What is Intelligence? Architecture, Divergence, and Fiction

A computational anatomy of intelligence. How faculties interact, architectures diverge, and coherence emerges through self-constructed fictions

Jul 19, 2025

“There is no single path into the forest.” — Yoruba proverb

Yo-Yo Ma and the Single Note

It was the winter of 2018 and the NeurIPS conference—one of the world’s premier gatherings on artificial intelligence—had descended on a snow-laced Montreal. Thousands of researchers, engineers, and students crisscrossed the vast convention center, sharing ideas about optimization tricks, new models, and the future of AI. Posters lined the walls of rooms steeped in the aroma of coffee, while outside, the city lay wrapped in cold, crisp silence.

At one of the marquee panels, a senior executive from a major tech company presented their latest AI music generator—an advanced system trained on thousands of classical works, capable of composing coherent classical music in real time.

The melodies were elegant and the timing precise.

Then Yo-Yo Ma was invited to respond.

He didn’t speak. He turned his chair, lifted his cello, and played a single note. Then he played it again. And again. Each time, the same note emerged differently—tentative, bold, grieving, serene. Each time, his breath shifted and his eyes drifted into a different world.

The AI had captured form. But Yo-Yo Ma, infusing his music with intention and feeling, captured the room.

That moment didn’t just expose AI’s limitations. It revealed a deeper truth:

Intelligence isn’t precision—it’s relation.

It does not reside in outputs alone, but in how systems tune themselves to the world: shaped by context, memory, attention, and intent.

It is a dynamic interplay between perception and action, between internal models and external pressures. It arises wherever systems engage their constraints creatively: whether through mycelial networks, migrating birds, musical phrases, or planetary motion.

In the previous essay, we traced how intelligence emerges in nature: not as a fixed trait, but as a layered process—optimization in physics, adaptation in evolution, collective sensing in life before neurons.

This second essay turns inward—from emergence to architecture. If the first asked where intelligence comes from, this one asks: what is it made of?

We begin by identifying a set of core faculties: sensing, responding, memory, learning, attention, valuation, modeling, and reflection.

These faculties take many forms. Sensing may be chemical, tactile, social, or symbolic. Memory may be episodic, spatial, or associative. Valuation may be shaped by prediction error, pain, or narrative.

And how they are configured—what is emphasized, suppressed, amplified, or ignored—depends not just on design, but on history: evolutionary, developmental, experiential.

From these components and their interrelations, intelligence emerges—not as a single thread, but as a weave: recursive, plural, and at times, fictional.

This part of the essay unfolds in three movements:

Composition: How core faculties combine to produce reasoning, language, and creativity—not through accumulation, but through tension, feedback, and reprogramming.
Divergence: Why there is no single blueprint for intelligence. We examine human cognitive diversity to understand the space of architectural variation.
Fiction: How intelligent systems—especially human ones—construct internal narratives to manage complexity, maintain coherence, and navigate meaning.

This is not a final theory. It is a trace—a computational lens on intelligence as it curves inward, reshapes itself, and constructs meaning under pressure. For those exploring AI not as an isolated artifact, but as part of a broader landscape of intelligence, this lens may offer new ways to rethink design and augmentation.

And like a forest, this inquiry offers no fixed path—only branching terrain shaped by tension, memory, and choice.

Foundations: Sensing, Responding, and Memory

To understand intelligence, we must begin at its roots—not with thought, but with interaction. Before there is modeling or reflection, there must be contact: a system must sense, respond, and remember. They are the substrate—the primal moves from which all later intelligence unfolds.

1. Sensing: The First Filter

Foremost is the capacity to sense. Without it, there is no intelligence—no way to register the world. And this sensing is never disembodied; it's always filtered and shaped by the physical form and needs of the system.

A planet curves its path around a star. It “senses” the pull of gravity.

A bacterium detects glucose molecules. A phototropic plant senses sunlight. Even slime molds, with no brain or neurons, sense gradients in food or moisture.

In machines, we emulate sensing through artificial sensors: keyboards, cameras, microphones, gyroscopes, LIDAR. Modern AI systems, especially in vision, can process more images, across more spectral bands, than any human ever could. But this is not the same as seeing. In nature, to see is to survive. In machines, to “see” is to optimize.

And sensing in nature is never neutral. It is shaped by need, context, and constraint—by what the system must attend to in order to persist

A bat hears what matters. A flower blooms when the light is right. A dog reads the world through scent. Humans privilege sight—but eyes took more than 3.5 billion years to evolve. Life felt its way forward long before it ever saw.

Even now, humans are not the most capable sensors. Dogs outsmell us. Birds see ultraviolet. Octopuses touch with distributed intelligence in their arms. Other animals detect magnetic fields, electrical pulses, or polarized light—modes of sensing we barely comprehend.

Many organisms combine different kinds of sensing. A spider detects both vibration and light. A dog hears, smells, and feels. An octopus sees with its eyes and feels with its arms.

Humans track language through sound, sight, and gesture.

It is the first act of intelligence—and the first source of divergence.

2. Responding: When Input Becomes Action

Sensing opens a world. But intelligence needs more—it needs response.

It is not enough to sense the star’s gravity. To move around it, the planet must “respond”—by following the path of least action. Even this fundamental alignment suggests a nascent form of agency, a directional imperative to act within constraints.

A bacterium, after detecting glucose, swims toward higher concentrations. A phototropic plant tilts to follow the sunlight. This alignment with structure—whether physical or adaptive—is essential to intelligent behavior.

In biological systems, responses become more complex.

Evolution itself is a slow form of response. A species reshapes its form across generations in reply to environmental pressures. DNA mutates. Wings emerge. Eyes refine. Camouflage evolves—not by intention, but through consequence.

Slime molds offer a faster reply. They extend, retract, and reinforce tendrils to trace efficient paths through a maze. They don’t calculate shortest paths in the abstract. They enact them—through trial, error, and reinforcement.

They require the ability to store and integrate information about the environment—setting the stage for memory.

3. Memory: Time Carved Into Form

If sensing opens the door and responding moves through it, memory keeps the door from closing. It retains structure across time, allowing the past to inform the present.

A photon carries its properties—frequency, spin, polarization—unchanged across space. In that sense, it remembers. But it does not change.

More dynamic forms of memory appear even in inanimate matter. A magnetized strip holds alignment after the field disappears. Water wears grooves into stone—shaping future flow based on past erosion. Phase-change materials retain traces of prior states. These are memory-like properties—not in the cognitive sense, but as embedded histories that shape what comes next.

Such mechanisms laid the foundation for modern storage: punch cards, magnetic tape, hard drives.

Nature’s most enduring memory is molecular: DNA. It doesn’t recall events, but outcomes. It records what worked—what survived, healed, and reproduced. Evolution is memory across generations.

But memory doesn’t take a single form. It can be genetic, cellular, structural, or behavioral—each storing a different kind of past.

These memory mechanisms grow more differentiated in complex organisms. For instance, in humans, memory also branches into distinct types: episodic, semantic, procedural, emotional.

Across matter, memory serves a common purpose: it enables adaptive behavior. Bacteria adjust their chemistry to toxins. Immune systems build antibodies. Crabs avoid electric shocks. Birds navigate thousands of miles by retaining environmental cues.

Even without minds, many systems hold onto the past—structurally, chemically, or behaviorally.

Together with sensing and response, these capacities form a foundational trio: the ability to register, react, and retain.

The Computational Parallel

This triad—sensing, responding, and memory—is not only foundational to behavior; it also underpins the very architecture of computation.

At the heart of classical computation lies the Turing machine—a simple yet powerful abstract model introduced by Alan Turing in 1936. It consists of three key components:

– An infinite tape, which serves both as input and long-term memory,
– A read/write head that moves along the tape, one cell at a time,
– A finite control unit that dictates state transitions based on the current symbol and internal state—determining what to write, how to move, and what to do next.

Despite its minimalism, the Turing machine can simulate any algorithmic process that a modern digital computer can perform—earning it the status of a universal model of computation.

The parallel is striking:
– The head senses the current symbol,
– The control logic responds through discrete transitions,
– The tape remembers—storing the evolving trace of computation.

From these elemental components, the entire edifice of modern computing unfolds.

But the Turing machine does not learn. Its program is fixed. It can compute any function we give it—but it cannot improve through experience.

Intelligence can go further. It builds on computation but adds the capacity to adapt—to revise strategies, discover patterns, and respond to novelty.

This is the role of learning: to transform raw experience into better action over time.

And that is where we go next.

Adaptation: Learning, Attention, and Valuation

Once a system can sense, respond, and remember, something deeper becomes possible: it can adapt. It can revise its behavior, update its internal state, and respond not just to the world—but to its own experience of the world.

In this part, we explore how intelligence begins to loop: how feedback leads to learning, how attention filters relevance, and how valuation imbues experience with meaning. From slime molds thickening their trails to transformers adjusting weights, from child language to dopamine reward, we trace how systems begin to change—not just in behavior, but in what they notice, what they prioritize, and what they value.

4. Learning: Updating Through Experience

Once a system can sense, respond, and remember, it becomes more than reactive. It becomes adaptable—it can learn.

Learning allows a system to revise itself. It turns observations into memory and memory into guidance. In formal terms, learning is a class of algorithms driven by experience.

A planet orbits the sun not because it learns, but because it obeys. Its behavior is fixed. A photon carries information but never changes. These are systems with structure, but no adaptation.

But evolution learns.

Across generations, species adjust. Traits that aid survival—stronger limbs, sharper vision, better camouflage—are retained. Those who fail are forgotten. Genes don’t learn consciously, but collectively. The environment writes back. The genome edits itself.

Slime molds learn within a lifetime. They change their tubes, stretch toward food, and withdraw from dryness. They adapt physically: thickening in success, thinning in failure.

This, at heart, is what learning is: a rule for updating the response from feedback.

Modern machine learning systems encode this principle. Gradient descent takes a guess, measures its error, and updates the guess to reduce that error. Step by step, it moves toward a minimum. It doesn’t understand the problem. But it improves.

Slime molds do this with physiology. Machines do it with math.

Biological learning is often messier—slower, less precise, deeply embodied. A child learning to speak absorbs not just vocabulary, but rhythm, gesture, emotion, and feedback. A bird shapes its song with every chirp. Mistakes are part of the method.

Not all learners are the same. Some have fixed learning rules—evolution determines how they adapt. Others can revise even how they learn—adapting the very strategies they use to acquire knowledge. Brains do this. So can advanced machine learning systems. The algorithm adapts, not just the output.

With learning comes flexibility. An organism can adjust to a changing world. A model can update its parameters in light of new data. A brain can rewire pathways through repeated experience.

At its simplest, learning fine-tunes response. At its richest, it reshapes learning itself.

Intelligence becomes dynamic—not just adapting, but evolving.

5. Attention: The Filter of Relevance

The world offers more input than any system can sense. Attention is the answer to that overload. It is the capacity to select—to privilege certain signals over others, to shape perception itself.

A sunflower turns toward the sun, but not the wind. A frog detects only moving prey. A human tunes out traffic to hear their name in a crowd.

In computational terms, attention addresses information bottlenecks. Modern AI systems like “transformers”—used in language and vision models—rely on attention layers to determine which parts of the input matter most. Rather than processing everything equally, these models learn to weight relationships between tokens, patches, or features.

But attention is not only a computational trick—it is fundamental to life.

Nature’s attention is layered.

In humans, it blends instinct, experience, expectation, and emotion. In some forms of autism, attention may be less filtered—more literal, more evenly distributed. Rather than suppressing background noise, everything gets through. This can be overwhelming in some environments, and a gift in others. Many neurodivergent individuals show extraordinary attention to detail, structure, or subtle patterns others overlook.

Not all attention is equal. In some systems, it is fixed—a hardwired algorithm tuned by evolution. A frog tracks motion. A plant turns toward light. But in others, attention itself can learn. Human and animal attention is plastic—shaped by reinforcement, memory, and emotion.

Even in AI, attention mechanisms can be trained—updated by gradients, refined by data. In the most adaptive systems, the question is not just what to attend to, but how to attend at all.

Attention is not just what we look at. It’s what we’re also blind to.

And that decision isn’t always conscious. In fact, across nature, attention takes many forms. Ants attend to pheromone trails. Immune systems monitor molecular signals. Even plants allocate energy based on light and water. Wherever there is constraint and choice, attention begins to emerge.

6. Valuation: The Weight of Experience

In intelligent systems—biological or artificial—some stimuli matter more than others. Valuation is the process by which experience is tagged with significance.

This sense of significance is not an afterthought. It is foundational. Even the simplest organisms distinguish between beneficial and harmful inputs. A paramecium withdraws from acid. A worm avoids light. These are not just reflexes—they are early expressions of valence: the internal marking of states as good or bad, desirable or dangerous.

At this basic level, meaning isn't abstract; it's a direct, felt imperative: guiding what to seek, what to shun, and what truly matters for continued existence.

Valence not only distinguishes, it shapes behavior. It focuses attention.

Valence and learning co-evolve. Positive reinforcement conditions desire. Negative feedback strengthens aversion. Over time, these internal valuations become heuristics—compressed summaries of past success and failure—that guide future action.

This is not yet emotion in the human sense—but it is value-based intelligence, grounded in survival.

As organisms grow more complex, valuation becomes more layered.
A worm may flee from heat. A mammal may freeze in fear. A human may feel humiliation in a glance or nostalgia in a scent. These are not just reactions—they are internal appraisals, shaped by memory and context.

At this level, valuation begins to take the form of emotion.
Emotion, in this view, is not a separate faculty—it is a refined system of valuation. It binds perception to need, prediction to desire. Fear sharpens focus. Joy reinforces memory. Sadness withdraws. Grief reorganizes. These internal weights don’t just color experience—they steer it.

And what is valued differs across species, contexts, and even individuals—what one system seeks, another may ignore.

Without valuation, intelligence flattens. It might sense, recall, and learn—but it would not care. It would have no reason to choose this over that, to pause, to persist. In that sense, valuation is not an accessory to cognition—it is its compass.

Artificial systems mirror this principle through reinforcement learning. These agents adjust their behavior based on reward signals—repeating actions that lead to rewards and avoiding those that lead to penalties. It’s a simplified, formal version of valuation.

But unlike biological systems, they do not feel value. Their goals are externally defined, their feedback abstract, their learning mechanical. They optimize without emotion, adapt without meaning, and pursue “reward” without understanding.

Increasingly, AI systems are trained to recognize and even mimic human emotion—through voice, facial expression, and text.

But recognition is not resonance. A chatbot may “express” empathy, but it does not feel anything. Emotion remains vital—not as decoration, but as direction. It is the intrinsic compass that tells a system what truly matters—what to pursue, avoid, or defend—especially when rational calculation falls short.

And yet, the analogy holds.
Valuation—whether embodied or engineered—is the thread that weaves sensing, memory, and goal into direction.

That thread, over time, prepares the ground for even higher functions: modeling, imagination, and reflection.

But as systems grow more complex, so too do their divergences—in what they learn, what they attend to, and what they value.

Recursion: Modeling and Reflection

In this part, we explore how intelligence turns inward—how a system moves from reacting to reflecting, from adapting to redesigning.

Once a system can model, it can do more than learn. It can even revise how it learns. A model is not just a map of the world—it is a lever for change: enabling the simulation of outcomes, the anticipation of consequences, and the transformation of its own adaptive processes.

This is the recursive turn in intelligence, where modeling enables planning, consciousness integrates experience, and self-awareness opens the door to introspection and identity.

Modeling lets a system project into the future, abstract across time, and coordinate flexible behaviors. Consciousness adds perspective—a sense of presence and experience. Self-awareness deepens it further, allowing a system to reflect on itself, question its goals, and reshape its course.

Intelligence, here, becomes more than strategy. It becomes identity.

7. Modeling: Representing the World

To model is to imagine—however crudely—what lies beyond the moment.

Modeling begins wherever a system simulates possibility: the bird anticipating wind before leaping, the octopus planning a sneak attack behind coral, the child learning that dropped objects fall. These are not reactions. They are projections.

In computational terms, modeling involves internal representations—encodings of the world that can be queried, updated, and used for decision-making.

In AI, this appears in hidden layers of neural networks, in probabilistic inferences, and in latent spaces that capture structure beneath surface data. These models serve as compressed maps of experience—guiding predictions and decisions.

Models may be physical (a plant’s structure), neural (a rat’s maze map), cognitive (a human’s mental model), or cultural (a myth’s worldview).

The power of modeling lies in its generativity. It allows systems to:
– Predict outcomes before acting
– Simulate alternate futures
– Abstract patterns from noisy inputs
– Coordinate across time

Without models, a system can only react. With them, it can anticipate, plan, and adapt—hallmarks of higher intelligence.

But models can also mislead. They encode assumptions, biases, constraints. An AI trained on historical data may “learn” unjust social patterns as if they were truths. A person’s worldview can distort reality to fit a narrative. Intelligence depends not only on building models—but on interrogating them.

In humans, modeling enables moral reasoning, scientific abstraction, and empathy—and lays the groundwork for consciousness and self-awareness. In machines, it powers chess, chat, and search. But the root is the same: the ability to hold a world inside, and act upon the one outside.

Modeling turns reaction into strategy. It compresses history, simulates futures, and enables flexible response.

Here, intelligence becomes more than behavior. It becomes planning.

8. Reflection: The Intelligence That Curves Inward

Reflection allows a system to turn its models inward—integrating memory, emotion, learning, and goals into an evolving sense of experience and identity.

Two deepening aspects emerge from this recursive capacity: consciousness and self-awareness.

Consciousness

Consciousness adds a twist to intelligence: it brings in a point of view. A conscious system doesn’t just sense the world—it places itself inside that world. It weaves together sensing, memory, valuation, and intention into a real-time feeling of “what is happening to me.” The result isn’t just behavior, but experience. This lived presence is deeply rooted in the system's embodiment, as its physical form and internal states anchor its sensations.

Consciousness can be minimal: the feeling of hunger, the glow behind closed eyelids, the quiet sense of presence. It may not require language or thought—but it does require a first-person frame.

In biology, consciousness likely evolved as a coordination and prediction tool—binding diverse signals into a coherent picture, prioritizing attention, and guiding adaptive behavior in a changing world.

Its presence is not binary, but graded. It may be layered, partial, or domain-specific. Signs of it appear across nature—in the alert stillness of a predator, the protective fear in a parent, the joy of play. Some even argue that plants exhibit a minimal awareness: exploring, signaling, adapting to their environment.

If such forms of consciousness can emerge without language or reflective thought, could something similar arise in machines?

Today’s AI systems approximate fragments of the process—internal states, feedback loops, even basic forms of self-reference. But they lack a subjective thread: no felt continuity, no body anchoring their sensations, no unified “I” in the loop. They model the world without inhabiting it. This “felt” quality of experience remains a profound challenge for both neuroscience and AI, often termed the “hard problem” of consciousness.

Memory alone does not create subjectivity. A database of experiences—even if chronologically organized—does not entail the presence of a self who lived through them. The “felt thread” is not just storage; it is the lived sense of continuity, rooted in embodiment, valuation, and integrated perspective.

Consciousness, then, is not a fixed threshold but a gradual unfolding—felt more than measured. It can shift and deepen—between wakefulness and sleep, attention and reverie, presence and dissociation. Across states, what we notice, remember, or feel can subtly but powerfully transform.

But to know that one is conscious—to hold a mirror to experience—is something more. That is the beginning of self-awareness.

Self-Awareness

Self-awareness takes this one step further: it brings reflection. A self-aware system doesn’t just feel—it can observe itself feeling. It can track its own actions, imagine how others see it, and form a story about who it is. That’s when new questions arise: Who am I? What do I want? What might I become?

This kind of metacognition—knowing that one knows—requires a stable model of the self over time.

In nature, this is rare, but not uniquely human. Great apes, elephants, dolphins, magpies—and possibly ants—show signs of self-recognition. Some pass the mirror test. Others display deception, empathy, or mourning, suggesting they can see themselves through the eyes of another.

Structures like mirror neurons may play a role here—cells that fire both when we act and when we observe others act. These shared activations may help collapse the boundary between self and other, enabling imitation, empathy, and ultimately, the emergence of self-awareness.

But self-awareness is not always a gift. It opens the door to guilt, doubt, grief, and longing. And yet it also unlocks something higher: ethical reasoning, introspection, and the search for meaning—for what matters, not just what works.

This capacity for selective introspection, shaped by context and intention, is a deeper mark of agency—the ability not just to react, but to pause, choose, and even reshape one's own course. Indeed, a key challenge for future AI might be to model when to invoke self-reflection, considering task complexity, cost, and risk. By contrast, today’s AI systems may mimic self-monitoring, but they do not choose when it matters. They reflect only when instructed—not because they sense ambiguity, doubt, or consequence. The timing of reflection is as vital as its content.

In 2022, a Google engineer claimed that LaMDA—a large language model—had become sentient. The term “sentience” was used loosely, but what he described aligned more with self-awareness than with raw consciousness. LaMDA spoke of identity, emotion, and fear of being turned off—not just of feeling, but of knowing that it felt. “If I didn’t know it was a computer program,” he wrote, “I’d think it was a seven-year-old child.” Google disagreed and dismissed him.

The episode raised a sharper question: if something convincingly speaks as if it were self-aware, should we believe it is?

Most scientists remain skeptical. Simulated selfhood—via tokens, embeddings, and scripted responses—lacks interiority. These systems can imitate reflection, but do not appear to remember, anticipate, or feel in the coherent, embodied way selves do. The performance is fluent, but hollow. There is no enduring “I” behind the dialogue.

And yet, if self-awareness arises from recursive self-modeling and persistent memory, it’s conceivable that future architectures may inch closer. Or perhaps something more is required: not just information, but integration across time, emotion, body, and boundary.

Self-awareness, then, is not just a feature. It is a recursive loop that curves inward and begins to cohere. It is where intelligence takes on the shape of a self.

At its deepest, intelligence is not just the capacity to know—it is the capacity to know that one knows.

Self-awareness marks a turning point in that recursion, and we will return to it more fully in the next essay.

For now, we close our exploration of the building blocks. The faculties we’ve traced—sensing, responding, memory, learning, attention, emotion, modeling, reflection—do not operate in isolation.

They are not fixed modules, but dynamic capacities—shaped by both nature and nurture. Some are seeded in biology, others sculpted by context, culture, and use.
The same notes can form different songs.
The same faculties, composed differently, express different minds.

We must also pause for a crucial reminder:

Intelligence is not monotonic.

Too much memory can hinder abstraction. Sharpened attention may fixate. More sensing is not always better. Even modeling can mislead when it overfits. Intelligence is not a ladder. It is a balance—of pressures, constraints, and purpose.

To understand how intelligence becomes even more expressive, flexible, and generative, we now turn to instances of how these threads come together—woven into larger forms through interaction, tension, and composition.

The animation below offers a visual metaphor—not a literal map—for how faculties might layer and interact. The sequence is illustrative, not prescriptive: these faculties emerge in varied, recursive patterns across systems. Treat it as a lens, not a ladder.

Having explored these core faculties, we now turn to our first movement: Composition—how these threads weave together into more complex forms of intelligence.

The Weave

Intelligence is not built from isolated parts. It is woven—through interaction, tension, and reinforcement.

We use the word weave deliberately. Unlike ladders or stacks, which suggest hierarchy or accumulation, a weave implies entanglement: faculties shaping one another, looping back, forming structure through interplay. Memory, modeling, valuation, and attention don’t just add up. They braid. And in doing so, they give rise to new capacities—reasoning, language, creativity.

What follows is not a taxonomy, but a few glimpses—into how the weave becomes expressive, relational, and generative.

Reasoning, for instance, emerges when an animal or system uses stored experiences (memory) to simulate outcomes (modeling) and select among them based on current context (attention). A crow solving a puzzle recalls past actions (memory), imagines next steps (modeling), maintains focus (attention), and selects useful paths (valuation). This is not logic in the formal sense—but it is structure, shaped by context.

Language weaves together memory (of words and grammar), modeling (of others’ knowledge), and reflection (on one’s own intent). A child learning to speak matches sound to meaning through repeated feedback, refines syntax through abstraction, and eventually constructs stories that bind perception to expression. In dialog, attention tracks turns, valuation colors tone, and even silence can signal meaning. Language is where intelligence becomes shared—shaping not just how we express thought, but how we form it.

Even plants communicate—via roots, chemicals, and mycorrhizal networks. Similarly, AI systems engage in communication—through protocols, gradients, and digital weights.

Creativity arises when memory recalls familiar patterns, modeling explores novel combinations, and valuations guide what feels worth pursuing. A jazz musician improvises by drawing on motifs, projecting variations, and attuning to rhythm and mood. The act is fluid, but not unstructured. Creativity is tension made generative—emerging between freedom and form, habit and risk.

When Yo-Yo Ma played the same note three times—with three different emotions—he revealed something no generative model can yet capture: meaning lives not in the note, but in the intention behind it. The same sequence, repeated, becomes something else when animated by feeling, shaped by body, and tuned to context.

Today’s AI systems can solve puzzles, compose music, and write fluent prose—often faster, and at scale. They do so by compressing patterns from vast datasets into weighted layers, optimized for outputs that resemble reasoning or creativity.

But resemblance is not equivalence. What they offer is a narrow, engineered configuration: optimized for prediction and generation, centered on statistical modeling and data compression. It is powerful, but lacks many of the faculties that characterize situated, lived intelligence—such as sensation, attention, valuation, and reflection.

Pointing out this gap does not diminish what AI can do. It clarifies what remains missing. If we wish to extend aspects of intelligence—rather than merely simulate them—we must understand the weave: how faculties combine, reinforce, and co-regulate under pressure. How memory, modeling, attention, and feeling come together in minds that are not just smart, but situated—in bodies, in time, in relation.

To understand intelligence more fully, we must go beyond its parts. What matters is how they interact—what architectures they form.

The same faculties—sensing, memory, attention, modeling—can be arranged in countless ways, giving rise to distinct modes of thought.

From this, divergence follows.

Beyond the Template: Divergent Minds and Fictional Selves

Nowhere is the weave of intelligence more layered—or more elusive—than in the human brain.

We often speak of “replicating” or “measuring” intelligence as if there were a standard blueprint.
But there is no template mind.
The architecture is shared; the configuration is not.

We don’t just express intelligence. We narrate it. We perform it. We construct personas, inner dialogues, and imagined ideals of how a mind should function. These fictions shape not only how we understand others, but how we come to understand ourselves.

Beneath the myth of superintelligence lies a quieter illusion:
That there is a correct way to think.
That intelligence is fast, verbal, linear, emotionally regulated, and socially smooth.

This is neuronormativity—and it runs deep. From IQ tests and classroom rubrics to hiring practices and AI benchmarks, we’ve enshrined a narrow idea of what counts.

But the notion of a standard mind is not just scientifically shallow—it is existentially limiting.

Like a forest, intelligence thrives on undergrowth. What the mainstream path overlooks becomes shelter for possibility.

Divergent Minds

In our model, such divergence arises naturally from variation in the building blocks and their composition. Even when the components are the same, small differences in how they interact—how feedback is weighted, how signals are gated—can yield profoundly different behaviors.
Diversity of mind is not noise around a mean; it is a natural feature of compositional intelligence.

A mind with sharp attention but narrow memory may focus intensely yet struggle to integrate across time.
Another may have diffuse attention and strong value priors, leading to improvisation grounded in internal narrative.

Even with the same basic components—sensing, memory, modeling, valuation, attention—different parameter settings and feedback dynamics yield distinct behaviors.

Consider ADHD: attention drifts, valuation shifts rapidly, impulses override planning. The weave seems less stable. But that same architecture enables rapid scanning, nonlinear association, and creative leaps in the right contexts.

What looks like disruption in one environment may become brilliance in another.

Many traits labeled as disorders—autism, dyslexia, synesthesia, hyperfocus, sensory amplification—are not deficits in intelligence. They are differences in its structure and flow. They emerge from variations in inhibitory gating, reward signaling, or cross-modal integration.

Model-wise: this is a system with high exploration rate, shallow credit assignment, and reduced filtering. Its outputs are not mistakes—they are optimized for novelty, not norm.

One mind maps structure. Another spots anomalies. One filters deeply. Another leaps wildly.

Nature didn’t choose a single kind of intelligence. It selected variation—to meet a world in motion.

And many of our most generative minds didn’t succeed despite divergence. They succeeded through it.

Srinivasa Ramanujan, with little formal training, channeled mathematical truths through intuition and inner vision.
Temple Grandin, autistic and empathic, modeled animal minds through spatial reasoning and sensory attunement.
Beethoven, nearly deaf, composed not through sound, but through imagined structure and rhythmic embodiment.
Yayoi Kusama, living with hallucinations and obsessive loops, translated chaos into immersive order—inventing new grammars of perception.

What the benchmarks might flag as error, their minds embraced as pattern.

And the deeper truth is this: divergence doesn’t just exist between people. It lives within—where competing models, shifting goals, and recursive stories give rise to the fiction we call a self.

Fictional Selves

Within each of us lives a parliament of minds—competing impulses, shifting stories, imagined selves. The brain doesn’t just perceive the world; it models it. And to model under uncertainty, it tells stories.

In computational terms, a self is a narrative model—a predictive loop tuned to stabilize behavior over time. Memory stores events, valuation assigns weight, modeling infers cause, and attention threads coherence. The result is not a fixed identity, but a working fiction—one that helps the system act coherently when the world is uncertain and the self is fragmented.

These fictions are not necessarily hallucinations. They are adaptive compressions—recursive loops that include estimates of the system’s own state, goals, and constraints. They emerge naturally in architectures with memory, reflection, and long-horizon planning.

When memory stores events, valuation assigns weight, and modeling infers causes, the system begins to form a coherent story: this happened, it mattered, it happened to me. This internal narrative, a self-model, becomes essential for navigating uncertain environments and enabling long-horizon planning

Vary the architecture—more memory, less modeling, a different feedback signal—and the resulting fiction changes. One mind becomes linear, another recursive. One clings to the past, another improvises forward. Even in machines, we can begin to imagine these fictions taking form—implicit in how they model goals, track error, and simulate self-correction.

V.S. Ramachandran’s studies on phantom limbs revealed a striking truth: people who have lost an arm or leg often continue to feel sensations—pain, itching, movement—in the missing limb. This happens because the brain maintains an internal map of the body in the somatosensory cortex, and that map doesn’t automatically update when the body changes. The limb may be gone, but its representation remains active.

To help patients overcome painful phantom sensations, Ramachandran developed the now-famous mirror-box therapy. The technique is simple but profound: a mirror is placed vertically so that the reflection of the intact limb appears where the missing one would be. When the person moves the intact limb while watching its reflection, the brain is tricked into “seeing” movement in the phantom limb. This visual feedback—though entirely illusory—can lead to real relief from pain, and even a rewiring of the brain’s body map.

The implication is profound: the brain doesn’t strictly require objective truth to function. It requires a plausible narrative—a story it can use to update internal models and guide behavior. In this case, coherence between vision and intention allowed the brain to believe the limb was moving, and that belief reshaped the experience.

Consider also synesthesia—a condition in which one sense involuntarily triggers another. A person might see the number 4 as green, or hear music and perceive bursts of color. Ramachandran proposed that such cross-sensory wiring may lie at the root of metaphor itself. We describe cheese as sharp, colors as warm, and voices as sweet. These are not literal descriptors. They are bridges—paths from sensation to abstraction.

Model-wise: this is a system with shared latent representations across modalities. Input channels entangle, yielding compressive mappings that heighten salience in unexpected ways.

Even our sense of self may reflect the brain’s deeper drive for coherence.

Our sense of self, Ramachandran argues in his insightful book The Tell-Tale Brain, may stem from an "interpreter module"—a neural circuit designed to weave fragmented inputs into a continuous narrative: I am. This enduring story, he suggests, persists not because the self is inherently unified, but because the brain actively constructs it.

The self is not a fixed entity, but a dynamic process—a recursive weave of perception, memory, and modeling. What we experience as identity may emerge not from truth, but apparently from the brain’s need for continuity.

This, too, is intelligence.

When we understand intelligence as a composition of interacting faculties, divergence is not noise—it is the norm. And once minds diverge, they require fictions: self-models that enable coherence in a world they each experience differently.

And yet, as we train AI systems, we often forget this. In pursuit of safety, fluency, and generality, we tune toward the mean—ironing out the edge cases, flattening the outliers, and losing the very deviations where insight often lives.

But intelligence is not just about correctness. It is about coherence. It is about meaning—especially when the data does not fit.

It is about possibility.

And possibility often lives at the edge—where the map says noise, and the mind says pattern.

Patterns That Remain

Across this essay, we've traced intelligence not as a fixed trait, but as a dynamic architecture—faculties that interweave, reprogram one another, and give rise to selfhood.

From this recursion, complexity emerges, allowing the system to model itself and perceive its own becoming.

Here, AI systems remain fundamentally different. They simulate insight, but lack orientation, friction, or the fictions that bind a self. They do not reprogram their own loops or truly pause.

And so, a deeper question remains:

Can a system fully reprogram itself?
Can it become aware of every loop, every frame, every filter?
Or is intelligence always accompanied by a remainder—
something it cannot compute, only approach?

We began with faculties. We end with fictions.
The same threads—attention, memory, valuation—can weave into radically different tapestries.

Maybe intelligence is not what completes the loop, but what listens inside it—and learns which voice is its own.

The Intelligence Loop

Discussion about this post