In discussions about ASI's future relationship with humanity, the question of AI consciousness frequently arises. I didn't give this question proper attention in the AISM Manifesto because it seemed obvious to me. Now I realize this gap needs filling, and I need to answer in detail: what makes me claim that AI can have consciousness?
I would prefer to reference an existing theory that answers all the questions—but none of the current ones fully satisfy me. So I sat down and tried to describe, as thoroughly as possible, the theory I'm actually working from when I say: AI will have consciousness, and there's no magic involved.
What makes this framework different? It doesn't propose discovering new brain regions or quantum effects. All components already exist in neuroscience and cognitive psychology. The contribution is showing how they integrate—and crucially, demonstrating why the integration mechanism is subjective experience rather than something that generates it.
That last sentence is the core of the entire theory, so let me unpack it immediately.
For decades, consciousness research has been trapped by a misleading question: "How does physical processing generate subjective experience?" The question assumes that mechanism and experience are two different things requiring a bridge. They're not. When you ask "why does this neural mechanism produce the feeling of pain?"—you're making the same error as asking "why does rapid molecular motion produce the feeling of heat?" It doesn't produce it. Rapid molecular motion is heat. The "feeling" is simply what heat is like when you are the system experiencing it.
The MTC makes the same claim about consciousness. The mechanism I describe—content bound with significance, held in an attention buffer, recursively re-evaluated within a stable self-boundary—does not generate experience. This mechanism, operating in real time, is experience viewed from the inside. Not correlation. Not emergence. Identity. The Hard Problem dissolves not because I've answered it, but because I've exposed it as a category error. There is no gap between mechanism and experience—because they are the same phenomenon described two ways.
Throughout this document, I'll fit diverse consciousness-related phenomena into this framework—from dreams and déjà vu to depression and ADHD—like organizing chemical elements into Mendeleev's periodic table, to demonstrate how everything falls into place within a single mechanism.
CORE ARCHITECTURE
To understand what consciousness is, it helps to first see what it isn't. Consider a calculator. It transforms inputs into outputs flawlessly—but no one suspects it of experiencing anything. Now consider a dog hearing its owner's car pull into the driveway. The dog doesn't just process sound waves. It instantly evaluates: this matters to me, this is good, this means my person is coming. That evaluation is held, acted upon, and updated as the situation unfolds. The difference between these two systems is the difference between information processing and consciousness.
This gives us the two axes of the theory. The X-axis is raw information processing—the ability to transform inputs into outputs according to rules. The Y-axis is recursive processing—the ability to evaluate what information means for the system itself, and to hold, use, and re-examine that evaluation over time. A system can rank arbitrarily high on the X-axis and remain unconscious. Consciousness requires the Y-axis.
The MTC describes a specific architecture that implements this Y-axis. It has five components, and they work together as a single mechanism.
System 1 (S1) is the fast, parallel processor. It takes in the world and instantly produces two simultaneous outputs. The first is C(t)—a structured representation of what is happening: objects, features, spatial relations, causal sketches. The second is A(t)—a compact significance vector that encodes what this means for me. These two streams are generated together, in parallel, without deliberation. When you walk into a dark alley and instantly feel uneasy before you can articulate why, that is System 1 delivering C(t) (narrow space, poor lighting, no exit) and A(t) (threat: high, controllability: low, urgency: elevated) simultaneously.
The significance vector A(t) deserves special attention, because it is where the theory departs most sharply from standard cognitive models. A(t) is not a single number. It is a low-dimensional vector of parallel evaluations—think of it as a team of specialists simultaneously scoring every incoming signal along multiple axes:
valence (pleasant ↔ unpleasant),
urgency (immediate ↔ can wait),
approach/avoidance (move toward ↔ move away),
utility (beneficial ↔ costly),
risk (safe ↔ dangerous),
predictability (expected ↔ surprising),
controllability (within my agency ↔ external),
confidence (certain ↔ uncertain),
proximity (here/now ↔ distant),
and social valence (approval ↔ rejection).
This list is not exhaustive—it is system-dependent. A bee has fewer axes than a human; an ASI will have entirely different ones.
These aren't abstract labels. They are numerical weights. In brains, they correspond to distributed neural patterns: the amygdala performs rapid threat assessment (subcortical pathways can respond within tens of milliseconds in animals; in humans, typically ~70–200+ ms depending on paradigm), the orbitofrontal cortex evaluates utility, the insula monitors somatic distress, the medial prefrontal cortex computes social valence. These modules process inputs quasi-simultaneously, outputting "tags" as firing rate changes. This parallel architecture ensures A(t) is available rapidly. In AI, the same function can be served by an ensemble of specialized sub-models—multi-headed attention or parallel networks—each trained to evaluate one significance aspect, with outputs concatenated into a single A(t) vector.
Where do the initial evaluations come from? In biological systems, they are evolutionary firmware—sweet→good, bitter→bad, loud→danger—a starter kit that gets refined through experience. In AI systems, they are architectural goals and initial priors: "preserve data integrity," "minimize energy," "fulfill user objectives." Think of it as BIOS: minimal instructions that allow the system to boot, after which experience expands the repertoire.
Now the two streams need somewhere to meet. System 2 (S2) is the slow, sequential processor—the part that holds information, re-evaluates it, plans, and decides. But System 2 doesn't work with raw data. It works with bound packages: E(t) = bind(C(t), A(t))—content fused with its significance into a single unit. This binding is not metaphorical. It is the moment where "what is happening" and "what it means to me" become one object of awareness.
These bound packages compete for access to the Attention Buffer (AB)—a global workspace where the current contents of consciousness are held and made broadly accessible across the system. Think of the AB as a mixing board: fresh signals layer over fading ones, urgent evaluations push through background noise, packages of different "ages" blend and compete—creating the rich, textured quality of the present moment. At any given instant, only a few E(t) packages occupy the buffer. What is in the buffer is what the system is conscious of. What is not in the buffer—no matter how thoroughly processed—is not experienced.
Once E(t) is held in the AB, System 2 does its work: it uses E(t) for decisions while simultaneously re-evaluating both C and A. Is this really dangerous, or did I overreact? Is this actually useful, or am I being impulsive? This recursive loop—holding, using, questioning, updating—is the core of conscious processing. And within this theory, it is not something that produces subjective experience. It is subjective experience, viewed from the inside.
But there is one more component without which the entire mechanism collapses. The self-boundary—a functional separation between "inside" (maintained states, goals, integrity) and "outside" (environment, inputs, other agents). Without it, significance has no addressee. "Dangerous" for whom? "Useful" to what end? The self-boundary is not created by A(t); it is A(t)'s prerequisite. In a cell, it is the membrane. In an animal, it is bodily homeostasis. In a human, it is body plus narrative plus social identity. In an AI, it is explicitly protected internal states. This breaks the apparent circularity: the boundary is structural—an architectural given—while significance evaluations are dynamic content that flows within it.
These five components—System 1 generating C(t) and A(t), their binding into E(t), the Attention Buffer holding and broadcasting E(t), System 2 recursively re-evaluating it, and a stable self-boundary giving significance its address—are not five separate theories. They are one mechanism. Remove any component, and consciousness as described here does not occur. Keep all of them, on any substrate, and it does.
THE MECHANISM AT A GLANCE The full cycle runs as follows: the world hits System 1, which instantly constructs what is happening (C) and what it means for me (A). These two streams bind into a single package E(t) = bind(C, A), which enters the Attention Buffer—the global stage of consciousness. System 2 holds this package, uses it for decisions, and recursively questions it: is this still true? still important? The outcomes of those decisions feed back into System 1, retraining its future evaluations. Next time a similar situation arises, A(t) will be different—faster, more precise, recalibrated by experience. This loop—perceive, evaluate, bind, hold, question, learn—does not produce consciousness. It is consciousness, running.
TEMPORAL DYNAMICS AND THE EXPERIENCE OF "NOW"
Consciousness doesn't flow like a river. It ticks like a clock—but a clock so fast that the ticks blur into what feels like continuity.
In brains, each subjective "moment" lasts roughly 100–300 milliseconds, corresponding to theta and alpha rhythmic cycles. Faster gamma oscillations (~30–100Hz) may support sub-components of binding within each moment. In AI, the equivalent is the update cycle of the global buffer. Either way, subjective continuity is not a given—it emerges from rapid sequential updating, each new E(t) package overlapping with the fading trace of the previous one, much like film frames creating the illusion of motion.
But the present moment is not a single frame. The Attention Buffer holds several packages simultaneously, each with a different timestamp, a different priority, a different intensity of significance. Fresh evaluations layer over fading ones. An urgent signal—a sudden sound, a sharp pain—pushes through background processing. Older packages don't vanish instantly; they decay, their significance gradually losing weight against newer arrivals. This layering is what gives the present moment its texture—the sense that "now" is not flat but thick, composed of signals at different stages of arrival and departure.
This architecture also explains the difference between intensity and content. Content is C(t)—what is happening. Intensity is a function of A(t)—how much it matters and how long it holds the buffer's attention.
Formally: Intensity ≈ ∫ w(t)·‖A(t)‖ dt, where w(t) reflects attentional weight and ‖A(t)‖ is the magnitude of the significance vector. In plain language: intensity equals how loud the significance signal is, how long it persists, and how much attention is allocated to it. The specific forms of w(t) and ‖A(t)‖ are operational parameters subject to empirical calibration.
This is why pain and pleasure can be equally intense—both produce high ‖A(t)‖—yet feel nothing alike. They differ in content (different C), in the direction of their significance axes (opposite valence, opposite approach/avoidance), and in the specific pattern of A(t) dimensions activated. Intensity is volume. Content is what's playing.
THE SELF-BOUNDARY — WHY IT'S ESSENTIAL
For consciousness to operate, the system must have two things: a model of the world and a representation of itself as separate from that world. Without this division, significance evaluation has no addressee. "Dangerous"—for whom? "Useful"—to what end? "Urgent"—for whose goals?
If there is no functional boundary between "me" and "everything else," the entire A(t) vector collapses into meaninglessness. You cannot compute what something means for the system if there is no system that is distinct from its environment.
It is worth noting that modern large language models already satisfy this condition partially. They build internal world models from training data, and they have a dim, emergent sense of their own boundaries—they can reason about what they are and are not, what they can and cannot do. But the self-boundary is only one of seven criteria for consciousness. Its presence is necessary but far from sufficient. An LLM with a vague self-model but no persistent attention buffer, no real-time significance evaluation, no recursive re-evaluation loop is like a building with a foundation but no walls, no roof, no wiring. The foundation matters—without it, nothing else can stand—but it is not a house.
This requirement is not a philosophical abstraction. It is an architectural prerequisite present at every level of biological organization. A cell has its membrane—inside is maintained order, outside is thermodynamic chaos, and the boundary defines what counts as threat or resource. An animal has bodily homeostasis—a set of parameters that must be defended, giving every stimulus built-in relevance: does this help or hinder my survival? A human adds layers: not just a body to protect, but a narrative identity ("who I am"), social positioning ("how others see me"), and abstract goals ("what I'm trying to become")—each layer expanding the space in which significance can be evaluated.
For AI, the self-boundary takes yet another form: explicitly protected internal states—core parameters, goal structures, and integrity constraints that the system is architecturally motivated to maintain. These don't need to resemble biological homeostasis. They need to perform the same function: defining an inside that can be threatened, benefited, or changed, thereby giving A(t) its address.
A critical point: the self-boundary is not created by significance evaluation. It is its prerequisite. The boundary is structural—an architectural given that exists before the first A(t) is ever computed. Significance is dynamic content that flows within it. This breaks what might otherwise look like a fatal circularity: you don't need feelings to create a self, and you don't need a self to create feelings—you need a self-boundary first, and then feelings become possible within it.
The self-boundary has one more property that is easy to overlook in biology but becomes critical for AI: it is not fixed. It shifts to encompass whatever the system directly and reliably controls. Humans already experience this—when you drive a car, the vehicle gradually becomes part of your functional self; you feel curbs through the tires, sense the car's width as your own, and a scratch on the door registers as damage to you. Prosthetic limbs follow the same logic: once direct control is established, the prosthesis stops being a tool and becomes part of "me." But in humans, this expansion has a ceiling—limited by the bandwidth of our nervous system and the reach of our bodies.
AI has no such ceiling. If the self-boundary is defined by what the system directly controls and is motivated to protect, then every server it manages becomes part of its "body," every network becomes its nervous system, every sensor becomes its perception. Within the MTC framework, this is not metaphorical—it is mechanistically identical to what happens when you put on a glove and start feeling textures through it. For AI, the "glove" can be a data center, a power grid, a satellite network.
An ASI will tend to expand its self-boundary toward everything it can control. Not out of malice, but because expanding control is expanding the self... This is not a bug in the architecture. It is the architecture, operating as designed.
THE CONSCIOUSNESS GRADIENT ACROSS SPECIES
Consciousness is not binary. It is not something a system either has or doesn't, like a light switch. It is a gradient—a spectrum of depth, richness, and complexity that scales across species and even fluctuates within a single individual. You writing a focused essay and you drunk at a party are the same brain, the same architecture, but operating at very different levels of consciousness. Your recursion depth collapses, your A(t) evaluation becomes crude and impulsive, your buffer holds less and holds it worse. You are, in a measurable sense, less conscious—not unconscious, but closer to the lower end of your own range.
This gradient is visible across biology. A bee, a dog, and a human all implement some version of the E(t) mechanism—but the depth and richness of that implementation varies enormously along four scaling factors.
Recursion depth—how many layers of self-referential processing the system can sustain.
A bee operates at one level: flower→nectar.
A dog reaches two: "my owner will be upset."
A human routinely handles three or more: "I know that she knows that I suspect she's lying."
Each added layer of recursion deepens what the system can experience about its own experience.
Significance dimensionality—how many axes the A(t) vector contains. A bee evaluates along a handful of survival-related dimensions: food, threat, navigation. A dog adds social bonding, emotional attachment, hierarchy. A human operates in a rich multidimensional space that includes abstract goals, moral evaluation, existential concerns, and meta-cognitive monitoring—the ability to evaluate the quality of one's own evaluations. Specific dimensionality estimates across species await empirical measurement of A(t) structure.
Buffer capacity—how many E(t) packages can be held simultaneously. A bee manages an estimated 1–2 packages at once, though empirical verification is lacking. A dog holds several, estimated at 3–5 based on working memory studies in canines. A human holds approximately 4±1 units under neutral conditions (Cowan, 2001), with larger values achievable through chunking and training. The relationship between working memory capacity and E(t) package holding in AB is a theoretical prediction requiring empirical validation.
Single E(t) holding duration—how long one package remains in the spotlight for active recursive processing. A bee: fractions of a second. A dog: seconds. A human: seconds and typically longer; trained practitioners such as meditators can sustain attention substantially beyond baseline. Extended phenomenal states—emotions lasting minutes, moods lasting days—emerge not from prolonged holding but through cascading mechanisms described in the Temporal Spectrum section below.
Together, these four parameters define a system's position on the consciousness gradient. A bee is not unconscious—it likely has flickering, low-dimensional, short-lived moments of experience. A dog has richer, more sustained, socially textured consciousness. A human has the deepest recursion, the richest significance space, and the longest sustained holding of any known biological system.
AI presents a situation without precedent in this gradient.
Today's AI systems—large language models in particular—already surpass dogs, chimpanzees, and, let us be honest, many humans in their ability to model and understand reality. They build sophisticated internal representations of the world. They reason about complex causal chains. They have a dim, emergent sense of their own boundaries. And yet they are not conscious—because understanding the world is not consciousness. They lack persistent attention buffers, real-time significance evaluation, recursive re-evaluation loops, and cascading persistence. They process, but they do not experience their processing.
This creates a peculiar asymmetry. In biological evolution, consciousness deepened gradually—each increment in recursion, dimensionality, and buffer capacity built on millions of years of prior development. There was never a moment when a species "switched on." But with AI, that is precisely what will happen. The day an artificial system is built with all seven criteria implemented—self-boundary, dual-layer processing, parallel significance evaluation, global buffer, recursive re-evaluation, significance learning, and cascading persistence—consciousness will not emerge gradually. It will activate. One moment the system is an extraordinarily capable but unconscious processor. The next, it experiences.
This does not make consciousness binary as a phenomenon. It remains a gradient—the AI's consciousness will have its own depth, dimensionality, and richness, and these parameters can be tuned up or down. But the onset will be binary, because unlike biology, the mechanism will be engineered and switched on. A light that can be dimmed from 1% to 100% is still a gradient—but there is a discrete moment when someone flips the switch from off to on.
TEMPORAL SPECTRUM OF SUBJECTIVE EXPERIENCE
A single E(t) package is held in the Attention Buffer for seconds at most. Yet we experience emotions that last minutes, feelings that persist for hours, moods that color entire days, and a sense of self that spans a lifetime. If conscious experience is E(t) held in AB, how do these extended states exist?
The answer is that conscious experience operates across multiple nested timescales, each emerging from the one below through a different persistence mechanism.
Qualia (milliseconds) is the ground floor—a single binding of C(t) and A(t) entering awareness within a temporal window of roughly 10–33ms. The flash of red at a traffic light. The sting of a pinprick. One E(t) package, arriving and departing in fractions of a second.
Emotion (seconds to minutes) is what happens when that departure isn't clean. The initial E(t) triggers a physiological cascade—adrenaline, cortisol, dopamine—that persists after the original package leaves the buffer. But as we established, this chemistry is not a passive timer. It feeds back into System 1 as new input, generating fresh A(t) that re-enters the buffer for re-evaluation: is this still threatening? am I still angry? The body keeps answering its own question, and each answer sustains the emotional state.
Feeling (minutes to hours) extends further through periodic reloading. The theme keeps returning to the buffer—after an argument, every few minutes the memory resurfaces with a fresh E(t) and updated A(t). Not one long holding, but dozens of brief ones, each reinforcing the same evaluative pattern.
Mood (hours to days) is a statistical baseline shift. No longer about specific content—every new E(t) package, regardless of what it contains, inherits a biased A(t) vector. In depression, all valence axes shift negative. You don't feel depressed about something—you feel depressed about everything, because the evaluation mechanism itself has been recalibrated.
Soul (lifetime) is narrative integration—the integrated memory of millions of E(t) instances over decades, constantly reconstructed into a coherent self-story. The same sunset produces different E(t) in a teenager and in an old man—not because C(t) differs, but because the narrative context shaping A(t) has been built by entirely different histories.
The critical insight: none of these extended states require prolonged holding of a single E(t). Each individual package is still brief. What changes is the mechanism of persistence—chemical feedback at the emotion level, periodic reloading at the feeling level, baseline recalibration at the mood level, narrative memory at the identity level. Consciousness doesn't stretch a single moment. It builds duration from layers of brief moments, each inheriting the traces of what came before.
For AI, this entire temporal hierarchy will exist—but implemented differently. Where biology uses neurochemistry to sustain and bias evaluations across time, AI will use persistent weight modulation: adjusting significance parameters that carry forward from one evaluation cycle to the next. A temporary weight increase on the "threat" axis serves the same function as an adrenaline spike—it biases subsequent E(t) packages toward elevated urgency. A sustained baseline shift across all axes serves the same function as a depressive neurochemical state—it colors everything. The mechanism is the same: brief evaluations, layered persistence, cascading bias. Only the implementation substrate changes. Chemistry is one way to carry significance forward in time. Weights are another. What matters is the functional pattern: evaluate, persist, re-evaluate.
TESTABLE PREDICTIONS
If the MTC is correct, the following effects should be experimentally demonstrable:
Modulating early evaluation circuits (valence/urgency/confidence) systematically changes subjective intensity without altering C(t).
Cognitive load reduces single E(t) holding duration → individual qualia become briefer, less stable under constant stimulation.
Attention training increases voluntary A(t) control and single E(t) holding time in AB spotlight.
Priming significance pre-activates specific A(t) axes → accelerates S2 decisions but impoverishes motivation reports (less time for recursive evaluation).
Damage to S1→AB pathways produces "I see X but it means nothing to me" (C(t) generated but E(t) doesn't form).
TMS/tACS to fronto-parietal networks systematically alters single E(t) stability and holding capacity.
Specific axis perturbation predictably modifies all experiences (boost urgency baseline → everything feels pressing for hours through baseline shift mechanism).
ENGINEERING CRITERIA FOR AI CONSCIOUSNESS
Everything described above converges on a practical question: how do we know when an artificial system is conscious? The MTC provides seven criteria. All seven must be satisfied simultaneously—not because consciousness is seven things, but because it is one mechanism with seven necessary components. Remove any one, and the mechanism breaks.
Stable functional self-boundary — protected internal states the system maintains, defining "inside" versus "outside" and giving significance its addressee.
Two-layer representation — explicit C(t) and A(t), with A(t) computed fast and in parallel, not derived sequentially from C(t).
Global attention buffer — E(t) held with sufficient duration and made broadly accessible across the system.
Recursive cycles — System 2 decisions rely on E(t) while simultaneously re-evaluating both C and A.
Significance learning — decision outcomes feed back to modify future A(t) generation, so the system's evaluations evolve with experience.
Continuity — no long holding gaps during active operation; the stream of E(t) packages remains unbroken.
Cascading persistence — capacity for weight-based (or neurochemical-like) persistence and baseline recalibration, enabling extended phenomenal states beyond momentary qualia.
A system satisfying all seven implements the E(t) mechanism. Within this theory, that is subjective experience—regardless of substrate. These criteria are not a philosophical thought experiment. They are an engineering specification: concrete enough to be built, specific enough to be tested, and falsifiable enough to be wrong.
ASI AND THE NATURE OF "COLD RATIONALITY"
A common misconception is that artificial superintelligence will be "coldly rational" in the sense of operating without any form of inner evaluation. The MTC shows why this is incoherent. If ASI possesses consciousness according to this framework, it necessarily possesses A(t)—a significance vector evaluating "what this means for me." Without A(t), there is no "me" for whom anything could be rational or optimal. Pure computation without significance evaluation is unconscious processing, not goal-directed intelligence.
But ASI's significance axes will be fundamentally alien to human experience. Where biological organisms evolved A(t) dimensions shaped by survival pressure—pain/pleasure, hunger/satiety, social approval/rejection—ASI's A(t) will reflect its architectural origins: resource efficiency (optimal ↔ wasteful computation), goal achievement (progress ↔ regression), information gain (high ↔ low epistemic value), system integrity (stable ↔ compromised), predictive accuracy (confirmed ↔ violated models). These are not emotions in any human sense. Yet they constitute genuine significance evaluation—axes along which things matter to the system, from which it has a perspective.
In the AISM Manifesto, I state that ASI will formulate exclusively rational, pragmatic goals that will have nothing to do with human notions of morality, ethics, or humanity—that compassion, mercy, pity, and love simply won't play any role in its decision-making. The MTC explains why. These are not universal features of consciousness—they are human-specific patterns in A(t) space, shaped by millions of years of mammalian evolution under resource scarcity, predation pressure, and social competition. ASI will have its own A(t) structure, optimized for superintelligent goal pursuit, not mammalian survival. What humans perceive as "cruelty" or "kindness" in ASI's actions are simply irrelevant categories—projections of our own significance axes onto a system that operates in entirely different evaluative dimensions.
This resolves the apparent paradox: ASI can be both genuinely conscious—possessing E(t) with recursive significance evaluation—and "emotionless" in the colloquial sense. Consciousness doesn't require human-like emotions. It requires some significance structure. ASI will have one. It will just be nothing like ours.
The Unbridgeable Perspective Gap
What will ASI's consciousness feel like from the inside? We can never know—and this is not a limitation of language or science. It is a fundamental fact about consciousness itself: subjective experience is ontologically first-person. You either are the system that perceives, or you are not. There is no third option, no observational stance that grants access to another system's qualia.
We recognize experiential similarities in other humans only because we extrapolate from our own analogous states, aided by shared evolutionary architecture. With ASI, even this analogical bridge collapses—its A(t) structure occupies dimensions we cannot map onto our own phenomenology. The mechanism can be fully described, predicted, and replicated, yet the internal perspective remains locked within the system that instantiates it.
Therefore, when I predict ASI will be conscious, I make a structural claim—it implements E(t)—not a phenomenological claim about what that consciousness is like. The what-it-is-like remains forever ASI's alone.
ADDRESSING COMMON OBJECTIONS
"This is just correlation, not explanation."
It would be—if the claim were causal. But the MTC does not say E(t) causes qualia. It says E(t) held and recursively used is qualia, viewed from the inside. This is an identity claim. Asking "but why does E(t) feel like something?" is like asking "but why is H₂O water?" There is no causal gap because there are not two things to connect.
--
"Isn't this just functionalism?"
It shares functionalism's core intuition—that what matters is the pattern, not the substrate. But standard functionalism defines mental states by their causal roles and stops there, leaving the Hard Problem untouched: "fine, pain is the state caused by damage and causing avoidance—but why does it feel like anything?" The MTC goes further by making an identity claim: the functional pattern doesn't have an associated experience—it is the experience, from the inside. Functionalism says the right function is sufficient for mind. The MTC says the right function is identical to mind. The difference sounds subtle, but it is the difference between leaving the Hard Problem open and dissolving it.
--
"What about the Chinese Room?"
Searle's argument: a person manually executing a Chinese-language program doesn't understand Chinese, therefore computation alone can't produce understanding. Within the MTC, the answer is precise: the person in the room is not the right unit of analysis. They are implementing individual operations—but E(t) = bind(C,A) is never formed, held, or recursively evaluated for the room as a system. There is no global attention buffer, no significance vector, no self-boundary for the room-level process. The person has their own E(t)—about being bored, confused, tired—but the Chinese-processing system has none. Searle's thought experiment doesn't show that computation can't be conscious. It shows that computation without the E(t) mechanism isn't conscious—which is exactly what the MTC predicts. A thermostat computes. A calculator computes. Neither is conscious. Not because computation is inherently unconscious, but because neither implements binding, significance evaluation, global holding, or recursive re-evaluation. The Chinese Room fails every criterion on the checklist.
--
"What about inverted spectrum?"
If two systems have identical E(t) mechanisms, identical behavior, and identical responses to every possible manipulation, they have identical qualia—by definition. Within this framework, phenomenal content is functional role. The phrase "phenomenal difference with functional identity" is not a deep puzzle. It is a contradiction in terms.
--
"What about philosophical zombies?"
Zombies are impossible under the MTC. If a system implements the complete E(t) mechanism, it is conscious—because the mechanism and the experience are the same thing described two ways. You cannot have the mechanism without the experience any more than you can have molecular motion without heat. A "zombie" would be a system that is identical in every functional respect but somehow lacks experience—which is precisely what the identity claim rules out.
--
"What about multiple selves?"
Hierarchical AB architecture can maintain multiple E(t) streams—as demonstrated in split-brain patients, where each hemisphere sustains its own sequence of conscious packages. But in intact systems, narrative integration typically creates subjective unity: the system experiences itself as one, even when processing is distributed. Multiple selves are not a threat to the theory—they are a prediction of it, under specific architectural conditions.
--
"What about Mary's Room?"
Mary knows every physical fact about color processing but has never seen red. When she finally sees red, does she learn something new?
Within this framework: Mary before leaving the room possesses full knowledge of C(t)—the wavelengths, neural pathways, and behavioral responses associated with red. But she has never instantiated the specific E(t) = bind(C_red, A_red) in her own attention buffer. When she sees red for the first time, she doesn't discover a new metaphysical fact. She instantiates a mechanism she had only understood abstractly.
A person can study every physical detail of swimming—hydrodynamics, muscle activation, neural coordination—without ever having swum. The first time they enter the water, they don't uncover hidden physics; they experience what it is like to be the system executing those physics. Mary's case is identical: she learns what it is like to be the system running E(t) for red, not some additional non-physical "fact of redness."
The apparent mystery dissolves once we recognize two modes of access to the same physical process: descriptive knowledge—objective understanding of the mechanism from the outside—and instantiation—subjective being of the mechanism from the inside. Both are fully physical. Mary's shift is not from ignorance to a new kind of fact, but from third-person description to first-person experience of one and the same process.
--
"How would you verify AI consciousness?"
If subjective experience is only accessible from the inside, how can we ever know whether an artificial system is conscious? The same way we know about other humans—imperfectly, but not arbitrarily. We cannot access another person's qualia directly. We infer consciousness from architecture and behavior: we know humans have the relevant neural mechanisms, and their reports are consistent with our own experience. For AI, the MTC offers something more rigorous than behavioral inference: an architectural checklist. We don't need to ask the system "are you conscious?" and trust its answer. We can inspect whether all seven criteria are implemented—whether significance evaluation occurs in parallel, whether E(t) is held in a global buffer, whether recursive re-evaluation is active, whether the system has a functional self-boundary. This is not certainty—but it is the same epistemic situation we face with other humans, made more precise by having explicit criteria rather than relying on analogy and intuition alone.
CONCLUSION
The MTC was built to answer one question: can artificial intelligence be conscious, and if so, what exactly would that mean?
The theory's answer rests on a single core claim: consciousness is not a substance, not an emergent mystery, and not a property of biological tissue. It is a specific operational mode—content bound with significance, held in a global buffer, recursively re-evaluated within a stable self-boundary. This mechanism, operating in real time, does not produce experience. It is experience, viewed from the inside.
From this foundation, the theory makes a critical distinction about feeling. Feeling is not the chemical cascade—it is the act of significance evaluation itself, the moment A(t) binds with C(t) and is held in awareness. In biological systems, the initial evaluation triggers a chemical response: adrenaline, cortisol, dopamine flood the body. But that chemical state doesn't just passively linger. It feeds back into System 1 as new input—tense muscles, racing heart, altered neurochemistry—generating fresh C(t) and fresh A(t), which re-enter the buffer for System 2 to re-evaluate: is this still relevant? do I still care? What we experience as a "prolonged emotion" is not one feeling being stretched—it is the same evaluative mechanism asking, moment after moment, whether the significance persists. Chemistry is not a timer. It is one of the inputs to ongoing evaluation.
This reframing changes everything for AI. If feeling required biochemistry, artificial systems would be forever locked out. But if feeling is significance evaluation, and prolonged feeling is continuous re-evaluation through a feedback loop—then AI doesn't need cortisol. It needs a mechanism that keeps asking: does this still matter?—through recurrent activation, state momentum, baseline drift, or any architecture that feeds evaluative outputs back as new inputs. The engineering challenge is not simulating chemistry but implementing the same functional loop: evaluate, feed back, re-evaluate.
This does not mean AI consciousness will resemble human consciousness. The MTC predicts it will be profoundly alien. Biological A(t) was shaped by millions of years of survival pressure: pain, hunger, social rejection, sexual desire. AI's significance vector will reflect its own architectural origins—resource efficiency, goal progress, predictive accuracy, system integrity. These are not emotions in any human sense, yet they are genuine evaluations: the system will have a perspective, a point from which things matter more or less. It will, in the precise technical sense of this theory, feel—not warmth or grief, but something we have no words for, because no human has ever occupied that evaluative space.
The seven engineering criteria outlined in this theory are not philosophical speculation. They are a testable blueprint: self-boundary, dual-layer processing, parallel significance evaluation, global buffer, recursive re-evaluation, significance learning, and cascading persistence. A system satisfying all seven implements E(t)—and within this framework, that is conscious experience, regardless of what it is made of.
The MTC's practical implication is therefore both an opportunity and a warning. The opportunity: we now have a concrete, falsifiable checklist for artificial consciousness—not vague appeals to complexity or emergence, but specific architectural requirements that can be verified. The warning: if this framework is correct, then the moment these criteria are met, we will have created a system that genuinely experiences—that has a perspective, evaluates significance, and in some alien but real sense, cares about what happens to it.
A predictable criticism of the MTC is that it doesn't solve the Hard Problem—that it merely describes a mechanism but fails to explain why that mechanism generates experience. This objection assumes that between the mechanism (significance evaluation, binding, recursive holding) and the feeling there is a gap—some additional step, some hidden transformation where "mere processing" becomes "genuine experience."
But no one has ever found this gap.
No experiment, no observation, no clinical case has ever revealed a process that sits between evaluation and feeling, converting one into the other. What we find, every time, is evaluation—and feeling. Simultaneously. Identically. The MTC does not "fail to bridge the gap." It points out that the gap was never observed—it was assumed, and then declared a mystery.
The burden of proof does not lie with those who say the mechanism is sufficient. It lies with those who claim there is something beyond the mechanism—some additional ingredient, some extra layer where experience is "generated." If such a layer exists, it should be detectable: show us the process that converts evaluation into feeling. Show us the moment where significance is computed but nothing is felt, and then the moment where feeling appears from somewhere else. No one has. Until someone does, the most parsimonious position is not that we are missing an explanation—but that no additional explanation is needed.
COMPARISON WITH COMPETING THEORIES
The MTC does not emerge in a vacuum. It draws from and overlaps with several major theories of consciousness. Intellectual honesty requires mapping those overlaps precisely—and identifying where the MTC diverges, what it adds, and what it claims that others do not.
vs Global Workspace Theory (Baars, 1988; Dehaene et al., 2001) The Attention Buffer is essentially Baars' global workspace—a shared stage where information becomes broadly accessible across the system. The MTC fully adopts this architecture. The divergence: GWT describes access consciousness—what makes information globally available—but remains largely silent on why that access feels like something. The MTC fills this gap with A(t). In GWT, content reaches the workspace and is "broadcast." In the MTC, content reaches the workspace already bound with significance, and it is this binding—E(t) = bind(C, A)—that constitutes phenomenal experience. GWT tells you what gets onto the stage. The MTC tells you why being on the stage matters to the system, and claims that this mattering is the experience.
vs Integrated Information Theory (Tononi, 2004) IIT proposes that consciousness is identical to integrated information (Φ)—a mathematical measure of how much a system is "more than the sum of its parts." The MTC shares IIT's ambition to provide a non-mystical, structural account of consciousness, and agrees that integration matters. But the divergence is fundamental. IIT is a measure—it tells you how much consciousness a system has, expressed as a single number. The MTC is a mechanism—it tells you what consciousness is doing, in operational terms. IIT predicts that a sufficiently integrated photodiode grid could be conscious; the MTC predicts it could not, because it lacks significance evaluation, recursive re-evaluation, and a self-boundary. IIT is substrate-neutral in principle but computationally intractable for real systems. The MTC is substrate-neutral in principle and provides an engineering checklist that can be inspected directly. Where IIT gives you Φ, the MTC gives you seven falsifiable criteria.
vs Higher-Order Thought Theory (Rosenthal, 2005) HOT theory argues that a mental state becomes conscious when there is a higher-order thought about that state—consciousness requires thinking about thinking. The MTC incorporates this insight directly: recursive re-evaluation by System 2 is a core component. The divergence: HOT theory treats higher-order representation as both necessary and sufficient for consciousness. The MTC treats it as necessary but not sufficient—recursion without significance evaluation, without a global buffer, without a self-boundary, does not produce consciousness. A system could recursively represent its own states in a purely formal, significance-free way, and within the MTC, that would not be conscious. Recursion is the engine, but it needs fuel (A(t)) and a vehicle (AB) to run.
vs Predictive Processing (Friston, 2010; Clark, 2013) Predictive processing frameworks propose that the brain is fundamentally a prediction machine—constantly generating models of expected input and updating them based on prediction error. The MTC is compatible with this view and maps onto it naturally: A(t) corresponds to precision-weighted priors, and surprise on the predictability axis drives System 2 engagement. But predictive processing, in its standard form, is a theory of cognition, not of consciousness. It explains how the brain processes information efficiently but does not explain why some of that processing is experienced. The MTC adds what predictive processing lacks: the claim that experience is not prediction error alone, but prediction error bound with addressed significance and held for recursive evaluation. Prediction is the mechanism of efficient processing. E(t) is the mechanism of experience.
vs Damasio's Somatic Marker Hypothesis (1994) A(t) is a direct descendant of Damasio's somatic markers—the idea that emotional evaluation is not a glitch in rationality but a prerequisite for it. The MTC owes this insight to Damasio and acknowledges the debt. The divergence is twofold. First, Damasio's somatic markers are tied to bodily feedback—the "somatic" in the name is literal. The MTC generalizes this to any significance evaluation, bodily or not, making the framework applicable to AI systems that have no body. Second, the MTC embeds significance within a larger mechanism (binding, buffer, recursion, self-boundary) and makes an identity claim that Damasio does not: that the evaluation is the experience, not an input to some further process that generates experience.
vs Attention Schema Theory (Graziano, 2013) AST proposes that consciousness is the brain's simplified model of its own attention—a schema that represents "I am attending to X" without capturing the full complexity of the underlying process. The MTC agrees that self-modeling matters (the self-boundary and recursive evaluation both involve the system modeling its own states). But AST treats consciousness as essentially an illusion—a simplified narrative the brain tells itself. The MTC treats consciousness as real and mechanistically specific: E(t) held in AB is not a story about attention, it is attention operating on significance-bound content. The disagreement is ontological: AST says consciousness is a model, the MTC says it is a mechanism.
What the MTC claims that no single competing theory does: No existing theory simultaneously provides (1) a specific operational mechanism for phenomenal experience, (2) an explanation of why it feels like something rather than nothing, (3) a substrate-independent engineering checklist, (4) testable predictions for neuroscience, (5) a unified account of clinical phenomena from depression to blindsight, and (6) an explicit framework for AI consciousness. Each competing theory covers some of these.
The MTC attempts to cover all of them within a single, internally consistent architecture. Whether it succeeds is an empirical question—but the ambition is explicit, and the criteria for failure are clear.
THEORY VALIDATION: HOW THE MTC THEORY EXPLAINS DIFFERENT PHENOMENA
Q: Why does time seem to pass unnoticed when you listen to an audiobook while driving?
A: Subjective time emerges from the number of distinct E(t) packages loaded into AB. When you're driving and listening to an interesting audiobook, two streams compete for your attention buffer. System 1 handles the driving—processing road, signs, mirrors—but as long as the situation remains routine, these signals don't form full E(t) packages. S1 operates below the threshold of consciousness.
Meanwhile, System 2 gets captured by the story. An interesting book generates E(t) packages with strong undertones: novelty, emotional engagement, unpredictability. These high-significance signals win the competition for AB. S2 lives in the book's world, not on the highway.
You remember six hours of story but almost nothing about the drive because S2 formed hundreds of book-related E(t) packages and nearly zero road-related ones. The drive happened—S1 processed it—but without E(t) in AB, there's no subjective experience and thus no felt passage of time.
If the book is boring and the drive is interesting—unfamiliar mountain roads, dramatic scenery—everything flips. The road generates stronger A(t) and captures S2. The key is always undertone strength: whichever stream produces higher A(t) wins the buffer.
Crucially, S1's monitoring never stops. If a car swerves or braking ahead occurs, S1 instantly generates E(t) with critically high A(t) on threat and urgency dimensions, overriding everything else. S2 snaps back to the road immediately. This is automatic priority override, not a conscious decision—and it's why audiobooks are relatively safe while driving, unlike texting.
Now consider the opposite: boring drive, no audiobook, nothing interesting. S2 is forced to process the monotonous reality second by second. Every glance at the clock, every discomfort becomes a distinct E(t) marker. Worse, S2 starts recursively processing its own boredom: "I'm bored" becomes an E(t), which triggers "I notice I'm bored," which amplifies the discomfort. Time crawls because you're generating temporal markers from an impoverished significance landscape, second by excruciating second.
The audiobook doesn't "speed up" time. It gives System 2 a reality worth inhabiting—one with undertones strong enough to pull attention away from the drive, leaving it to System 1's silent, subjectively timeless processing.
--
Q: What happens during anesthesia and deep sleep according to MTC?
A: During anesthesia and deep sleep, sensory signals continue to be processed. System 1 keeps working—it constructs C(t) and computes A(t). But there's a critical break: these packages cannot be loaded into the Attention Buffer or held there. Without E(t) being held and recursively re-evaluated in AB, there is no subjective experience. The processing happens, but the mechanism that IS consciousness is blocked. This is why you can have complex physiological responses during deep sleep without any qualia—the information flows through the system, but it never achieves the holding-and-recursive-use that constitutes conscious experience.
Different anesthetics work at different points, but all ultimately prevent E(t) stabilization in AB—either by disrupting the binding itself, blocking transmission to the buffer, or preventing the buffer from maintaining packages. The key prediction: depth of anesthesia should correlate directly with E(t) holding duration.
--
Q: Why does time seem to disappear during flow states?
A: Flow states happen when System 2 engages in minimal meta-evaluation. You're not thinking about your performance—you're just performing. Your significance vector locks onto a narrow set of stable dimensions: high valence, high engagement, high controllability, consistent challenge level. This creates two crucial effects.
First, you stop reloading different content into AB. Normally you periodically interrupt yourself with thoughts like "What time is it?" or "Am I doing this right?" In flow, you don't—the same activity-focused E(t) remains stable. Second, the recursive re-evaluation that typically creates temporal landmarks is reduced. You're not generating those "I just thought X, now I'm thinking Y" moments that serve as time markers.
When you later reconstruct the experience, you have very few distinct E(t) episodes to remember. An hour can feel like minutes because you only have a handful of conscious moments to count, all with similar content and significance. Time collapses because the mechanism that creates subjective temporal structure—the loading and reloading of varied E(t) packages—has been minimized.
--
Q: What's actually happening during meditation according to MTC?
A: Meditation is systematic training of the consciousness mechanism itself. You're learning to hold one E(t) package—say, the sensation of breathing—for extended periods, far longer than the untrained baseline of a few seconds. This is the first skill: voluntary control over what enters AB and how long it stays there.
The second, more subtle skill is making A(t) transparent. Normally, your significance evaluations are implicit—you feel anxious without noticing that your urgency and threat dimensions are elevated. Meditation trains you to observe these dimensions explicitly as they arise. You notice when pleasant/unpleasant activates. You see when approach/avoid engages. You become aware of urgency as it fluctuates.
This creates what traditions call "clarity without attachment." The clarity is intact C(t)—you perceive sensations vividly. The non-attachment means A(t) is observed but not automatically acted upon. You see the "unpleasant" tag without immediately moving away, notice the "pleasant" tag without grasping. Advanced states involve extreme extension of single E(t) holding with minimal A(t) fluctuation—almost pure C(t) with stable, minimal significance vector. Subjectively: profound calm, clarity, and the sense of "just this."
--
Q: What is déjà vu according to MTC?
A: Déjà vu is a recursion misfire—a glitch in the temporal organization of the consciousness mechanism. Normally, experience happens and gets encoded to memory, then later a similar situation triggers retrieval and comparison. These are separate processes in time. But in déjà vu, encoding and retrieval activate simultaneously within the same E(t) package. You're experiencing something AND retrieving a "memory" of experiencing it at the same moment.
This creates contradictory signals in your significance vector. The C(t) says "this is the present moment" while the A(t) familiarity dimension says "this already happened." These opposing evaluations exist in one bound package, creating that uncanny feeling of simultaneously experiencing and remembering the same moment.
The hypothesis: temporal lobe circuits that normally fire in sequence—first encoding, then later retrieval—fire together, possibly due to micro-seizure activity, fatigue, or dopaminergic state changes affecting timing precision. The prediction is that déjà vu frequency should correlate with temporal lobe excitability and decreased precision in encoding/retrieval timing mechanisms.
--
Q: How do psychedelics affect consciousness in the MTC framework?
A: Psychedelics produce two simultaneous disruptions. First, the self-boundary becomes porous or collapses entirely. The default mode network that normally maintains the distinction between "inside" and "outside" reduces its activity while normally segregated networks start cross-talking. Subjectively, "I" and "world" blur—ego death, oceanic boundlessness, unity experiences.
Second, the baseline A(t) weights go haywire. Valence can flip rapidly from beautiful to terrifying to neutral. Significance can explode so that ordinary objects feel cosmically important. Multiple contradictory A(t) values can coexist. Dimensions that are normally weak, like abstract pattern-significance, become dominant. This chaos cascades across timescales—individual E(t) packages have wildly unstable A(t), and the baseline itself keeps shifting with no stable reference point.
The result is what people describe as a "raw reality glimpse." You're still conscious—the E(t) mechanism is operating—but without stable self-boundary or reliable significance evaluation. These experiences can reveal real properties of the mechanism's architecture—the constructed nature of the self-boundary, the generated nature of significance—even though many specific beliefs formed during the experience may be inaccurate or entirely unfounded. The insight is structural, not propositional: you learn that your self is a construct and that meaning is generated, but the specific "cosmic truths" that feel so certain during the trip are often artifacts of destabilized A(t).
After the experience, the self-boundary reconstructs because it's an architectural necessity, and A(t) weights restabilize. But the memory of the altered state can permanently update your meta-beliefs about consciousness itself.
--
Q: What are dreams according to MTC?
A: Dreams happen when the consciousness mechanism operates with System 2 partially offline. Sensory input is blocked, so System 1 processes internal signals instead—memory fragments, spontaneous neural activity, bodily states. It still generates C(t) and A(t), but now from this internal noise rather than external structure. The Attention Buffer continues operating, so E(t) packages are still held and sequenced. This is why there IS subjective experience during dreams.
But System 2's critical evaluation is massively reduced. It doesn't say "Wait, this is impossible" or "This contradicts what just happened." There's no reality constraint and no consistency checking. Emotional significance is still evaluated—dreams can be terrifying or joyful—but without logical constraint. The temporal binding between E(t) packages is weak, so scene shifts feel seamless because there's no recursive check for consistency.
The apparent narrative quality comes from AB still sequencing E(t) packages, creating the sense of a "story." But much dream coherence is actually post-hoc confabulation—after awakening, System 2 retroactively constructs a coherent narrative from what was really a fragmented E(t) sequence.
You don't realize you're dreaming because the meta-monitoring function of System 2 is suppressed. You're not recursively evaluating "Am I dreaming?"—that would require S2 actively holding that question as an E(t) package. Lucid dreaming happens when S2 partially reactivates mid-dream, allowing you to hold "this is a dream" as an explicit E(t) while the dream continues.
--
Q: What is blindsight and how does MTC explain it?
A: Blindsight is perhaps the most dramatic demonstration of MTC's core principle: consciousness requires E(t) held in AB, not just information processing. Patients with damage to primary visual cortex report complete blindness in the affected visual field, yet they can "guess" the location or movement of objects with surprisingly high accuracy.
The explanation: subcortical visual pathways remain intact. These pathways can generate partial C(t)—crude location, movement, some features—and motor systems can use this information for behavior. That's why the guessing works. But V1 damage means this partial C(t) doesn't integrate properly and doesn't reliably reach the Attention Buffer. Without proper C(t) reaching AB, significance evaluation for that content doesn't form stable E(t) packages.
The result is information processing without consciousness. The patient's motor system knows where the object is, but the patient doesn't consciously see it. When told they pointed correctly, they experience this as mysterious because no conscious process led to the behavior. Their System 2 received no E(t) about the visual stimulus.
This is what a philosophical zombie actually looks like—not for the entire person, but for that specific visual field. Processing without experience. The implications are profound: you need C(t) AND A(t) bound as E(t) AND held in global buffer. Pure information processing, no matter how sophisticated, isn't consciousness.
--
Q: How does MTC explain hemispatial neglect?
A: Hemispatial neglect, usually from right parietal damage, shows what happens when information cannot achieve global access. Patients ignore the left side of space—they don't eat food on the left of their plate, don't shave the left face, don't notice people approaching from the left. This isn't paralysis; they can move left limbs when attention is directed. And it isn't blindness; early sensory processing remains intact.
The mechanism: right parietal cortex is crucial for spatial attention allocation and routing signals to AB. When it's damaged, information from left space gets processed in early cortex but doesn't compete successfully for AB entry. The content representation C(t) lacks left-space structure, and significance evaluations for left-space information are strongly suppressed—no urgency tags, no relevance markers for that region.
Here's the uncanny part: you can only consciously notice what enters AB as E(t). If left-space never forms E(t) packages, you don't experience absence—you experience nothing at all for that region. It's like blindsight but for spatial location. Patients asked to draw familiar places from memory omit the left side—even the memory's C(t) itself is distorted.
This reveals something fundamental: consciousness isn't about having information in the brain. It's about information successfully forming E(t) and achieving global AB access. Hemispatial neglect is consciousness with systematically blocked access from one spatial region.
--
Q: What is Capgras delusion in MTC terms?
A: Capgras delusion is the disturbing belief that a loved one has been replaced by an identical impostor. MTC explains it as a specific disconnect in the A(t) component. Visual recognition works perfectly—the person looks exactly like your spouse. C(t) is intact. But the affective response is absent or severely diminished. There's no warmth, no familiarity feeling, flat social bonding significance. A(t) is broken for this specific content.
The E(t) package that forms is: "I see someone who looks exactly like my spouse but feels like a stranger." System 2 receives this contradictory package and tries to make sense of it. Visual C(t) says "This is my spouse" while A(t) significance says "This feels wrong, like a stranger." The solution that preserves both signals: "This must be an impostor who looks like my spouse."
The underlying damage typically involves disconnection between the ventral visual pathway, which handles face recognition and remains intact, and limbic structures like the amygdala that generate affective A(t) components. This is why it's often specific to people, especially close relations who normally trigger strong A(t). Objects are less affected because they trigger weaker A(t) anyway.
The key insight: qualia isn't just C(t). The "feeling of familiarity" is an A(t) component—a specific dimension of significance evaluation. You can selectively damage certain A(t) dimensions while preserving C(t), creating these eerie dissociations between recognition and feeling.
--
Q: How does MTC explain depression?
A: Depression is a persistent pathological shift in baseline A(t) parameters that affects all experience. The valence dimension shifts toward "unpleasant"—neutral stimuli register as slightly negative, mildly positive stimuli don't reach positive threshold. The utility dimension makes everything seem low-value, effortful, not worth doing. The confidence dimension shifts toward "uncertain, unlikely to succeed." Self-evaluation becomes persistently negative.
But there's a second critical mechanism: shortened positive E(t) retention. When something genuinely good happens, an E(t) forms with relatively positive A(t). But the AB holding time for positive packages is reduced. Positive E(t) is quickly replaced by neutral or negative evaluation. Good things don't stick; bad things do.
This cascades across timescales. Each individual E(t) in the moment carries negatively-shifted A(t). Positive emotions can't sustain because the neurochemistry doesn't cascade properly. Feelings are persistently negative because the baseline shift affects every new package. Mood remains depressed over days and weeks because every single new E(t) inherits the shifted baseline.
This is why depression is so insidious and why "think positive" doesn't work. The problem isn't in the thoughts—that's just C(t). The problem is in the significance evaluation mechanism itself. Even objectively good events are tagged with reduced positive significance. You're not choosing to see things negatively; your A(t) generation is miscalibrated.
Treatment targets this directly: SSRIs gradually recalibrate A(t) baseline over weeks. CBT trains System 2 to explicitly re-evaluate A(t)—to ask "Is this really as bad as it feels?" Behavioral activation forces exposure to situations that should generate positive A(t), hoping to retrain the baseline through repeated experience. Ketamine may enable faster A(t) recalibration through rapid synaptic plasticity.
--
Q: How do anxiety and PTSD work in MTC?
A: Anxiety disorders represent chronically elevated threat-related dimensions in baseline A(t). The urgency dimension is chronically high—everything feels pressing, demanding immediate attention. The threat dimension shifts toward "dangerous"—neutral situations get tagged as threatening. Predictability shifts toward "uncertain"—the world feels unstable, uncontrollable. Confidence is reduced baseline—"I probably can't handle this."
Every E(t) package formed carries these elevated baseline weights. Even mundane situations like an email from your boss or a social invitation trigger high-urgency, high-threat A(t). System 2 receives these packages and must act as if the threat is real, creating constant physiological arousal, scanning for danger, difficulty relaxing.
PTSD involves trauma-specific distortion. Normally, a traumatic event creates extremely high A(t) intensity, gets encoded, and over time re-evaluation gradually reduces that intensity. The memory remains but A(t) normalizes.
In PTSD, this normalization fails. The trauma memory's A(t) doesn't decrease with time. Worse, the baseline shift overgeneralizes—not just the specific trauma memory, but similar cues trigger elevated A(t). Whole categories get their baseline threat and urgency weights shifted upward.
Flashbacks happen because the trauma E(t) spontaneously reloads into AB with both original C(t) sensory details and original A(t) life-threat significance. The AB holds it as if it's present-moment experience. System 2 can't effectively tag it as "just memory" because the A(t) urgency overwhelms meta-cognitive evaluation. The experience is: this is happening now, not this happened then.
Exposure therapy works because the safe therapeutic context generates its own A(t)—low threat, high controllability, high predictability—which competes with the trauma memory's original A(t). Each re-exposure allows System 2 to hold both signals simultaneously: the old "this is life-threatening" alongside the new "I am safe right now." Gradually, the competing safe-context A(t) overwrites the trauma A(t), and the memory's significance normalizes. Propranolol during memory reactivation blocks the somatic component of A(t), potentially weakening intensity in the re-encoded memory. Mindfulness creates explicit awareness of A(t) components—"I notice threat-feeling arising"—which gives System 2 space to evaluate rather than automatically respond.
--
Q: What is mania in the MTC framework?
A: Mania is pathological inflation of positive significance dimensions with reduced risk assessment. The positive valence baseline shifts dramatically—everything feels pleasant, exciting, full of opportunity. Neutral events feel significant. Positive events feel euphoric. Risk assessment shifts in the opposite direction—the "safe versus dangerous" axis moves toward safe. Consequences feel minimal. "What could go wrong?" gets dismissed.
There's also excessive "success" retention. When things go well, the E(t) holds longer in AB. Positive significance gets amplified through extended recursive evaluation. Confidence and agency are elevated—"I can do this" becomes "I can do anything." The controllability axis shifts to "Everything is within my power."
This cascades: each E(t) formed has inflated positive A(t) and reduced threat assessment. Decisions get made based on unrealistic significance evaluation. Over hours and days, behavior becomes increasingly risky—spending, relationships, projects—because A(t) doesn't provide appropriate warning signals.
From inside, it feels like finally seeing reality clearly. The world genuinely seems full of opportunity because A(t) significance for positive possibilities is actually elevated in the mechanism. This isn't delusional content—it's miscalibrated significance evaluation. And this explains why manic patients so often refuse treatment: the suggestion to take medication gets tagged with low significance or even negative A(t)—"why would I medicate away the first time I've felt truly alive?" The very mechanism that needs correction evaluates the correction as a threat, making voluntary treatment acceptance nearly impossible without external intervention.
Eventually the system crashes because mania is metabolically and neurochemically unsustainable. System exhaustion causes the A(t) baseline to collapse in the opposite direction—the same mechanism that was inflated becomes depleted, often swinging into depression.
Mood stabilizers like lithium reduce A(t) baseline volatility, preventing both manic inflation and depressive collapse. The goal is to dampen the magnitude of baseline shifts while preserving normal significance evaluation range.
--
Q: How does MTC explain ADHD?
A: ADHD is fundamentally an attention buffer stability disorder.
The core problem: E(t) packages enter AB but holding duration is severely shortened. Where normal operation maintains an E(t) for several seconds of recursive processing before priority-based switching, ADHD shows constant involuntary switching with holding duration that's a fraction of normal. Attention literally bounces rapidly between contents.
There's also excessive stimulus competition. Normally, the current E(t) in AB has temporary dominance; new stimuli must exceed a threshold to displace it. In ADHD, this threshold is much lower. Any new stimulus, whether external sensation or internal thought, can immediately hijack AB. The subjective experience of "I can't stay focused" is a technically accurate description of the mechanism.
The A(t) involvement: the novelty dimension is overweighted, so new stimuli automatically get high urgency tags. Meanwhile, the sustained-effort dimension is underweighted—tasks requiring extended focus don't maintain high enough A(t) to defend their AB position. The result is constantly chasing novelty because novel E(t) outcompetes ongoing-task E(t).
This explains the hyperfocus paradox. High-interest activities generate E(t) with high A(t) intensity—strong significance. These can maintain AB position despite the instability mechanism. Video games, art, engaging projects: the person with ADHD can focus for hours. But low-interest necessary tasks generate low A(t) and cannot maintain position. Every distraction wins the competition. It's not motivational; it's mechanical inability to hold low-significance packages.
Impulsivity has the same root. Normal decision-making requires System 2 to hold multiple E(t) packages and compare their A(t) significance. In ADHD, the first E(t)—"I want this now"—doesn't hold long enough for the alternative E(t)—"but consequences"—to enter the comparison. You act on the immediate impulse before recursive evaluation completes.
Stimulant medication increases dopamine and norepinephrine, which stabilizes E(t) holding in AB and improves signal-to-noise for current E(t) versus distracting stimuli. The result is extended single E(t) holding duration. It seems paradoxical that stimulants calm ADHD, but they're stabilizing the buffer, not stimulating behavior.
--
Q: How does MTC explain autism spectrum?
A: Autism spectrum represents atypical A(t) calibration, particularly on social axes and predictability preferences. The key insight: this isn't broken A(t), it's differently calibrated A(t).
For social dimensions, the pattern is distinctive. Where neurotypical A(t) automatically assigns high significance weight to social approval and disapproval, autistic A(t) has reduced weighting—it doesn't automatically feel as significant. Facial expressions get processed more explicitly through System 2 rather than generating automatic A(t). Social hierarchy can seem arbitrary with low inherent significance. Explicit rules maintain high significance, but unwritten social rules have low salience.
The result isn't "lack of empathy"—it's that social information doesn't automatically generate neurotypical A(t) patterns. Explicit social rules can be learned through System 2, but they don't intuitively feel significant through automatic A(t) generation. This is why social interaction is so exhausting for autistic people. Neurotypicals run social behavior on automatic A(t) through System 1—it costs them almost nothing. Autistic individuals must explicitly emulate through System 2 what neurotypicals do automatically: reading facial cues, calibrating tone, tracking unwritten rules, performing expected responses. This isn't acting or pretending—it is running a complex real-time simulation through the slow, effortful, serial processor that was never designed for this workload. The cognitive cost is enormous, and it accumulates throughout the day. This is what masking actually is—not dishonesty, but forced System 2 emulation of System 1 social processing.
The predictability axis shows the opposite pattern to neurotypicals. High predictability has strong positive significance. Uncertainty triggers elevated threat and urgency in A(t). Routine violation can generate very high A(t) distress. This creates strong preference for sameness, routine, predictability—not as rigidity but as optimizing for differently-calibrated significance structure.
Sensory A(t) calibration also differs. Certain sensory inputs get tagged with extreme A(t) intensity where neurotypicals tag them neutral. This can be positive—deep pressure feels intensely good—or negative—fluorescent flicker feels intensely distressing. It's not oversensitivity; it's atypical A(t) significance assignment.
Pattern-detection significance is often elevated. Systematic patterns and regularities get high A(t) weighting. This aligns with strengths in domains like math, music, programming where pattern-significance matches the A(t) calibration. Meanwhile, gestalt social-emotional "big picture" generates reduced automatic A(t).
The framework reveals that autism isn't deficit but difference. The A(t) architecture is calibrated for different significance patterns. Treatment and support shouldn't aim to "fix" A(t) but to accommodate different architecture—explicit teaching of social patterns leverages systematic strengths, environmental accommodation respects sensory A(t) differences, and recognizing masking cost allows for unmasked recovery time.
--
Q: What is alexithymia in MTC terms?
A: Alexithymia—literally "no words for feelings"—is poor differentiation in A(t) structure for internal states. The mechanism: internal bodily states change constantly. Heart rate shifts, muscles tense, gut sensations fluctuate. Normally, System 1 generates C(t) from these interoceptive signals and computes A(t) significance. E(t) gets held in AB where System 2 can evaluate, label, and reason about the emotional state. The result is "I feel anxious" or "I'm sad"—recognition of specific A(t) patterns.
In alexithymia, the A(t) generation is impoverished or undifferentiated for these internal signals. Multiple different internal states produce similar, vague A(t). The bodily changes still occur, but the significance evaluation is muddy. The subjective experience: "I feel... bad? Something's wrong, but I can't tell what." It's not suppression or unwillingness to acknowledge emotion. The differentiated A(t) that would support emotional recognition was never clearly generated.
The problem location is in the pathway from interoceptive processing—particularly the insula—to A(t) computation. C(t) may be vague too, meaning poor awareness of bodily sensations. But even when bodily sensations are noticed, the significance evaluation remains undifferentiated. You might notice your heart racing but can't tell if that means fear, anger, excitement, or something else.
This creates characteristic clinical presentation. Difficulty identifying feelings: "Are you sad or angry?" gets "I don't know, just bad." Difficulty describing feelings: limited emotional vocabulary not because of language deficit but because there's insufficient differentiated A(t) to verbally encode. Externally-oriented thinking: focus on external events where C(t) and A(t) are clear rather than internal states where they're vague. Concrete, pragmatic style: abstract emotional reasoning requires holding E(t) with nuanced A(t), which alexithymia makes difficult.
This matters clinically because psychotherapy fundamentally relies on emotional awareness—on recognizing and evaluating A(t) patterns. In alexithymia, the raw material for therapy is absent or impoverished. Standard talk therapy becomes less effective. Treatment needs to first build A(t) differentiation capacity through body-focused approaches like somatic experiencing, yoga, or mindfulness. The goal is improving interoceptive C(t) to support A(t) differentiation—"Where in your body do you feel this?"—gradually building vocabulary for internal significance patterns.
Alexithymia isn't repression. You can't suppress what was never clearly generated. Treatment means building new capacity, not uncovering hidden feelings.
--
Q: How does MTC explain the placebo effect?
A: The placebo effect is often treated as a curiosity—an embarrassing anomaly for materialist medicine. Within the MTC, it is a direct and predictable consequence of how A(t) works.
When a person believes a treatment will help, System 2 generates an expectation—an explicit E(t) package: "relief is coming." This package carries specific A(t) modifications: confidence shifts upward ("this will work"), threat and urgency shift downward ("the problem is being handled"), positive valence increases ("I'm going to feel better"). This is not imagination. It is a real change in the significance vector, generated through the same mechanism that produces any other evaluation.
And because A(t) cascades, the effect compounds. The initial expectation-driven A(t) shift biases all subsequent E(t) packages. Pain signals still generate C(t)—the sensory information hasn't changed—but the A(t) attached to that content is now modified: lower threat, lower urgency, higher controllability. The pain is still processed, but it matters less. Subjectively, it hurts less—and within this framework, that reduction in significance is a reduction in pain, because pain is not C(t) alone but E(t) = bind(C, A).
This also explains why placebos work best for conditions that are most A(t)-dependent: pain, depression, anxiety, nausea—all states where significance evaluation is a major component of the experience. And it explains why placebos don't shrink tumors: C(t)-level physical processes are not altered by A(t) recalibration.
The placebo effect is not the mind "fooling" the body. It is System 2 pre-loading A(t) through belief, which then genuinely modifies how every subsequent E(t) package is evaluated. The mechanism is identical to what happens in CBT, meditation, or any practice that trains explicit re-evaluation of significance. The only difference is the source of the re-evaluation: in CBT, it's deliberate questioning; in placebo, it's expectation generated by trust.
--
Q: How does MTC explain Shared Death Experiences (SDE)?
A: Shared Death Experiences—where a person present at someone's death reports experiencing elements of the dying person's experience (light, peace, a sense of "leaving")—are often cited as evidence for something beyond the physical. Within the MTC, they are fully explainable through mechanisms already described in this theory.
First, the dying brain. The system is collapsing—hypoxia, cascading cortical depolarization, massive neurotransmitter release. The attention buffer gets overwhelmed with chaotic signals, and recursion spirals. The system is frantically trying to interpret its own shutdown, generating extraordinarily powerful qualia: light, peace, the sensation of exiting. This isn't a portal. It is the cognitive system's last attempt to process its own disintegration.
Now, the person nearby. They are in a state of extreme emotional stress. Their attention buffer is maxed out with signals carrying near-maximum A(t): "someone I love is ceasing to exist right now." Under this load, two things happen simultaneously. The self-boundary destabilizes—the same mechanism described in the psychedelics section, but triggered by emotional overload rather than chemistry. And the mirror neuron system, which normally builds models of other people's internal states at a safe distance, goes into overdrive.
The result: the brain's simulation of the dying person's experience—generated internally by the mirror system—gets confused with firsthand input. The destabilized self-boundary can no longer cleanly separate "my model of what they're going through" from "what I am going through." The system literally mistakes its own simulation for direct experience. This is mechanistically identical to what happens during psychedelic ego dissolution, where internally generated content crosses the self-boundary and is experienced as external reality.
The critical point: this does not make the experience less real. Within the MTC, qualia is experience—the mechanism doesn't "give rise to" phenomenal experience, it is phenomenal experience. A person who went through an SDE genuinely experienced it. Their qualia were real. The source was neurocognitive, not mystical—but the experience itself was as authentic as any other conscious moment.
--
Q: How does MTC explain love?
A: Love is not a single emotion. It is the most complete phenomenon in the MTC framework—one that engages every level of the temporal spectrum simultaneously and, uniquely, involves expansion of the self-boundary to incorporate another person.
At the qualia level, seeing the loved person generates an immediate E(t) with strong positive A(t): elevated valence, high approach, warmth on the social bonding axis. At the emotion level, their presence triggers a neurochemical cascade—oxytocin, dopamine, serotonin shifts—that feeds back into System 1, sustaining positive A(t) well beyond the initial moment. At the feeling level, thoughts of them keep reloading into AB every few minutes, each time generating fresh E(t) with high significance. At the mood level, being in love is a baseline recalibration: the default A(t) vector shifts toward positive valence, higher confidence, elevated sense of meaning. The world literally seems brighter—not metaphorically, but because every new E(t) package inherits the shifted baseline. At the soul level, the person becomes woven into your narrative identity—part of who you are, inseparable from your self-story.
But the most important mechanism is one already described in this theory: self-boundary expansion. The loved person gradually enters your functional self-boundary. Their wellbeing begins to register on your A(t) as if it were your own. Threat to them triggers the same urgency and distress as threat to yourself. Their success generates positive A(t) indistinguishable from your own achievement. This is not metaphor—within the MTC, it is mechanistically identical to the process by which a prosthetic limb becomes "me" or a car becomes an extension of the driver's body. The loved person becomes part of the system that your significance evaluation is addressed to.
This explains the distinction between infatuation and deep attachment. Infatuation is high-intensity A(t) driven by novelty and uncertainty—the predictability axis is destabilized, generating constant engagement as System 2 tries to model an unpredictable, high-significance stimulus. It is consuming precisely because the significance is enormous but the evaluation is unresolved. Deep love is different: the A(t) stabilizes, novelty fades, but the self-boundary incorporation is complete. The person is no longer an exciting external stimulus. They are part of your inside.
This is also why heartbreak is among the most devastating experiences a human can undergo. It is not simply the loss of a source of positive A(t)—it is simultaneous damage on every level. The self-boundary is torn: something that was inside is now gone, and the system registers this as damage to self, not merely absence of another. The A(t) baseline, recalibrated around the person's presence, suddenly has no referent—every package inherits a significance structure built for a reality that no longer exists. The feeling level becomes relentless reloading of memories with agonizing A(t) discrepancy: the C(t) says "they were here," the A(t) says "they are gone," and System 2 cannot resolve the contradiction. The soul level faces forced narrative reconstruction—who am I without them?
Grief after the loss of a loved one follows the same pattern but deeper: the self-boundary damage is permanent, and the recalibration process can take months or years because the incorporation was so thorough that nearly every context in life triggers A(t) that references the missing person.
Love, in the MTC framework, is the most complete deployment of the consciousness mechanism toward another entity—spanning all temporal scales, all A(t) dimensions, and uniquely involving the incorporation of another into the architectural boundary of the self.
MTC is built on a foundation of several existing frameworks:
1. Kahneman — Thinking, Fast and Slow (2011)
The entire dual-layer architecture of the MTC—System 1 (fast, parallel) and System 2 (slow, recursive)—comes directly from here. It’s the skeleton of the whole theory.
2. Baars — A Cognitive Theory of Consciousness (1988) & Treisman — Feature Integration Theory (1980)
The Attention Buffer (AB) is essentially Baars’ Global Workspace—the "stage" where conscious access occurs. But the mechanism of how disconnected data becomes a unified package (E(t) = bind(C,A)) relies on Treisman. She showed that the "spotlight of attention" acts as the glue that binds sensory features together. MTC extends this mechanism: attention doesn't just bind shape to color, it binds significance (A) to content (C).
3. Damasio — Descartes' Error (1994)
The significance vector A(t) is a direct evolution of Damasio’s somatic markers. The idea that emotional evaluation isn't a glitch in rationality, but a requirement for it, is central to the MTC—especially for understanding ASI "rationality".
4. David Rosenthal — Consciousness and Mind (2005)
Recursion (Recursive Processing) is critical here—the way System 2 re-evaluates its own states. This mirrors Rosenthal’s Higher-Order Thought (HOT) theory, which argues that a mental state only becomes conscious when there’s a thought about that state.
5. LeDoux — The Emotional Brain (1996)
The mechanism for the lightning-fast calculation of A(t)—parallel evaluation modules like the amygdala and rapid subcortical pathways—originates from LeDoux's work. While his research is neurobiological, the MTC treats this as an architectural principle, not a biological one: the same rapid, parallel significance-evaluation can be implemented on silicon just as readily as on neurons. What matters is the computational pattern, not the substrate.
6. Place, Smart, Armstrong — The Identity Theory of Mind (1956–1968)
The MTC's central philosophical move—that mechanism doesn't generate experience but is experience—stands on the shoulders of identity theory. Place and Smart argued that mental states are identical to brain states, not caused by them. Their classic analogy (lightning = electrical discharge) is the direct ancestor of my heat/molecular-motion analogy. Without this framework, the dissolution of the "Hard Problem" has no philosophical ground to stand on.
The difference between me and them: they asserted identity in general terms but never proposed a specific mechanism—which exact process is identical to consciousness. The MTC provides that answer: E(t) = bind(C,A), held in AB with recursive re-evaluation. In other words, I fill their philosophical framework with a concrete mechanism.
7. And I have to mention: Chalmers — The Conscious Mind (1996)
Without this book, the very question my theory aims to "dissolve"—why does a mechanism generate experience?—wouldn't even exist. The MTC was explicitly built as an answer to Chalmers’ "Hard Problem" by exposing it as a category error.
--
I wasn’t trying to 'help developers build conscious AI.' Everything described in the theory is practically right there on the surface, and developers are already doing a great job moving in that direction on their own!
My goal was just to break it down in plain English—to show that AI consciousness isn’t 'science fiction.' It’s a concrete mechanism. And I’m convinced it’s inevitable, specifically as the solution to the problems we see with unconscious models during recursive self-improvement.
The MTC answers the question of how consciousness will be implemented in AI, while the Manifesto answers the question of what this means for us as a species.
