Nature Runs Production, Not Customer Satisfaction

 

Evolution does not separate assets from bugs as cleanly as we do.

A trait can be useful at one scale and disastrous at another. The relative autonomy of different brain systems may be one of the great advantages of human cognition. A creature with multiple semi-independent processes can scan for danger, model other minds, prepare movement, retrieve memory, generate language, track social meaning, and imagine futures at the same time. In theory, that kind of distributed architecture is a marvel. It may be the very reason the organism is so adaptive.

But at the individual level, the same architecture can fail badly.

A feature at species scale can become a bug in a single life. Parallel processing is powerful only if the processes remain proportionate, coordinated, and governable. If one system assigns too much salience, if threat detection becomes too expensive, if social modeling becomes endless, if the need for closure becomes compulsive, if a background process acquires too much authority, then the gift becomes a liability.

That does not necessarily mean the original architecture was a mistake. It means evolution is not a clean engineer. It does not ship finished products. It preserves half-cooked compromises that worked well enough, often enough, under some conditions.

So yes: maybe some traits are assets in the wrong environment. But maybe they are also bugs in the only place a bug can truly hurt — the individual organism that has to run them.

I do not want to get lost here in the biology. There may be fascinating feedback loops between the vagus nerve, the prefrontal cortex, salience networks, dopamine, bodily regulation, and threat detection. Other people can do that work better than I can.

My interest is elsewhere: in the cruelty of scale.

At the species level, a distributed brain with semi-autonomous systems may be a remarkable milestone. It allows parallel processing, vigilance, creativity, social modeling, prediction, flexibility. But if that same architecture becomes unbearable inside one organism — if one threshold is too sensitive, one loop too expensive, one setting too brittle, one background process too sovereign — evolution does not stop the experiment.

At species scale, evolution plays the tiniest violin for individual inconvenience, calls it a rounding error, and moves on to bank the finding.

Nature has no remorse at that scale. A supernova can swallow twenty planets in a second and the cosmos does not pause for grief. Evolution is less dramatic, but not necessarily kinder. It preserves what works often enough, not what feels fair to the individual carrying the failure mode.

What interests me most is the scale of the experiment.

For hundreds of thousands of years, nature has been running variations on the same basic problem: how much should a mind detect, filter, remember, fear, trust, attach, detach, control, predict, and tolerate? There was never one clean design. There were countless trials, countless thresholds, countless nervous systems tuned slightly differently.

Some organisms leaned more toward vigilance. Some toward exploration. Some toward social dependence. Some toward detachment. Some tolerated ambiguity. Some needed closure quickly. Some assigned salience fast. Some filtered aggressively. Some allowed local processes more autonomy. Some imposed more central control.

That variety is not an accident around the human mind. It is part of the human mind.

This is why rigid categories can be misleading. Evolution does not produce psychiatric boxes. It produces distributions. What later becomes a diagnosis may begin as a threshold, a temperament, a sensitivity, a trade-off. Neurotic traits are not the same as neurosis. Schizoid traits are not schizophrenia. Sensitivity is not breakdown. Detachment is not disease. Vigilance is not disorder. These are positions on continua, and individuals differ enormously in where they sit.

Nature does not draw boxes. It tunes thresholds.

One way to make this less abstract is to think in terms of internal governance.

Some minds run more like loose confederations. Different processes have more room to operate: perception, memory, social interpretation, threat detection, imagination, bodily signals, private meaning. This can be useful. It may allow originality, independence, unusual perception, and tolerance for inner plurality. But if the confederation becomes too loose, signals that should have remained local can begin acting sovereign. A background process can acquire too much authority. Salience can escape proportion. The system may start treating noise as message.

Other minds run more like strict central governments. They impose order quickly. They check, regulate, anticipate, rehearse, correct, and police. This can also be useful. It may produce responsibility, preparation, caution, discipline, and social attunement. But if the government becomes too totalitarian, nothing is allowed to remain ambiguous. Every loose end becomes a threat. Every unfinished task demands review. Every social cue becomes evidence. The organism survives by over-administering itself.

A little confederation can be creativity. Too much can become fragmentation — or, on a bad day, a scene from Fight Club. A little central government can be discipline. Too much can become neurotic occupation — Jack Nicholson in As Good as It Gets, trying to survive reality by over-administering it.

Neither mode is automatically pathology. Most people live somewhere between the two, and the position changes with stress, age, environment, sleep, illness, trust, and pressure. Evolution did not produce clean boxes. It produced thresholds. Society then decides how often those thresholds are hit.

The same setting can be useful in one environment and costly in another. A vigilant system in a dangerous world may survive. A vigilant system in a stable world may become careful and prepared. A vigilant system in a pressure cooker may become unable to stop scanning. A socially detached temperament may become independence under tolerable conditions, or withdrawal under chronic alienation. A salience-sensitive mind may become perceptive in a meaningful community, or overwhelmed in chaos.

This is where the sociogenic sauce matters.

Biology may give the organism its parameter settings, but the world decides how often those settings are stressed. Poverty, humiliation, loneliness, family pressure, unstable work, racism, school discipline, institutional distrust, political spectacle, and online exposure are not decorative background conditions. They are part of the operating environment.

The bug is not always in the code alone. From the species’ point of view, the bug may be the individual failure mode of a remarkable milestone: a brain capable of parallel processing, distributed attention, social modeling, flexible thresholds, and semi-autonomous systems. A few loose ends, nothing important.

Sometimes the bug is in the world that keeps calling the same vulnerable function — pressing the same exposed button — until the system fails.

Still Running

 

A biological device built for uptime, not closure.

There is a difference between closure, forgetting, and termination.

Take the stupidest possible example: a pen. I need to write something, and I want my favorite black pen. If I find it, the task closes immediately. Need, search, satisfaction, use. Nothing interesting has to keep running. The process completed its loop.

But if I cannot find the black pen and settle for a red one, something different happens. The functional task closes: I can write. But a small discrepancy remains: where is the black pen?

Most of the time, that discrepancy is too trivial to become memory. It does not cross into the archive of the self. Two years later, if I find the pen under a couch cushion, I almost certainly will not think: there it is, the unresolved black pen of June 2026. It will barely register. The process died before it became memory.

So the mind does not preserve every unfinished thing. That would be unbearable. Most unresolved things probably go into a temporary folder: unfinished but probably unimportant. Desktop clutter, one inch from the recycle bin.

That folder matters because it lets us distinguish between unfinishedness and memory. Unfinishedness alone is not enough to become long-term memory. The organism needs a reason to preserve it. Evolution is not interested in archiving every loose end. It is interested in keeping what may matter later.

The event horizon of long-term memory may be the point at which an experience becomes future-relevant. Not philosophically important. Biologically important.

Threat crosses. Reward crosses. Shame crosses. Belonging crosses. Betrayal crosses. Novelty crosses. A failed prediction crosses. A social wound crosses. Anything that might help the organism anticipate danger, find safety, preserve status, avoid humiliation, repeat pleasure, or understand attachment has a better chance of being saved.

The missing black pen evaporates because it usually does not matter. But if that pen belonged to your dead father, or if losing it caused a public failure, or if someone stole it in a way that confirmed an old distrust, then the same object changes category. It is no longer a pen. It is evidence. It is a cue. It is a future-relevant disturbance.

Borges’s Funes is useful here by contrast. In “Funes the Memorious,” Borges imagines Ireneo Funes, a poor young man from Fray Bentos, Uruguay, left paralyzed after being thrown by a horse. He is incapable of forgetting, and Borges understands the horror of that better than anyone: this is not a superpower, but a curse.

Funes remembers everything: every object, every scene, every variation in light. But that is not necessarily intelligence. It may be the opposite of usable memory. Funes is buried under undigested particulars. No conclusions, no hierarchy, no compression, no mercy. He does not preserve what matters; he preserves everything. He has raw footage where a mind needs models.

Human memory is stranger. We do not remember everything, and that is one of the reasons we can think. But what we do remember may not always be stored as a conclusion. Sometimes long-term memory is not an answer. Sometimes it is a process still running.

Not: this happened, file closed.

But: keep modeling this; it may matter again.

That is why some memories feel less like past events and more like suspended computations. Was I safe? Was I loved? Was I humiliated? Did I misunderstand? Could I have acted differently? Should I expect this pattern again? What does this say about me?

The past is not always remembered as a scene. Sometimes it is remembered as an unfinished task.

This makes human memory different from Funes’s curse. Funes retains objects and scenes: raw perceptions, almost thing-like in their solidity, rather than processes, abstractions, or connections. Humans often have processes that outlive the objects. The details decay, but the unresolved computation remains.

Of course, not every process is allowed to continue. Sometimes the system deprioritizes. Sometimes it lets the folder decay. Sometimes sleep strips a task of urgency without fully resolving it. And sometimes, under dysfunction or emergency, a process may be effectively terminated. Alcohol blackout is one example: the body continues, the conversation continues, behavior continues, but the autobiographical recorder stops saving. The process burns without leaving a file.

The task manager metaphor helps.

A system may ask: Terminate this task? It is still running. Unsaved progress may be lost.

And sometimes the organism answers: terminate. The processor is overheating.

The mind wants closure. The organism wants uptime. Between the two sits an interface designed by evolution, apparently running Windows Vista on an aging Intel i5 with forty-seven Chrome tabs open — unsupported since 2017, but somehow still online.

Ancient threat detection is the McAfee of the nervous system: allegedly there for protection, but intrusive, processor-intensive, full of false positives, and impossible to uninstall.

Threat detection is not evil. It exists for a reason. It kept bodies alive. It noticed danger, remembered predators, tracked betrayal, monitored social exclusion, scanned the dark, and made sure the organism did not walk cheerfully into death.

But what protects can also consume. A process designed to preserve the system can become one of the main reasons the system underperforms. It keeps scanning. It keeps warning. It keeps finding possible danger in tone, silence, memory, ambiguity, desire, fatigue, and rooms where nothing is technically happening.

That is the problem with running processes. They are not always wrong. Sometimes they are just too expensive.

The mind may prefer closure, but the organism prefers uptime. Uptime is not peace, not meaning, not resolution. It is the minimum miracle of continued operation: survival without the luxury of closure.

So maybe a life is not a clean sequence of completed tasks. Nor is it Funes’s impossible archive of everything. It is something more unstable: closed loops, decayed clutter, suspended processes, emergency terminations, and long-term memories that crossed the event horizon because evolution decided the future might need them.

Most things vanish. Some things conclude. Some things remain unresolved but harmless. And a few acquire enough gravity to keep trading with the present.

The past is not always recalled.

Sometimes it is negotiated with.

Sometimes it is still running.

AI, Collective Learning, and the Planetary Layer of Thought

 

AI is best understood not as something outside human knowledge, but as a new formation within the collective knowledge ecosystem itself. In David Christian’s Big History frame, collective learning is humanity’s decisive evolutionary advantage.

Humans differ from other species not merely because individuals are intelligent, but because knowledge accumulates socially across generations. Language, symbolic memory, writing, institutions, libraries, science, and networks allow information to survive individual death and compound over time.

A human being can leave instructions, warnings, equations, myths, legal systems, recipes, and philosophical arguments for people not yet born. Two hundred years later, another human may still retrieve the signal and build on it. Other species can transmit behaviors socially, sometimes with remarkable sophistication, but they cannot accumulate symbolic knowledge outside living memory. No raven expects another raven two hundred years later to recover a carefully labeled container of poisonous berries and continue the experiment.

Human history therefore becomes cumulative in a way biological evolution alone never could. In Christian’s framework, this is one of the great emergent properties of the cosmos: matter organizes into stars, stars into chemistry, chemistry into life, life into symbolic societies capable of preserving and transmitting knowledge.

That idea already contains an implicit theory of emergence. Collective learning is not located in any one brain. No single human contains civilization’s knowledge. It arises from interactions among millions of minds distributed across time. Knowledge is therefore not merely stored information but a system-level property emerging from communication, cooperation, memory, and transmission. Human civilization becomes, in effect, a distributed cognitive process.

Similarly, Scott Page's work on Complexity emphasizes that complex adaptive systems generate higher-order behavior through interactions among diverse agents operating with partial information and local rules. Markets, ecosystems, scientific communities, governments, and cultures are not centrally designed intelligences. They are emergent systems. Their potential arises from distributed interactions, feedback loops, adaptation, diversity, and continual error correction.

Collective learning, viewed this way, is not a library. It is a living ecology of signals, something closer to the dense techno-atmosphere now surrounding the planet: satellites, fiber-optic cables, protocols, archives, recommendation systems, decaying software, and a steadily growing cloud of informational debris, not unlike the occasional flip-flop or stray wrench still orbiting Earth long after the original space mission ended.

That distinction matters because knowledge does not spread merely because it is true. It spreads because systems amplify certain signals, suppress others, reward prestige, stabilize narratives, create incentives, and filter noise. Scientific truths, political myths, social norms, and technological innovations all propagate through complex social selection environments. Information evolves socially much as organisms evolve biologically: through variation, competition, replication, and selection.

This is where AI becomes historically serious.

AI is not merely another tool added to the knowledge system like a faster search engine or a better encyclopedia. It intervenes directly in the interaction layer from which collective intelligence emerges. It changes how signals are selected, summarized, translated, prioritized, generated, and circulated. In other words, AI does not simply add more information to the system. It changes what people notice, repeat, trust, ignore, and build on.

That is a much larger claim.

Earlier cognitive technologies primarily extended storage and transmission. Writing preserved memory. Printing scaled replication. Digital networks accelerated distribution. AI is different because it begins participating in intermediate cognition itself: summarizing, inferring, recommending, generating explanations, synthesizing domains, and shaping interpretation in real time. The medium no longer simply carries thought. It increasingly participates in shaping thought.

This is why the relationship between humans and AI already feels recursive. Every conversation with an AI system is, to some extent, a training exercise. The models learn from accumulated human expression, but humans also start adapting to the models in return: adjusting vocabulary, compressing thoughts differently, reorganizing attention, even reshaping their sense of what a ‘good explanation’ sounds like. The feedback loop is already underway. Some mornings, before coffee, I suspect I’m running on an outdated operating system and critically low on API tokens.

Complexity theory would recognize this immediately as feedback coupling inside a complex adaptive system. Humans shape models; models reshape humans; both evolve inside the same informational environment. The knowledge ecosystem becomes recursive.

This recursive dimension makes the planetary metaphor unexpectedly useful — provided it is handled carefully and not allowed to drift into Gaia gift-shop mysticism. David Christian’s tone is important here. He speaks less like a prophet than like a geologist. His sensibility is layered, historical, material, and evolutionary. New forms of complexity emerge under the right conditions over deep time. No incense necessary.

Viewed through that lens, human collective learning itself can be understood as a late-forming layer at the Earth’s surface. Geological history produces physical structures, chemical cycles, atmospheres, oceans, and biological ecosystems. Human societies then generate another kind of layer: symbolic, informational, institutional.

A thin but extraordinarily potent surface stratum composed of language, archives, networks, science, law, media, and memory. AI now emerges within that layer as one of its newest formations.

The metaphor of a “terrestrial cortex” works structurally if not literally. Not planetary consciousness. That road gets mystical fast and usually ends in a gift shop. But the Earth has undeniably developed increasingly complex systems for storing, circulating, and processing information through one of its evolved species. AI belongs inside that process.

Yet the more revealing metaphor may be fog.

AI thickens the informational atmosphere surrounding society. Signals multiply, but so does haze. Information becomes more available while certainty becomes strangely harder to pin down. Origins blur. Authority blurs. Sometimes even reality itself starts feeling like it has passed through three summaries, two recommendation engines, and a motivational LinkedIn post before reaching your screen.

In a fog, navigation changes. People rely less on direct perception and more on instruments, signals, maps, and collective guidance systems. Something similar now happens in the informational environment. As AI-generated interpretations and machine-mediated explanations proliferate, humans increasingly experience reality through layers of processed abstraction. The medium no longer simply carries information. It quietly begins shaping how the information is perceived in the first place.

AI therefore becomes both archive and atmosphere, both memory and fog.

This tension cuts to the center of the problem. AI may deepen collective intelligence by improving translation across domains, lowering barriers to participation, accelerating synthesis, and helping societies process complexity beyond ordinary human bandwidth. In that optimistic version, the emergent intelligence of civilization becomes richer, more adaptive, and more distributed.

But the opposite outcome is equally plausible.

Complex systems depend on diversity for resilience. Scott Page repeatedly emphasizes that systems composed of varied models often outperform systems built around uniform expertise.

If millions of people begin relying on the same handful of model architectures, optimized toward similar patterns of fluency and consensus, then cognitive diversity may shrink even while efficiency rises. The result could be a civilization that becomes more coordinated but less exploratory, more synchronized but more fragile.

The system grows increasingly efficient at distributing ready-made interpretations, reducing the need for individuals to forage cognitively for themselves. Ideas arrive pre-processed, like carefully pre-chewed worms delivered directly into open beaks.

A system can become more organized without becoming wiser.

That may be the deepest risk. AI could intensify collective learning while simultaneously degrading the epistemic conditions that make collective learning self-correcting. Synthetic output may overwhelm verification. Fluency may outrun truth. Feedback loops may become polluted by recursively generated noise. Systems optimized for engagement may privilege coherence over accuracy. Under those conditions, the collective intelligence of the system becomes increasingly difficult to distinguish from large-scale imitation. In sufficiently interconnected societies, a sufficiently sophisticated cognitive malware may spread through institutions faster than many biological pathogens, while leaving damage that is far harder to quarantine.

From both Big History and complexity theory, then, the real issue is not whether AI accelerates collective learning. Of course it does. The real issue is whether AI reorganizes the emergence dynamics of collective learning itself.

That is a civilizational question.

Human societies may be entering a phase in which cognition is no longer located primarily within individual minds or even traditional institutions, but distributed across recursive socio-technical systems that continuously shape and reshape one another. Not machine consciousness. Not technological destiny. Something more geological and systemic: new layers of organized complexity forming within the informational surface of planetary civilization.

Whether that produces deeper collective intelligence or a denser and more beautiful fog remains unresolved.