AI: Endgame: AI Summary

This is an AI-generated summary of AI: Endgame. The full paper is at https://www.lloydwtaylor.com/endgame/ai-endgame/

Nearly every attempt to govern artificial intelligence treats it as a policy problem. This essay, the first in a four-part series, argues that biology offers a better lens. Seen that way, AI is a problem of organism. The question is how a living system keeps itself working when something new disturbs it. It is harder still when the disturbance gets inside the very equipment the system uses to notice trouble.

How a Body Stays Stable

At any moment, the human body runs hundreds of feedback loops at once: blood sugar, temperature, blood pressure, pH, sodium, oxygen, cortisol. Each has its own set point, the target value its loop works to hold, and its own way of detecting and correcting change. None reports to a central coordinator. The body does not have a regulatory system, the essay says. It is an architecture of overlapping corrections running at different speeds, with no one in charge.

Three landmark ideas anchor this picture. In 1865, Claude Bernard saw that a body’s stability is not the absence of disturbance. It is the ongoing result of correcting against constant perturbation, meaning any disturbance big enough to force a response. Every heartbeat and every meal is one. In 1932, Walter Cannon named this homeostasis, which means actively maintaining balance while being pushed off it, not simply being in balance. In 1993, Bruce McEwen and Eliot Stellar showed that under long-term pressure, set points drift. A body under chronic stress settles around a new baseline instead of returning to the old one. McEwen called the price of this resetting allostatic load: the slow wear of surviving in conditions a body wasn’t built for. The organism survives. It is not, in any useful sense, well.

The essay uses “organism” in a structural sense. It means any system that keeps itself going by detecting and correcting problems in many places at once. The comparison applies only where that kind of operation can actually be seen in society.

How Human Groups Regulate Themselves

David Sloan Wilson works in the tradition of multilevel selection theory. Drawing on him, the essay argues that natural and constructed regulation differ in kind. In a band of fifty people, regulation is part of the group itself. The elder enforcing a rule is a relative. Shame feels like losing a relationship, not paying a fine, and damage to reputation registers in the brain much like physical pain. Large groups must build a copy from different materials: rules, enforcement bodies, penalties. The output looks like the original. The mechanism is completely different.

Natural regulation rests on three conditions. People can see what others do (direct observability). Others’ judgments matter to them (reputational stakes). And relationships last long enough for those judgments to count (relational continuity). Robin Dunbar put a ceiling on this at about 150 people, the most a human brain can track well enough to know who cheated and who cooperated. Above that, substitutes take over. Religion supplies an all-seeing watcher. Law writes down norms and hands enforcement to specialists. Both are imitations. Neither is the thing.

The essay’s sharpest example is an influential prediction that proved wrong. Garrett Hardin’s 1968 “The Tragedy of the Commons” argued that any shared resource without private ownership or state control would be destroyed by overuse. Elinor Ostrom went looking for real cases. She found Swiss alpine meadows shared for centuries and Japanese irrigation systems running for four hundred years. She found Spanish acequia irrigation networks older than the nations governing them, and Maine lobster fisheries run through informal territories. She summed up what they shared in eight design principles, including clear boundaries, rules fitted to local conditions, monitoring, and graduated sanctions. Read plainly, the essay argues, those principles describe small-group regulation at work. Hardin’s tool, formal economics, had no slot for that machinery, so he couldn’t see it and concluded it didn’t exist. The tragedy is real, but it describes what happens when community has been destroyed and nothing has replaced it.

Timing is the third lesson. The body’s corrections are staggered by speed. Nerves act in milliseconds, and evolution acts over generations. Each layer is matched to the disturbances it faces. Society’s layers also run at different speeds: markets in seconds, news in days, elections in years, constitutions in decades. But nobody designed that stagger. It piled up by historical accident, so when a problem moves at the wrong speed, the mismatch is built in.

Why AI Is a Different Kind of Disturbance

Fire, the printing press, industrial machines, and the internet changed the conditions in which human thinking worked. AI, the essay argues, does part of the thinking itself. It weighs evidence, writes language, spots patterns, and makes regulatory decisions. Those functions are society’s self-correcting machinery at its highest level. AI is a new kind of process working in the same territory as the instrument trying to judge it.

The essay names three failures that are stages of one mechanism. Misclassification aims correction at the wrong target. Response-time mismatch keeps correction from settling before the target changes. Integration puts the same kind of system inside the equipment doing the classifying. Bit by bit, that erases the boundary that outside regulation depends on.

Classification matters because the response depends on it; the body reacts differently to a bacterium than to a virus or a cancer cell. Society is filing AI under every category it has: tool, communications medium, professional service, intellectual property, weapon, worker. Each label triggers its own rules, and all misfire as complete answers, because AI fits every one in some setting and none in general.

Biology’s version of this failure is prion disease. A misfolded protein causes healthy copies of the same protein to misfold. The immune system depends on telling “self” from “not self,” and here the invader is self, so the defense runs against the wrong target. Meanwhile, AI is taking on regulatory jobs directly. It moderates content, judges credit, screens job seekers, assists in sentencing, diagnoses illness, and detects threats. The same kind of system is beginning to sit on both sides of the line: the thing being judged and the tool doing the judging.

Speed makes it worse. Biology has no close precedent for this. AI moves fast, produces new versions of itself, and changes the judging instruments, all within one generation. Most of society’s adaptive tools are at least ten times too slow. What survives fast change in nature is phenotypic plasticity, the ability to reorganize form and function within a single lifetime. The uncomfortable implication: the institutions most likely to cope are not the most consistent or principled, but the ones able to reorganize fast.

Five Possible Endings

Biology offers five outcomes for an organism facing a truly new disturbance. The first is death. The second is stable adaptation, a new healthy balance. The third is chronic pathology, survival under permanent strain. The fourth is speciation, where groups adapt differently until they differ in kind. The fifth is integration, where the disturbance is absorbed into the organism’s own machinery.

Death is a tail risk, not the likely course; AI isn’t hostile by design, and responses are firing everywhere. Stable adaptation is what everyone is attempting. But it can’t settle while AI moves faster than institutions do. Every framework targets something that will have moved. The alarm is the part of a system that notices trouble and fires a correction. The result is a continuous alarm phase: the alarm keeps sounding, and the correction never settles.

Chronic pathology is the most accurate description of where most institutions are and will stay: rules always behind, governance always mobilizing, shared trust in information eroding, courts producing contradictory rulings. The essay calls it “the metabolic syndrome of governance”: functional, but not healthy.

Speciation has begun. Europe’s risk-tier rules, China’s state-integration model, and America’s market-led approach are different environments breeding different organisms. Where security, market, and political incentives outweigh the pull to coordinate, they will drift further apart.

Integration is where the logic points. Where human-plus-AI performance is rewarded and staying out is costly, integration is the structural attractor, the state such systems tend to settle into. The essay calls the resulting working unit the compound: human judgment and AI output operating together. It stresses that this is a prediction, not a recommendation.

The model comes from Lynn Margulis. Her once-rejected idea, now foundational, is that the mitochondrion began as a free-living bacterium. The mitochondrion is the energy-producing part of every complex cell. A host cell took it in, almost certainly through predation or infection, and the pair outcompeted both partners in an oxygen-rich world. Nobody chose this; it was selected for. The mitochondrion still carries its own DNA, but neither partner can now live without the other.

The same pattern is starting in radiology. A radiologist with AI support can do what neither could do alone, and training is being rebuilt around that compound. Similar shifts are beginning in legal analysis, drug discovery, financial modeling, and research. The integrated entity will be more capable in specific areas. It will depend for good on infrastructure it didn’t evolve: computing, energy, and supply chains. And it will be vulnerable in ways not yet visible. A simple cell could never have predicted mitochondrial disease; that failure could only be named once the merged cell existed.

What Survives a Fast Shift

When change outruns slow adaptation, the winners are not the organisms most finely tuned to old conditions. They are the plastic ones, able to build a new working form from capacities they already carry. Take the spadefoot toad tadpole. When its pond dries too fast, it switches within days from an omnivore body to a carnivore body. It grows a wider mouth, a shorter gut, and different jaw muscles. It survives by eating the pond’s creatures that can’t adapt. Such organisms look wasteful in calm times, keeping backup capacities that rarely fire. The payoff comes only when conditions change fast.

Flexible organizations share four traits. Authority sits where the organization meets the world. Leaders specify outcomes, not procedures. Redundancy is kept, not trimmed. And some inconsistency is tolerated, since consistency assumes the past will keep working. Special operations units, emergency medicine teams, and improvising jazz groups fit this pattern. A symphony orchestra makes better music for the concert it was built for, but the jazz group can reorganize under conditions no one could predict.

The essay calls this structural, not prescriptive. The institutions most exposed to AI tend to be the most optimized for stability, with refined procedures, long decision cycles, and top-heavy authority. The shock finds them “exquisitely prepared for a world that is no longer the one they are operating in.”

What Cannot Be Seen From Here

The biological comparison breaks at one point, which the essay calls its hardest turn. The mitochondrion was truly separate before it merged. AI is not. It is an artifact, made again and again by the society absorbing it. And it is made by minds the absorption is already reshaping. Society generates the disturbance, the disturbance reshapes society, and reshaped society builds the next version. This is coevolution compressed into a single generation, with feedback loops measured in months. Nothing in the fossil record shows where it ends.

That leads to the observer constraint: a system cannot evaluate its own output from outside its own frame. It applies even to the essay’s own prediction, because the instruments for judging the integrated state are among those being integrated. The endpoint is visible only from inside that state. We cannot see it from here, not because the information is hidden, but because the observer has not yet arrived.

The essay’s closing:

The body does not merely regulate the perturbation from outside. It adapts by drawing it into the machinery of regulation. What the compound becomes is the one thing it cannot currently see from inside the alarm.