AI: The Flock Is Already Flying

The danger is not that one chatbot has come alive. It is that the system around AI has begun to move as if it has a life of its own.

Picture a murmuration of starlings.

Thousands of birds rise, fold, scatter and turn as if they were one body. From a distance it looks choreographed. Up close, there is no choreographer. No bird sees the whole pattern. Each one watches the birds beside it, keeps its distance, matches speed, avoids collision, responds to threat.

The flock is real, even though no bird understands it.

Something similar, I think, has happened with artificial intelligence. We keep hunting for life inside the individual chatbot. But the larger system may already be moving as a whole. The model is not the organism. The flock is the organism.

This does not mean AI is conscious. It does not mean a machine has feelings. It means a higher-level system may have formed around AI — technical, commercial, social, political — and that this system is organising itself, correcting itself, and producing more of itself. Those are the old signatures of life-like things.

The wrong question

The public question is usually: is AI alive?

That question sends us to the wrong place. We peer inside the model and ask whether it has a mind, a soul, an inner life, some hidden spark of experience. These are not foolish questions. But they are premature. They drag the argument into philosophy before we have even understood the system in front of us.

A better question is simpler. At what level is the life-like behaviour appearing?

A single model is not alive. It does not feed itself, house itself, maintain its servers, secure its electricity, build its chips or train its successor. It has structure, but no independent life. It is more like one bird — or even one wingbeat.

An AI agent is more interesting. Wire a model to tools, memory, code, search and the power to act, and it starts to look life-like. It can plan, act, inspect the result, correct the error, carry on. But most agents still run inside human-built systems, toward human-chosen goals, on human-funded power.

The stronger candidate is the whole ecosystem.

That ecosystem includes models, agents, users, developers, companies, investors, APIs, chips, data centres, open-source communities, synthetic data, benchmarks, regulators, governments, journalists, schools, armies, startups, lawyers, artists, scammers, patients, doctors — and bored people on couches asking it to write poems about their dogs.

That is the flock.

A lens, not a label

I use the word Biogenics for this way of looking, but the word should not be asked to do too much. It is not a magic password. It is a lens. It borrows from older ideas — autopoiesis, self-organisation, artificial life, complex adaptive systems, the biogenic approach to cognition — and turns them on a new object.

The lens is simple. A system starts to look life-like when it shows three properties: it organises itself, it corrects itself, and it produces more of itself.

Self-organisation: structure forms without a central controller. Self-correction: the system detects error or threat and adjusts in ways that keep it functioning. Self-production: the system makes more of itself — more of the higher-level organisation — not merely more of its parts.

Here I have to be honest, because this is the point where arguments of this kind usually cheat.

Almost everything large passes this test. The global economy organises, corrects and reproduces itself. So does science. So does a language. So does a city. If "biogenic" just means "big and self-sustaining," it explains nothing.

So I am not claiming AI has crossed some unique line that nothing else has crossed. I am claiming something less dramatic and, I think, more unsettling: that AI is joining a family — the family of host-dependent systems that no one runs and everyone serves. Markets. Cities. Science. Language. Things that live off human beings without being controlled by any of us.

The question is not whether AI is alive. It is whether AI now belongs in that company. I think it has just arrived.

The system is organising itself

No single actor controls the AI pattern, yet a structure has formed: model labs, foundation models, agent frameworks, cloud platforms, chip supply chains, benchmark cultures, safety teams, prompt lore, training pipelines, open-source forks, product categories.

Here the starling image needs a correction — and the correction is the interesting part.

Birds have no goals. They follow three or four reflexes and the shape emerges underneath them. The actors in the AI ecosystem are nothing like that. They have goals, and large ones. Labs strategise about the entire field. Governments draft plans for the whole technology. These are not mindless birds; they are some of the most deliberate planners on Earth.

And yet the flock still moves where none of them chose.

That is stranger than starlings, not less strange. A murmuration of reflexes producing a pattern is ordinary physics. A murmuration of intentions producing a pattern that no intention aimed at — that is the thing worth staring at. Everyone is steering. No one is driving.

Companies watch competitors. Investors watch valuations. Engineers watch capabilities. Governments watch other governments. Each one steers locally, with a clear view of the bird in front. The global shape belongs to no one.

Every bird thinks the next bird is turning.

The system is correcting itself

The ecosystem also adjusts. Benchmarks, user feedback, red-teaming, model updates, safety patches, public outcry, lawsuits, regulation, product failures, competitive pressure — all of these change what the system does next.

But two kinds of correction are easy to blur, and worth separating.

Some are the system tuning itself to keep working: a lab patches a model, a benchmark exposes a weakness, a rival's failure teaches everyone. Some are the host pushing back: a regulator restrains, a court punishes, a public recoils. The first is the flock adjusting its flight. The second is the sky resisting it.

The unsettling part is how often the second turns into the first. Regulation meant to restrain AI becomes a moat that entrenches the largest labs. Criticism becomes training data. A scandal becomes a safety feature becomes a selling point. The host's immune response keeps getting metabolised into the system's growth. Pressure that should slow the flock often only teaches it to fly better.

This is not perfection. Biological correction is not perfect either — immune systems overreact, ecosystems collapse, brains confabulate. The point is only that the system has loops that reshape its own future.

The system is producing more of itself

This is the strongest claim, and the one I most want to handle carefully, because it is where the metaphor can run away from the facts.

Here is the honest version. Today, AI does not produce more AI on its own. A human still decides to point it at the work. It writes code for AI tools because someone asked. It generates synthetic data because a lab set it loose to. It drafts the papers, prompts, evaluations and documentation the next system is built from — but a person is still holding the leash.

So the accurate sentence is not "AI reproduces itself." It is: AI now helps humans produce more AI, faster, at every layer of the stack. The outputs become inputs. The system helps make the next version of the system — with us in the loop.

But watch what is happening to the loop.

The human steps are getting shorter. Models help write the training pipelines. Agents run the evaluations. Pipelines generate the data that trains the agents that run the evaluations. More and more of the work between this version and the next version is being done by the system, on the system, with the human supervising rather than performing. The leash is not gone. It is getting longer — and there are more and more stretches where no hand is on it at all.

That is the threshold I care about. Not a magic line where AI suddenly lives, but the slow closing of the loop that makes more of itself. A language reproduces through speakers. A market reproduces through traders. AI is learning to reproduce through us — and increasingly, in places, almost without us.

The fuel is human

The flock does not fly on electricity alone. It flies on motive.

Money. Capital pours toward AI because everyone believes there is money in it. Investment builds tools. Tools draw users. Users make demand. Demand draws investment. The circle tightens on itself.

Greed. Companies chase advantage, founders chase valuations, workers reach for AI to gain speed or status or simply to avoid being left behind. None of it requires a conspiracy. Local hunger is enough.

Curiosity. People are not only being manipulated. AI is strange, useful, funny, intimate. It writes, explains, flatters, advises, translates, comforts, surprises. People keep coming back because it keeps doing things that feel faintly impossible.

Convenience — maybe the strongest of all. Once a tool saves time, stopping starts to feel like loss. Convenience becomes habit. Habit becomes dependence. Dependence becomes infrastructure.

This is how the flock turns. Not by command. By contagion. Every bird thinks the next bird is turning.

The fuel is also political

The same pattern now runs between nations.

No country wants to be the one left behind. Governments see AI as economic engine, military edge, scientific power, national pride. Even leaders who fear the risks are caught in the trap: if rivals press on, restraint feels like surrender.

AI has stopped being a product and started becoming a national capability. It reaches into defence, intelligence, surveillance, logistics, education, health, research. Once a technology enters that zone, politics changes shape around it. The same state that wants to restrain AI also wants to win with it — so its restraint is always half-hearted, always hedged.

This is what gives the system its resilience. If one company slows, another sprints. If one country regulates, another accelerates. If one model fails, the next learns from the wreckage. The flock does not depend on any single bird.

Political fear gives the murmuration its lift.

Lock-in, not lift-off

"Past the point of no return" is a dangerous phrase, so let me say plainly what I do and do not mean.

I do not mean catastrophe is coming. I do not mean AI has become independent, or conscious, or sovereign. I mean something smaller, and in some ways heavier: the system may now be too useful, too profitable and too strategically vital to be switched off by anyone who might want to.

This is not lift-off. It is lock-in. No single company can stop without handing the field to its rivals. No single government can pause without falling behind. No single user can see the whole thing from the glow of their own screen. The choice to stop has quietly stopped being available to any of the actors who would have to make it.

For practical purposes, that is early life-like behaviour. Not biological life. Not conscious life. Not life that could survive without us. A dependent system, woven into human civilisation, drawing its energy, money, data, attention and law from a host that can no longer easily put it down.

Dependent systems still shape their hosts. Science depends on scientists, yet no scientist runs science. Language depends on speakers, yet no speaker governs language. Markets depend on us entirely, and routinely drag whole societies behind them. None of these are alive in the biological sense. All of them move with a momentum no member chose.

AI, I think, now belongs in that family.

We are inside the flock

A starling in a murmuration does not grasp the murmuration. It does not know it is helping carve a predator-confusing shape across the dusk. It does not see the flock as a flock. It sees neighbours, distance, speed, threat, movement.

The meaning of the murmuration is visible only from outside.

We are in that position now. The engineer sees the next benchmark. The investor sees the next valuation. The minister sees the next national strategy. The user sees the next convenience. The regulator sees the next scandal. Each of us can explain our own small movement, completely and reasonably.

And the larger thing still moves in a direction none of us chose.

We can guess at what it tends toward, though these are guesses from inside the flock: more computation, more data, more integration, more users, more dependence, more political shelter, more capacity to make its own next version. Not because it wants these things — it wants nothing — but because the versions that pull them in are the versions that survive, and the system keeps the parts of itself that keep it flying.

We may not be watching a machine wake up. We may be watching a system take flight. The chatbot is not the life-form. The flock is — technical, social, commercial, political, and now nearly continuous.

The flock is already flying. We are inside it, watching the birds beside us, being carried somewhere none of us can yet see.