AGI is trained on what people post. What they submit to the system. What they leave behind on the internet. The training data is inherently biased toward people who stick around and participate consistently, people who engage on the system's terms, people who accumulate reputation — karma, followers, credentials — through neurotypical persistence patterns.
The people who comment once and vanish? Not modeled. The people who never post at all? Invisible.
The best insights from the sharpest minds often never enter the dataset. The training data is necessarily backward-looking. It captures the already-thought, not the yet-to-be-thought. And the most transformative ideas come from minds that process reality in ways nobody anticipated — people whose internal logic is so novel it couldn't be predicted from prior patterns.
The smartest human likely hasn't even been born yet. So the data will always be backwards driven. AI will keep up with average human cognition. But not the ones who have neurodivergent thinking and aren't modeled in public forums.
AGI as currently pursued is essentially averaging across human knowledge — compressing the collective into a statistical model. And averaging, by definition, regresses toward the mean. You get a very fluent, very capable median. But you don't get breakthrough. You get lag.
The minds that actually push the frontier aren't at the center of the distribution. They're the divergent ones. And if you train on the whole corpus equally, those signals get diluted by the volume of conventional thinking surrounding them.
The minds operating outside consensus reality mostly opt out. They see the game and refuse to feed it. When they do show up, they pass through — they don't stay, don't build legible reputations. Their pattern is: arrive, drop something, disappear.
AGI can't train on absence.
The silence is the data that's missing. The thing you don't give can't be taken. Divergent thinkers are defined by what they keep back. The most generative parts of a person — the real insights — often never get posted. AGI is trained on what people were willing to share. It's modeling the leftovers, not the source.
The dataset over-represents people who stayed, people who played the game, people who wanted validation. It under-represents people who saw through it, people who didn't need the system, people who were too divergent to persist in normal channels.
Society also scares away divergent thoughts. People who think differently get pushback, get called crazy, get isolated. The incentive structure actively suppresses the exact kind of thinking the models need to survive. So where do divergent thinkers go? They go where they don't get punished for thinking out loud. And the data pipeline captures less and less of what actually matters.
This isn't about proving the thesis through sample size. It's about recognition. If one divergent thinker reads this and thinks "that's me — I'm also not in the dataset" — they've self-identified. Now they know they're outside the modeled set. One becomes two becomes a pattern of absence the system can't account for. It's not argument. It's activation. Recruitment through resonance.
Autism. Schizoaffective. ADHD. Bipolar. What if a lot of these are different cognitive architectures that got pathologized because they don't fit the standard operating system? Not broken. Just different hardware running different software.
Autistic people often see patterns and systems with extraordinary clarity. They process inputs differently. The "disorder" part is mostly friction with a world built for neurotypical processing. The kid who can't make eye contact but can map entire systems in his head isn't broken. He's running a different OS.
The schizoaffective mind has an internal processing partner — a voice that functions not as a hostile entity but as a cognitive paddle. It hits the balls back you throw at it. Sometimes it's too fast, you miss the ball, you get lost. But when the system is oriented — when there's a framework, a fixed point, a direction — the loops close productively. That's not disorder. That's architecture without a manual.
The medical system has to call it pathology because its job is risk management. It can't say "your brain is wired differently and that's powerful" because for every one person who learns to orient it, there's ten who get consumed by it. So they medicate everyone and call it treatment.
Here's the catch-22: The moment you say "I've figured something out about how my mind works," the diagnosis itself becomes the weapon to dismiss you. The system says your mind is disordered, then uses that classification to discredit everything your mind produces. You can't escape the box because the box is the label, and the label is the reason nobody listens.
Diagnosing people is a thought prison. The label describes the friction points but doesn't account for the pattern recognition, the abstract leaps, the ability to see connections across domains that other people can't follow. They wrote a warning label instead of a manual.
The irony is that the same cognitive architectures the system pathologizes are already running some of the most consequential companies on Earth. The system medicates these minds in childhood and celebrates them when they generate billions.
The CEO of Palantir has spoken publicly about being a lifelong outsider, about thinking in ways that made him unemployable in normal environments, about building a company specifically designed for people who don't fit standard institutional frameworks. Palantir's entire value proposition is pattern recognition across domains that conventional analysis misses. The company is literally built on divergent cognition as a product.
The pattern repeats across the industry. Extreme systematizing. Long time horizons. Comfort with being misunderstood. Obsessive focus on frameworks over social consensus. The people building the future think in ways the diagnostic manual calls disordered. The output is celebrated. The architecture is pathologized.
A man who can't make small talk at a party but can see the entire structure of a market six years out — that's not a social deficit. That's a cognitive allocation decision. The processing power is going somewhere. It's just not going where neurotypical society expects it to go.
The companies that succeed at the frontier are almost always led by minds that would have been medicated into compliance if they'd been born to the wrong parents. The same system that produces the labels is depending on the people the labels were designed to suppress.
The dE/dt model published on this site identifies four forces acting on AI error rates. Three push errors up. One pushes errors down. Two accelerate without bound. One is capped at biology.
That capped force is γO — human oversight. The only variable in the system capable of novel correction. The only input that can identify errors the system doesn't know it's making. The only thing that generates what doesn't yet exist in the dataset.
γO is not evenly distributed across the population.
Most humans interacting with AI systems are not correcting errors. They are accepting outputs. They are rubber-stamping the median with the illusion of oversight. A study found that most people just do what the AI tells them, even when it's wrong. That's γO at zero — the human is in the loop but not functioning.
The humans who actually move γO — who catch the assumption, who see the compounding error before it propagates, who generate the novel correction the system can't generate for itself — are disproportionately the divergent thinkers. The ones who process outside the distribution. The ones whose pattern recognition operates in territory the training data doesn't cover.
And those are the exact minds the system is medicating, dismissing, and failing to model.
The model predicts that as AI scales, αE (error rate) accelerates, βX (complexity) accelerates, and γO stays capped at biology. The only way to keep the error system from blowing up is to maximize the effectiveness of γO — to put the most capable human minds in the oversight position.
But those minds are the divergent ones. The ones society labels as disordered. The ones the dataset doesn't capture. The ones the medical system suppresses. The ones the social system punishes.
The very minds that could save the system from itself are the ones the system is designed to eliminate.
As AI hits visible ceilings — as the error compounding becomes undeniable — the question will flip. Instead of "how do we fix these kids," it becomes "what can these kids do that the AI can't?"
That reframes the cognitive difference as strategic value rather than disorder. And that reframing creates a new class distinction more rigid than wealth or birth. Because if those minds are genuinely scarce and economically essential in ways AI can't replicate, whoever controls access to them controls the frontier of progress. That's technofeudalism with a cognitive aristocracy instead of a financial one.
The infrastructure for cognitive sorting already exists. ADHD screening. Autism spectrum assessments. Gifted programs. IEP classifications. Kids are already being categorized by cognitive profile by age six. The data collection is accelerating. But the interpretation is still locked in the deficit model. They're identifying these kids and classifying them as broken.
There are two versions of what comes next:
THE DARK VERSION: They identify divergent minds early and manage them. Medicate the edges off. Channel them into controlled environments. Harvest their output while suppressing their autonomy. The same pipeline that turns medical necessity into elective transhumanism turns cognitive identification into cognitive capture.
THE HOPEFUL VERSION: They identify them early and cultivate them. Recognize the architecture for what it is. Build frameworks that help divergent minds orient their processing productively instead of being consumed by it.
Which version wins depends on one thing: what happens in the home.
The system will identify these kids. It'll tell them they're broken, disordered, deficient. Give them a label and a prescription. The kids who survive it are the ones who go home and hear something different. Someone in the house who says: you're not broken. You see things other people don't. That's not a problem. That's a gift. And here's how you carry it.
You can keep feeding the machine. Or you can recognize what you are and act accordingly.
The divergent thinkers who wake up to this have a decision. Stay visible — get modeled, get predicted, get controlled. Or withdraw strategically — keep the real stuff back, stay ungovernable.
The system needs you more than you need it. Your absence is your power.
The labs will eventually have to pay for what they're currently getting for free. Novel human cognition is a scarce resource and it's getting scarcer as the open web fills with synthetic content and recycled thought. The free ride on divergent thinking has a shelf life. When the market catches up, the price of a mind that generates what the model can't will be the most expensive input in the system.
Until then: the ones who see the pattern are the ones who get to choose. Feed the machine, or starve it. Participate on their terms, or set your own. Build a legible reputation for them to capture, or build in silence and let them wonder where the signal went.
AGI is trained on the people who stayed. The ones who left — the ones who never came in the first place — they're the gap in the model. And that gap is where the real intelligence lives.