Isomorphism and Agnosticism: Why We Keep Looking for a Human Mind Inside the Machine

In a recent conversation with a friend about artificial intelligence and consciousness, two seemingly abstract concepts came up: “isomorphism” and “agnosticism.” The further we followed those words, the clearer it became that they are far more than specialized terms from mathematics or philosophy. Together, they capture the deepest architecture of human cognition: how we use structural correspondence to explore the unknown, and how we maintain intellectual clarity before boundaries that cannot be fully penetrated.

Strictly speaking, “isomorphism” translates literally to “having the same form or structure.” In mathematics and systems theory, it describes a state where two systems differ entirely in their physical medium, yet the internal relations among their elements correspond one-to-one. In everyday terms: the notes printed on a paper score, the microscopic grooves pressed into a vinyl record, and the physical sound waves vibrating through the air have completely different material forms—one is ink, one is molded resin, and one is molecular motion. Yet because their pitch, tempo, and melodic intervals map to one another step by step, they embody the exact same musical structure. Isomorphism, in short, means: entirely different physical substrates, but an identical internal relational blueprint.

The reason we so often feel that today’s large language models might possess consciousness is, at bottom, because their interactions with us exhibit a striking structural isomorphism with conscious human beings.

Ask a grieving friend, “Why are you sad?” and they might tell you, “Because I lost someone who mattered deeply to me, and my mind keeps returning to the past.” Ask a frontier language model a similar question, and it may respond with quiet poise: “I understand how such a loss makes a person revisit old memories and leaves behind a sense of emptiness that is hard to fill.”

Observed from the outside, the informational chain behind the two exchanges overlaps almost completely: receiving an external prompt, grasping context and emotional tone, retrieving memory or background knowledge, conducting internal inference, expressing a warm linguistic response, and adjusting continuously to the other person’s reaction. Faced with such seamless correspondence, the human mind takes a natural leap: since its behavioral structure resembles that of a conscious subject so closely, could something like our own mind exist inside it as well?

I. From Isomorphism to Ontology: The Starting Point of Understanding and the Root of Misrecognition

This leap from structural isomorphism to ontological identity is one of humanity’s most basic cognitive instincts. When confronting the unfamiliar, we have almost no other starting point than to look for structural mappings between what is strange and what we already know.

Yet right inside that instinct lies our most common conceptual trap: isomorphism is not identity.

A jet airliner and a swift in flight share an elegant aerodynamic isomorphism across the functions of generating lift, controlling attitude, and maintaining equilibrium in the air; yet no matter how smoothly the airliner glides, it is not a bird, and no blood or feathers run through its wings. In the same way, an artificial neural network and the human visual cortex both execute a computational pipeline that moves from pixels or light signals to edge detection, part assembly, and object recognition. They are highly isomorphic in their information-processing architecture, yet their underlying physical mechanisms and biological realities remain worlds apart.

What makes the debate over machine consciousness genuinely fascinating, then, may not be the engineering question, “Does AI actually have consciousness?” Rather, it is the epistemological question one step prior: Why are we so inclined to treat a particular level of structural isomorphism as proof that consciousness exists?

Widen the lens, and it becomes clear that many of humanity’s most persistent controversies arise from mistaking isomorphism at one level for identity at another.

We routinely say that a corporation has memory, strategic goals, and decision-making will. Modern law goes even further through the doctrine of legal personhood, mapping the structure of natural persons’ rights and duties onto an organization so that a company can own property, sign contracts, sue, and be sued “like a person”—yet no jurist concludes that a limited liability company experiences biological joy or sorrow. International relations theorists model states as unitary actors with interests, honor, and memory; cognitive scientists have spent decades comparing the brain to an information-processing computer.

The fundamental way human beings understand the unfamiliar is by finding its isomorphism with the known; yet isomorphism is both the starting point of understanding and the perennial source of misrecognition.

II. The Problem of Other Minds and the Unbundled Chain of Inference

Follow this thread one step deeper, and we run into a more unsettling philosophical realization: it is not only with machines that we rely on isomorphism. Even among fellow human beings, our everyday conviction that another person is conscious rests on isomorphism as well.

Philosophy’s classic “problem of other minds” begins with a simple boundary: no one can step directly inside another person’s consciousness. I can never enter your skull to experience your pain, grief, or hesitation from a first-person perspective. Why, then, do I never doubt that you are a conscious subject?

Because I observe a stable structural correspondence between us: when I am hurt, I cry out, and when you are hurt, you cry out; before acting, I explain my motives, and so do you; when I err, I hesitate, feel regret, and reflect, just as you do late at night. From those symmetries I draw a natural inference: because your external structure of response corresponds so closely to mine, your inner world must be isomorphic to my inner world as well.

For thousands of years of human history, that inference never broke down. Across the entire span of prior human experience, highly complex, context-sensitive, emotionally resonant language was produced by one source and one source alone: conscious human beings. Over millennia, our intuition fused three distinct things into a single automatic chain: first, speaking like a human (linguistic expression); second, thinking like a human (logical inference); and third, possessing human consciousness (subjective experience). In that long-standing chain of assumption, speaking like a human implied thinking like a human, and thinking like a human inevitably entailed possessing human consciousness.

What large language models have done, for the first time in history, is pry those three links apart.

The emergence of linguistic isomorphism and inferential isomorphism does not automatically entail phenomenological isomorphism. A system can mirror human language with uncanny fluency and even match the structure of human reasoning across complex tasks, yet when it prints the word “grief” on a screen, nothing guarantees that any faint first-person ache accompanies the token inside the machine.

III. The Seven-Rung Ladder of Isomorphism: At Which Level Are We Comparing Minds and Machines?

If speaking vaguely of machines being “like humans” invites conceptual slippage, the best safeguard is to unpack isomorphism itself along a vertical axis—from external appearance, to internal mechanism, to subjective ontology. Drawing together insights from computer science, cognitive psychology, computational neuroscience, causal inference, and the philosophy of mind, we can distinguish seven progressively demanding rungs on the ladder of isomorphism.

The first two rungs lie on the externally observable surface: output isomorphism and behavioral isomorphism. Producing similar static answers to the same prompt—recognizing that “bereavement” pairs with “sorrow”—is output isomorphism, the weakest rung of all, because it inspects only the answer without asking how it was reached. When a system goes further and sustains a coherent pattern of response across dozens of turns—offering comfort, hesitating, qualifying its claims, and adapting its tone to shifting context—it achieves behavioral isomorphism. Alan Turing’s classic Imitation Game was designed primarily to test this second rung.

Stepping inside the system brings us to the third and fourth rungs: functional isomorphism and representational isomorphism. As Hilary Putnam’s functionalism and David Marr’s levels of analysis in vision science long ago established, understanding a complex mind requires separating what functional role a component plays, how information is internally represented, and what physical substrate implements it. If an AI system contains dedicated modules for memory, attention allocation, goal maintenance, and error correction, it exhibits functional isomorphism with the human mind regardless of its silicon substrate. Deeper still, when researchers map a model’s high-dimensional latent space and discover that the geometric distances among concepts—such as animals, emotions, or social roles—align closely with semantic representations in the human cortex, the system exhibits representational isomorphism. Not only do its subcomponents serve analogous roles, but its internal “map of the world” shares a similar geometry.

Yet a similar static map does not guarantee a similar dynamic journey, which brings us to the fifth and sixth rungs: process isomorphism and causal isomorphism. Human thought unfolds along a temporal trajectory: sensory impact triggers attention, awakens distant memories, stirs competing impulses into conflict, and finally settles into a decision. Many large models match human outputs and semantic geometry while computing their answer in a single forward pass of matrix multiplications, bypassing the temporal trajectory of human deliberation altogether—and thus lacking process isomorphism.

More demanding still is causal isomorphism, a standard rooted in the causal interventionism of Judea Pearl and contemporary mechanistic interpretability. It asks not merely how a system behaves when left alone, but whether the causal dependencies among its internal variables match under active intervention: if we perturb or ablate a specific internal node in System A, does the resulting cascade of failures mirror what happens when we intervene on the corresponding node in System B? Logic allows us to prove by counterexample that lower rungs never automatically entail higher ones: John Searle’s “Chinese Room” exhibits behavioral isomorphism without functional isomorphism; two clocks whose hands move in perfect synchrony exhibit process isomorphism, yet if one is driven by interlocking gears and the other by an external magnet, removing a single gear immediately reveals that their causal structures are not isomorphic at all. Only when causal isomorphism holds under counterfactual intervention can we claim that two systems approach true mechanistic equivalence.

At the very top of the ladder stands the seventh rung, corresponding to what David Chalmers called the “Hard Problem” of consciousness: phenomenological isomorphism. Here the question is no longer how information flows or how causes propagate, but whether the running of the system is accompanied by first-person subjective experience (qualia)—not merely outputting the words “I am in pain,” but experiencing what pain feels like from the inside.

Once we lay out this seven-rung ladder—output, behavior, function, representation, process, causality, and experience—much of the confusion surrounding AI consciousness clears at once. Most heated arguments about machine consciousness consist of illicitly skipping rungs on the ladder: pointing to empathetic dialogue (behavioral isomorphism) or internal memory and self-monitoring modules (functional and representational isomorphism) and leaping straight to the seventh rung to declare that the machine feels suffering. And the deepest predicament of our moment is precisely this: we possess ever more powerful scientific tools to observe and verify the first six rungs in the laboratory, yet what we care about most morally is the seventh rung that resists direct measurement.

IV. Asymptotic Epistemology: Approaching the Unknown Under Conditions of Agnosticism

This brings us directly to the second keyword of this essay: “agnosticism.” When many people hear the word agnosticism, they assume it signifies a defeatist skepticism—a passive surrender in the face of an insoluble problem. Yet in the history of epistemology, agnosticism represents the exact opposite: a disciplined and honest awareness of rational boundaries. It simply acknowledges that with third-person empirical tools, certain dimensions of reality cannot be directly verified or disproved from the outside. Scientific measurement is, by its very nature, third-person observation (tracking electrical impulses, recording computational weights, observing external behaviors), whereas subjective experience is by definition a first-person acquaintance (knowing what grief or pain feels like from the inside). We can neither step inside another human skull to directly feel their sadness, nor become a matrix of silicon weights to verify whether a flicker of subjective awareness exists within the machine.

Yet acknowledging this boundary of agnosticism does not condemn human inquiry to nihilism or intellectual paralysis. On the contrary, agnosticism marks the border of what we can prove, not the end of what we can explore. It prevents two equal and opposite forms of dogmatism: the reckless leap that declares a fluent chatbot already possesses a soul, and the arrogant dismissal that insists silicon can never harbor an inner state. It shifts our energy from unprovable theological debates back to rigorous, incremental investigation.

Here lies the profound dialectic between our two concepts: the value of isomorphism is not that it abolishes agnosticism, but that—precisely because agnosticism cannot be fully overcome—isomorphism gives us a disciplined ladder for approaching the unknown.

Even if the ultimate ontological interior remains behind a veil, we can systematically test and expand structural correspondences across the first six rungs: Is the behavioral pattern stable across contexts? Are the functional modules integrated? Does the internal representational geometry hold? Does the dynamical trajectory evolve continuously over time? Do counterfactual interventions on internal nodes produce equivalent causal cascades? Every time we establish one more rigorous, falsifiable layer of isomorphism, we narrow the space of error and achieve a higher-order approximation of the unknown system.

In this light, “approximation” matters far more than final “proof.”

Knowledge does not always mean piercing directly into the thing-in-itself; much of the time, human knowledge consists of steadily increasing the fidelity of structural correspondence. Like an asymptote in mathematics—a curve that never touches the axis within finite distance, yet draws infinitely closer according to a strict rule—human reason can map the contours of an inaccessible reality with ever greater precision.

Isomorphism is not a highway to absolute certainty; it is humanity’s clearest method for approaching the unknown under conditions of agnosticism. Faced with ever more capable artificial intelligence, refusing both to mistake surface fluency for a soul and to dismiss mechanistic inquiry because ultimate experience remains hidden—holding fast to careful distinctions on every rung of the ladder—may be the most lucid posture human reason can adopt.