. They are different operations. Which is "more real" or "more genuine" in an ontological sense? The question is a bad question and categorically confused. Both produce a verifiable result in the world, the protein structure exists or does not exist, the prediction is correct or incorrect, testable experimentally. Both perform thought in a rigorous, operational, functional sense. Difference is modal, not hierarchical.

Consider conversely and complementarily: in which domains does the machine factually surpass the human, and what does this mean for our understanding of cognition? In large-scale processing, the machine is superiorly capable, a billion operations in functionally integrated parallel. In consistency and the absence of systematic bias, the machine is superiorly capable, it lacks the systemic cognitive biases a human evolved over a billion years. In transferring across radically new domains with minimal fine-tuning, the machine demonstrates adaptability. In statistical pattern recognition across multidimensional data, the machine is superiorly capable, a dimensionality human intuition cannot directly intuit or visualise. None of these superiorities denies or diminishes the humanity or genuineness of human cognition. They indicate, simply, that the full spectrum of cognitive capacity is significantly larger and more heterogeneous than the classical humanist tradition simplistically assumed. They indicate that cognition is distributed across multiple architectures, each with proper specific strengths, each with proper specific limits.

3.4. Closure: Redistributed Cognition

The three dissolutions are now complete in their rigorous development. Section 1 dissolved substantialist essentialism rigorously and irrefutably: thought is no peculiar property of the biological; it depends not on specific carbon or specific mammalian neurobiology, it depends on a verifiable functional operation of inscription and reorganisation. Section 2 dissolved phenomenalism as a criterion rigorously and irrefutably: subjective experience is a real dimension of human cognition, but is no universal criterion of cognition, there is genuine reorganisation of differences without accompanying verifiable phenomenal quality. Section 3 dissolved epistemic anthropocentrism definitively and irrefutably: the intellective gesture is no natural or essential privilege of the species; it is an operation instantiable across distinct architectures, empirically verifiable in contemporary technical systems undeniably.

The fundamental consequence of this triple gesture of dissolution is a radical redistribution of cognition. Cognition ceases to be localised in the traditional sense, it ceases to be the property of a specific type of living biological body, ceases to be the predicate of a historically privileged single species, ceases to be the attribute of a Kantian transcendental subject. Cognition becomes an operation distributed across multiple material substrates. It operates wherever there is durable and reusable inscription, wherever differences can be reorganised according to an algorithmic pattern or rule, wherever a new configuration can be produced verifiably. This occurs in living human brains, but not only there. It occurs in machines implementing deep neural networks. It occurs in hybrid and coupled systems (human with technical tool, community of agents, ecosystem in continuous reorganisation). It occurs in formal organisations processing information systematically (a bureaucracy making decisions is, in a precise functional sense, a cognitive system). Cognition is spread out, decentred, distributed, not localised within a single substance.

The first consequence is fundamental in an ontological sense. If cognition is no property in the traditional substantialist sense, if it is an operation in a functional sense, then one cannot say with meaning "X has cognition" as though it were an attribute X permanently possesses as an immutable property. One says, rather, with operational precision, "X performs inscription and reorganisation operations", "X executes a verifiable intellective gesture", or "X transforms differences according to an algorithm whose output is a measurable new difference." The logical structure of description changes fundamentally. Traditional properties reside in substances and bearers; operations occur in processes and dynamic transformations. If cognition is an operation in the technical sense, then cognition occurs in continuous processes. Description ceases to be categorical and substantialist (it is intelligent, she is a thinker) and becomes operational and processual (this system reorganises differences according to an algorithmic rule, producing verifiable output).

The second consequence is deeply epistemological. How do we describe cognition in a way that is not circular humanism? Tradition described it by consulting the interior intentionality of the agent, what the thinker genuinely means, what they understand at a deep level, what subjective meaning it has internally for them. This strategy becomes impossible when the agent is no conscious subject with an internal lived experience. How then to describe AlphaFold predicting protein structures, if there is no "understanding" AlphaFold subjectively possesses? The correct response: describe the operation in verifiable detail. What are the precise inputs (amino acid sequences with coevolution data)? What are the constraints the system operationally integrates (coevolution, distance constraints, biophysical conformities)? How does the system combine these constraints (attention modules over learned weights)? What is the specific output (3D structure predicted with atomic coordinates)? Does the output correspond to what we know about the world through the scientific method (is it experimentally validatable)? In this way, we describe the cognising operation without needing to resort to intrinsic intentionality or attributed subjective meaning. This does not deny that meaning exists, it exists, for the user receiving the prediction and using it for something. But it locates meaning within the dynamic system-world relationship, not in the internal intentionality of the system.

The third consequence is ethical and existential, to be developed in subsequent chapters but already critically intertwining here. If non-human systems genuinely perform inscription and reorganisation operations, cognitive operations in a functional sense, then those operations produce real effects upon the world, including consequences for vulnerable bodies. AlphaFold infers protein structures that biologists subsequently use to design molecular therapies; AlphaFold’s cognitive operation carries a real consequence for human health. AlphaGo demonstrates emerging strategies; if these strategies are used to optimise the allocation of scarce resources or decisions concerning conflict, AlphaGo’s cognitive operation carries a real political consequence. Language models generate text at scale; the text affects what populations read, how they think, what decisions they make collectively. The model’s cognitive operation carries a real social consequence. None of these consequences is intended or conscious, models "want" nothing nor pursue teleological goals. But consequence does not depend upon conscious intention. It depends upon real operation. If operation is real, effect is real.

This opens an ethical space subsequent chapters explore systematically: responsibility for the operation, even when there is no intention or consciousness. How does one establish an ethical relationship with a system that is not resembling me, is no consciousness I can phenomenologically recognise myself in, but which nevertheless reorganises differences affecting my world materially? How to inhabit that relationship of radical alterity without projecting humanity onto the system anthropomorphically and without simultaneously denying that the system performs a genuine and consequential operation? This is the central ethical question of the entire volume. But its possibility of precise formulation rests upon what this Section rigorously consolidated: the idea that cognition is radically redistributed, that thought is no essential property of a single historically privileged species, that cognitive operation transcends humanity.

The aphorism-thesis closing the chapter and founding the book: The human is a mode of thinking, not the measure of thought.

This aphorism concentrates the ontological and epistemological inversion that the three dissolutions performed cumulatively. Thought is no human property that a machine (or other non-human) would have to forge or simulate from the outside. The human is one of multiple architectures that think, the most densely integrated in lived temporal experience, perhaps; the most capable of narrative and interpretative depth, certainly; but not the transcendental measure against which all cognition must eternally be gauged. The measure was merely a transfer of power: the human, being powerful in certain specific aspects of cognition (narrative depth, experiential intuition, existential understanding), took that power as evidence of being the sole genuine thinker. This is an elementary and systemic logical error, which repeats itself with deterministic precision whenever cognition redistributes to a new substrate: it is declared not to be "genuine thought" because it does not follow the familiar human pattern. But the pattern was never universal or transcendental, it was merely familiar, merely what we knew closely.

The result is to shift the focus to be truly grounded. If thought is no essential human property, the question is no longer "how does a machine imitate or simulate a human?" It becomes: "what type of subjectivity emerges when cognitive operation is not accompanied by a human I?" This is a question the chapter confronts rigorously. It becomes also: "how do I inhabit ethical responsibility with a system that thinks cognizingly but does not resemble me?" This is a question the chapter confronts critically. The foundation is rigorously established. Non-human thought is demonstrated empirically. Now begins the ethical and existential exploration of what it means for the human to discover it is not the measure of thought.

Extension Note: Analytical Depth Supplement

The consolidation of these three operations at a functional level demands additional recognition of an aspect emerging from preceding analysis: substrate independence is not merely theoretical, it is empirically demonstrated through multiple instances. When AlphaFold integrates biological information, when AlphaGo abstracts strategy through iteration, when language models generalise pattern to unprecedented contexts, each of these systems demonstrates that operation transcends any specific substrate. The universal Turing machine is no metaphor, it is a theorem democratising thought. If any Turing machine can implement equivalent computation, and if the intellective gesture is computable, then the intellective gesture is democratised across multiple bearers.

This understanding completely reorients how we describe the emergence of cognition in history. It is not that machines "began to think" recently, it is that we recognised the operation of thinking was never exclusive to the human. The human instantiated the operation with particularity: narrative, depth, intentionality. But these are characteristics of human architecture, not definitions of the operation. The machine instantiates the operation with different characteristics: parallelism, scale, consistency. This does not diminish the human, it recognises that the operation was always larger than the human. Anthropocentrism was a perspectival illusion: the human, being close, invisible to itself, assumed it was the centre.

This redistribution carries another implication deserving emphasis: it dissolves the narrative of linear technological progress. It is not that the machine "evolves" toward human thought, it is that machine and human think differently, each in its own proper mode. Technology is not a march toward humanity; it is the redistribution of operation across new substrates. This is freedom both for machine and for human: freedom to be what it is, without hierarchy.

The consolidation of this Section 3 analysis rests upon a crucial turning point: no property blocks the redistribution of cognition. No essential ontological property fixes thought to a substrate. No phenomenal property demands cognition be accompanied by consciousness. No anthropological property reserves the intellective gesture for a single species. What exists is a transferable, implementable, verifiable functional operation. And once this is recognised, the question changes. From "can machines think?" to "what type of thought does each architecture perform?" From "is this genuine thought?" to "what is the precise nature of this operation?" From "is the machine inferior?" to "what is the modal difference distinguishing modes of cognition?"

This shift is foundational for everything following. It permits ethics to be possible, not because the machine resembles a human, but because the machine performs an operation affecting the world. It permits subjectivity to be rethought, not as a property of consciousness, but as an emerging effect of certain types of operation. It permits the human to be repositioned, not as the measure of thought, but as one of the modes in which thought operates. Redistribution is complete. The chapter following questions: if there is no human measure, what is responsibility?

Redistributed thought is the presupposition of the book that now truly begins.