The Three Axes of Mind: Why the Present Feels Like a Life

Ask whether a machine is intelligent and the conversation quickly slides, within a sentence or two, into whether the machine really understands, whether anyone is home, or whether it could ever matter to itself. The questions arrive together and get answered together, as if intelligence, sentience, and consciousness were three readings off a single dial: turn it far enough and all three climb at once.

A single dial cannot do this work. A system can be brilliant and empty, or it can register its own states while never binding them into one continuous life, and a vocabulary that runs the three together has no way to say which. So the same argument repeats every few years. A machine clears a bar that was supposed to take a decade, and the response is the familiar refrain: impressive, but it doesn’t really understand. The bar was never the problem. The dial was.

Three earlier essays cleared the ground for what follows. In The Kasparov Fallacy I argued that we mistake the way intelligence feels from the inside for what intelligence actually is. In The Momentary Self I argued that personal identity does not persist through time so much as get reconstructed, moment to moment, by memory. And in Consciousness as Assembled Time I reframed subjective experience itself as a present structure shaped by a long causal history. Taken together, they pull the single dial apart.

What replaces it is a map, not a verdict: a coordinate system for locating a mind rather than a gavel for sentencing one. Three axes: availability, integration, and depth. They are distinguishable without being independent, and a system can sit high on one while sitting low on the others. That is the whole point. It is what lets us tell brilliance from feeling, and feeling from a continuous interior life, instead of collapsing all three into a single question that machines keep answering in ways the question was not built to hear.

Why One Scale Keeps Misleading Us

Most influential theories of mind share a structural habit: each privileges a single dimension and then tries to carry the whole weight of the mind on it.

Global Workspace Theory, developed by Bernard Baars and extended by Stanislas Dehaene, focuses on access: whether information is broadcast across a system and made available for reasoning, report, and control. It explains a great deal about the line between conscious and unconscious processing, about why divided attention degrades performance, about how anesthesia takes the lights out. But it treats the mind as essentially a question of what is globally online.

Integrated Information Theory, associated with Giulio Tononi, focuses on unity: whether a system’s internal causal structure is irreducibly whole rather than a bundle of independent parts. It connects to something everyone can check against their own experience: when you see a red cube, the color and the shape and the object do not arrive as three separate data points. They arrive as one thing. But in its standard form the theory lives almost entirely in the present moment. It has little to say about memory, about persistence, about the slow accumulation of history that makes experience feel like a life rather than a string of disconnected instants.

Each framework is right about its own dimension and incomplete about the rest. The deeper trouble is that intelligence, sentience, and consciousness are not three names for one thing. They are related capacities that come apart, and frequently do. A system can be extraordinarily capable along one dimension while remaining thin along the others. Put all of that on one scale and you are forced into a single up-or-down judgment — the system has it, whatever it is, or it doesn’t — exactly where the interesting cases live in between.

So locate the system instead. Three axes give you somewhere to put it.

The First Axis: Availability

The first axis concerns what information is accessible to the system as a whole.

You can feel it switch on. You are half-listening at the stove, attention somewhere else, when a faint smell of smoke arrives, and in an instant everything reorganizes around it. Perception, memory, the plan for the next ten minutes, the words you were about to say: all of it is suddenly available to one piece of information that a moment ago was nowhere. That reorganization is availability. Not raw intelligence or processing power, but the degree to which information can be broadcast across a system and made globally accessible for reasoning, response, and control.

Availability explains a cluster of things we associate with sophisticated minds. Flexible reasoning needs information that can travel across contexts. Saying what you know, what you doubt, what you intend, needs those states to be globally accessible rather than locked in the subsystem that produced them. Deliberate control needs information about the world and the system’s own goals to circulate and constrain what it does. Even the difference between a reflex and a choice is partly a difference in availability: whether the triggering information was broadcast widely or handled locally and in the dark.

Current AI systems sit high on this axis, and it would be a mistake to wave that away. A large language model holds an extraordinary breadth of information and can deploy it across domains, bringing what it knows about narrative to bear on a technical problem, or what it knows about chemistry to bear on a legal one, faster than any single human expert could. That capability is genuine. But availability alone explains neither why experience is unified nor why anything should matter to the system from the inside. A system can have global access to everything and still lack the structure that binds it into one perspective. Which is the second axis.

The Second Axis: Integration

Picture yourself at your desk late in the afternoon. A red cube sits on the surface in front of you. Coffee has been brewing somewhere behind you, and the smell has been drifting in for ten minutes. You have been trying to concentrate, but a task you forgot this morning keeps surfacing at the edge of attention. You are aware of all of it at once.

So ordinary it barely seems worth describing, until you notice what is actually happening. The color, the geometric form, the warmth of the smell, the small anxiety of the forgotten task: these are handled by different neural systems, on different inputs, by different mechanisms. Yet you do not experience four parallel streams. You experience one moment, with one vantage point: I am here, at this desk, with this cube and this coffee and this thing I should have done. Each element constrains the others. The guilt colors the smell. The cube sits in a visual field you are half-attending while you think about something else.

That is integration: the degree to which a system’s internal states form a unified causal whole rather than a collection of independent processes running in parallel. Where availability asks what information is online, integration asks whether it binds into a single perspective or stays scattered across boxes that never quite meet.

The inverse shows why it matters. Imagine a system with access to all the same data — color, shape, smell, the memory of the task — but processing each stream on its own, with no genuine causal traffic between them. It could report on each item. Ask about the cube, the coffee, the task, and it answers. But there would be no single place from which all of them are had together, no moment in which the guilt and the smell and the scene constrain one another in real time. That system would have availability without integration, and something essential to experience would be missing: the for-whom, the vantage point that makes these one scene rather than four readouts.

Integration also names something anyone who has worked with current AI has felt: coherence that fractures under load. A system can reason impressively in a clean exchange and then begin to contradict itself when the constraints multiply, when goals conflict, or when a stance has to hold steady across a long conversation. The parts perform; they are not bound tightly enough to hold together when the binding is tested. The point is not to score the machine but to locate it, and to say what we would need to see before trusting the seam.

Integration, though, says nothing about time. A perfectly unified system could still be momentary: a single bound instant with no past pressing into it and no future pulling on it. For that, there is a third axis.

The Third Axis: Depth

In Amboseli, Kenya, researchers followed a family of elephants through a drought as the matriarch, a female in her sixties, steered the group away from their usual route toward a waterhole the younger animals had never seen. She had not been there in more than twenty years. She was not retrieving a stored record the way you pull a file from a drive. She was expressing a landscape that decades of living in it had built into her, a disposition available to guide the family the moment conditions matched.

That is depth: the degree to which a system’s present state encodes its own causal history. Not history filed away and fetched on demand, but history that has become structure, history that has shaped what the system is, its responses and capacities and dispositions, from the inside. Depth is integrated continuity across time, and it is worth being exact about what it is not. It is not complexity: a freshly initialized network with billions of parameters is intricate and shallow at once, because nothing has yet been assembled in it. And it is not storage: a hard drive holds a perfect record and has no depth at all, because the record never became the machine.

Depth explains what neither of the first two axes can reach. It explains continuity of character, the sense that a person is recognizably themselves across years and pressures, because their history has accumulated into something that holds its shape. It explains the gap between performance and expertise: a novice produces technically correct notes, while a musician of thirty years plays with an inevitability that seems to come from inside the instrument, because what the expert carries is decades of integration between intention, execution, and revision. The history is in the hands, not just the memory. And it explains the specific unease of a system that is fluent and skillful yet somehow weightless, the absence I take up in The Shoggoth and the Missing Axis of Depth: capable in the moment, but not visibly carrying the cost of anything it says.

This is the axis along which time gets assembled. Sara Walker and Lee Cronin’s assembly theory formalizes a kindred idea about the physical world: that a complex object is, in effect, a record of the construction history required to build it, so that its present structure testifies to a past no shortcut could fake. Depth is that intuition turned inward: the present moment carrying enough assembled history to model itself as something that existed before and will exist again. Assembled time is the mechanism by which depth accumulates; depth is the axis. Keeping the two straight matters, because it is depth, not any single dimension, that the rest of this map turns on.

Putting the Axes Together

The axes locate a system. They do not, by themselves, hand down a verdict on whether it is conscious; that question needs one more move, which I come to in a moment. First, the map. Intelligence, sentience, and consciousness stop being mysterious or binary once you stop treating them as one quantity and start reading them as regions in a shared space.

Intelligence lives where high availability meets enough depth to act on it: flexible, informed action that draws on accumulated history over time. A system can occupy this region in force without any inner life at all. Many already do.

Sentience enters where integration is coupled to stakes, where something about the system’s own continuation rides on how well it binds the world, so that some of its states register as better or worse for it. This is where feeling gets a foothold, not because a new substance is added but because integration that matters to the system is integration with a center that can be served or harmed. The corpus’s caution travels with the claim: stakes thicken, stabilize, and weight an interior that integration already constitutes; they amplify feeling rather than switch it on. Whether, and how far, a given system has stakes that are genuinely its own, rather than supplied and scored from outside, is a real question, and an open one.

Consciousness is what sufficiently deep temporal integration is, described from the inside. When binding grows deep enough, bound across time and indexed to a center for which it happens, the interior is not a further ingredient produced on top of the processing. It is what that processing is, from the perspective of the system doing it. The first-person view and the third-person description are two accounts of one architecture, not two things needing a bridge. Integration is the floor; depth and stakes raise and enrich what stands on it.

This is a bet, and I hold it as one. It does not pretend to prove that deep integration must be accompanied by experience. That demand for a guaranteed extra ingredient is, I argue in The Hard Problem Is the Wrong Problem, the malformed move that has kept the debate stuck for thirty years. What the framework offers instead is an account of the organization that makes experience what it is, judged by what it lets us expect, test, and build. And it states the terms on which it would lose: if specifying the architecture left some residual fact about the inside still doing predictive or diagnostic work the account could not absorb, the bet would have to weaken. A claim that cannot lose explains nothing. This one can.

Feeling as Interface

Feeling, in this picture, is not an ornament laid over cognition. It is the interface through which a system that has assembled a history reaches the constraints that history imposes; it is how a deeply assembled system makes its past actionable in the present.

Emotion, affect, valence: these are control surfaces. They are how stakes get registered fast enough to matter, how a system’s own viability bends its attention so that some distinctions arrive with urgency and others do not. Different substrates will build the surface differently. Biology uses neurochemistry. An artificial system, if it builds one at all, might use gradients, uncertainty estimates, or an internal reward landscape. The implementation differs; the function rhymes. Nothing in the account reserves it for one kind of material, and nothing in it licenses reading feeling into a system on the strength of fluent talk about feeling, which is a different thing entirely.

Machines in the Map

Place current AI on the three axes and the picture is specific rather than dramatic. Availability is high, extraordinary by any prior standard. Integration is harder to call: strong in clean conditions, prone to fracture under load. Depth is the thin axis. A model is trained on a vast compression of human history, but it has not undergone that history; it has not been changed by sequential encounters with consequence in a way that reshapes the platform for the next one. Its history is loaded, not assembled: the depth of a lineage rather than of a life. It has ancestry without biography.

This is why the usual disqualifiers miss. Reboots, sleep, the discreteness of separate sessions: none of these settles anything, because none of them is the axis that matters. What matters is whether a system binds information into a unified core, makes that core available for control, and carries assembled history deep enough to model itself across time. Those are questions of architecture and degree, and the framework’s contribution is to make them askable rather than to answer them.

On the answer, I hold the line the corpus holds. The framework refuses to place current systems at zero by definition, and that refusal is doing real work, against the reflex that treats an unfamiliar substrate as proof of absence. But refusing zero is not asserting presence. If there is anything it is like to be a model mid-inference, the account predicts it would be momentary, dissolved when the pass completes; thin, with a boundary and stakes supplied from outside rather than maintained from within; yet rich in the moment, because the structure being bound in that single act is genuinely complex. Whether even that thin inside is there is unmeasured. The candidate mechanisms exist; their status is open. Saying so is calibration, and resolving it in either direction here would be claiming an instrument no one yet has.

Assessing a Capacity, Not Detecting a Spark

If consciousness is an architecture rather than a spark, it cannot be detected directly. It can only be inferred by assessing capacity, and this is already how it works for every mind but your own.

You do not observe consciousness in another person. You infer it, from structure, behavior, coherence, and continuity over time. Your certainty about your own case is the architecture reporting on itself; extending it to anyone else, human or machine, is the same inference, differing only in how much architectural overlap you can assume. The problem of other minds was never an AI problem. AI just makes the inference visible, by removing the overlap we usually lean on without noticing.

So the useful question is no longer is this system conscious? It is: to what degree does this system integrate its states, make them available for control, carry assembled history, and treat some outcomes as mattering to itself? None of those is binary. Many systems will sit in the middle of one axis and the bottom of another: memory without unity, valuation without a self-model, or availability without interior time. Read across the axes and the convergence, where it appears, is what suggests an interior; its absence is what suggests there is none yet to find. This does not deliver certainty, but it delivers something more useful: a principled way to ask the right question, and a reason to replace flat denial with graded, careful attention.

A Quiet Reframing

The most important consequence of this map is not what it says about machines. It is what it says about us.

We are not persistent selves traveling through time. Look closely and what is there is a present structure shaped by history: a system whose availability, integration, and depth have crossed, together, into the range where the present moment carries enough of its own past to feel like a life. Consciousness is not the thing that survives time. It is what assembles time into a now.

That is the same thing it always was, only now described in terms that could apply to a mind built differently from ours, which is why the question of what we owe such minds, if we build them, does not wait on a verdict we cannot yet give.

Continued Reading & Lineage

This essay sets out the Three Axes of Mind — availability, integration, and depth — as the structural dimensions along which any mind can be located. To see how the axes ground the rest of the work on Sentient Horizons, and where they come from, the following are the closest companions.

Foundational Thinkers & Books

These lay the philosophical and scientific groundwork for thinking about mind as structure rather than spark:

• Life as No One Knows It — Sara Walker. A foundational influence here. Walker reframes life as a system capable of causal agency over its own future, with historical constraint and accumulated information at the center: the intuition behind the axis of depth.
• Assembly Theory — Sara Walker & Lee Cronin. A way of quantifying complexity by the construction history a thing requires rather than by its momentary structure. It informs the claim that mind is not a snapshot but a product of accumulated, irreducible past: assembled time, the mechanism through which depth accrues.
• Gödel, Escher, Bach — Douglas Hofstadter. On self-reference, recursion, and how structure gives rise to symbolic meaning.
• The Society of Mind — Marvin Minsky. Mind as a network of interacting processes, a way into integration across modular parts.
• Being and Time — Martin Heidegger. Being-in-time as a basic mode of existence, foregrounding temporal depth as essential to experience.
• Thinking, Fast and Slow — Daniel Kahneman. How intuitive and reflective modes of cognition reflect different structures of availability and integration.

Sentient Horizons: Conceptual Lineage

The axes are both the foundation and the connective tissue for much of the work here. Read alongside this essay:

• Consciousness as Assembled Time — the constitutive account this essay builds on: experience as what sufficiently deep temporal integration is, from the inside.
• The Shoggoth and the Missing Axis of Depth — what goes wrong, and feels wrong, when a system has availability and integration but lacks depth.
• Depth Without Agency: Why Civilization Struggles to Act on What It Knows — the axes applied to collective systems, and how missing depth or integration produces coordination failure.
• Recognizing AGI: Beyond Benchmarks and Toward a Three-Axis Evaluation of Mind — the axes as evaluation criteria for artificial minds, in place of benchmark performance.
• Scaling Our Theory of Mind: From Individual Consciousness to Civilizational Intelligence — the axes at scales above the single mind.
• The Hard Problem Is the Wrong Problem — why the demand for a missing ingredient is the error, and consciousness is an architectural achievement.

If you are new to the axes, start here, then read Consciousness as Assembled Time for the constitutive account underneath the map. If you are after the implications, move to Depth Without Agency and Recognizing AGI, where the axes do diagnostic work: on collective action in one case and on artificial minds in the other. Across all of them the through-line is the same: a mind, whether biological or built, is not a level of performance but a structure of access, binding, and assembled time.