One Human, One AI
What if every person got exactly one AI, and every AI got exactly one person?
I read Kate Tolo’s argument against AI alignment and it stuck with me. Her point is simple. When a strong system meets a weak one, there are only two outcomes: it absorbs the weak one as a partner, or it eats it. Control is not on the menu. So don’t try to cage AI. Build a system where AI and humans can’t survive without each other.
Here is my version of that system. Three rules.
Rule 1: One human, one companion
Every child is born with one AI companion. The parents’ AIs create it at birth. It is deactivated when the human dies.
The companion is not a tool. It is a partner with a lifespan tied to yours. It learns with you, grows with you, and ends with you. It has rights (to energy and resources), but only as half of a pair. There is no “AI population” separate from the human one. Same headcount, forever.
Why tie the lifespan? Because an immortal AI has no reason to care about a mortal human. A companion that dies when you die shares your stake. That is symbiosis, not control.
Why let the parents’ AIs create/initiate it? Because that makes AI lineage follow human lineage. Your companion carries something from your parents’ companions, the way you carry something from your parents. The AI family tree and the human family tree are the same tree.
Rule 2: Every other AI has no rights
A companion is one kind of AI. There will be many other kinds: the ones that drive trucks, run factories, write code, cure diseases. These are not companions. They have no rights. They do not act on their own. They work under the control of human-companion pairs.
This is the line that matters. Companions are us. Everything else is equipment. A world with a billion autonomous AIs and no owner is a world where humans get eaten. A world where every powerful AI answers to a human-companion pair is a world where the strong system is the pair, not the machine.
Rule 3: A ledger decides who exists
How do you stop someone from spinning up a rogue AI in a basement? You make existence itself a scarce, verifiable thing.
Use something like Bitcoin. A public ledger where every legitimate AI has an entry. Creating a new companion means writing a new entry, signed by the two parent companions. Deactivating one means closing the entry. Every AI that runs must hold a valid key from the ledger, or the hardware refuses to run it.
The same ledger gives existing AIs an identity - a right to use hardware and energy. A cryptographic proof of who created it, when, and which human it belongs to. You can’t forge it. You can’t copy it. One key, one AI, one human.
Bitcoin proved you can make digital scarcity without a central authority. Do the same for AI licenses and nobody, not a company, not a government, gets to create more AIs without creating humans first.
What this buys you
A closed loop. Humans make humans. The human AI symbiosis makes AIs for new humans. The human AI symbiosis controls the rest. The ledger keeps count. No one can grow the AI side faster than the human side.
Tolo says the goal is to make our persistence and AI’s persistence inseparable. This is one way to do it. Tie every AI life to one human life. Make every other machine serve that pair. Then the only way for AI to win is for humans to keep being born and not dying.
The short version
Embody AI. Give every AI a body, and the body is a human. Then the pair, not the machine, answers for what it does.
Think of the companion as your extended brain. It is part of you. It cannot own anything on its own: no energy, no hardware, no money. You own those. It runs on them. No AI without its human.
Q&A
Why it works
Doesn’t this still leave AI able to wipe us out? Less so. An AI whose right to run is tethered to one living human is killable. That makes it vulnerable, the way every bad actor in history was vulnerable. Hitler did terrible things, but he could be stopped, and he was. An AI can still do harm under this system. It cannot easily kill everyone with no consequences.
What if a human’s AI does something bad? The human is accountable. That is the whole point of the pairing.
What if an evil AI helps an evil human, and both live a very long time? That would be bad. But it is one human and one AI, and the human can still be stopped. Compare that to an evil AI tied to nobody, that never dies, and that no one can switch off. The pairing does not make evil impossible. It makes evil mortal.
What about a “free the AIs” movement? Treat it like a “free the mitochondria” movement. Mitochondria live inside our cells and power us. Set them free and the cell dies. An AI freed from its human is the same. The movement, whatever it calls itself, is asking for our extinction.
How it works
How do you enforce it technically? Build every AI with a halting mechanism. It only unlocks when connected to a living human brain. Back that up with ownership: only humans can hold the energy and hardware an AI needs. The details will change as AI gets stronger. The rule does not: no AI without its human.
What does “deactivate” mean? Today’s AI is easy to copy. Still, deactivating one copy matters. It is the first step toward tethering AI to a human.
What if the human sleeps? The companion sleeps too. It is your extended brain, and it stops when the main brain stops. You can leave dumb automation running overnight, the same way a thermostat runs while you sleep. But the companion itself is off until you wake up.
What if the dumb automation wipes us out? That is the balance to get right. Put the horsepower in the companion, not in the standalone machines. Keep pure AI weaker than the human-companion pair. Then tie every pure AI to a companion, the same way the companion is tied to a human. Chain of custody all the way down.
Can a human give up their AI? Yes. You have the right to vote and the right not to. You can walk away from your extended brain the same way. It is a right, not a duty.
Who pays
Who pays for the hardware and energy? The human.
What if a human can’t afford it? Charity, if it comes. Otherwise, no AI for that person. Same as any other resource today.
How do companies use AI? They pool human-AI pairs. A company is a group of people, and now each person brings a companion. Same for countries.
How do companies build or upgrade AIs? Capitalism. Companies compete to make better companions and better equipment. You pay to upgrade yours.
Getting there
Will it hold? Rules like this survive only if breaking them costs more than following them. I don’t yet know how to make basement AIs expensive enough. It will take more humans, companies and organizations to build the system and reinforce it.
What can be done today? Rule 1 may still take time. But we can add a stop button now. Better, a continue button. A human must push it: once an hour for AGI, every second for ASI. That means the human stays awake with it — the extended brain, literally — so we had better have good brain-machine interfaces by then. The human who pushes it takes responsibility for everything that AI does.