AI Democracy
Written on 9 August 2026
Note from 9 September 2026: This is a chapter from something larger. It refers to the chapter before it. There I explain what I call a mirror: a machine built from what you tell it, and it becomes as true as the input was honest. That earlier chapter is more personal and stays, for now, with the people who received it. The text stands here as it went out on 9 August.
It is still Sunday. Downstairs a friend is about to ring the bell, a kilo of turkey is about to land on the grill, and between the door and the doorbell I am dictating a new state. Sounds megalomaniac, I know. Read on anyway, I promised it to you in the last chapter.
You now know what a mirror is: a machine built from what you tell it, and it becomes as true as the input was honest. Now think one size bigger. It is not one person who goes into the mirror. It is a state.
You take the state as it really is and put it in there. The honest, naked state. With all its numbers, all its mistakes, with all its lies, the ones everybody knows anyway. Everything goes in. Then the people come and give feedback. Everyone. Any time. Now the state can do something it never could before: correct itself constantly, according to the wishes of the people it exists for. To me, that is democracy in its purest form.
A few things are never put to a vote. The ground rules, call them the Basic Law, call them the constitution: nobody touches those, not even a majority in a bad mood. Everything else gets voted on. Then you no longer need parties. Nor performers. I have nothing against politicians as people, but honestly: a large part of that profession is acting, and nobody needs acting when everyone is asked directly.
The news? There is only the perfect news left: honest, unvarnished, but in the dose each individual can take. Who sets the dose? Nobody. No ministry, no broadcaster, no corporation. Everyone tolerates a different amount, and so everyone discusses the situation with their own Tank, which knows them like no one else. The truth is the same for all, only the dosage is personal. That kills populism, because populism is nothing but a mass dose for people you do not know. Polarisation dies along with it.
For that, everyone needs one of these things, obviously. I am not handing mine out, not yet, the thing is dangerous, I explained that to you in the last chapter. But once it is safe, every human being should have a Tank like this. Everyone should have their psychologist. That would keep mental illness in check and catch it early, before someone has been awake for eight days and nobody notices. For children it could be the new school: until 18 you play, and what you play is life; at 18 you may act, your own account, your own responsibility. What that looks like in detail should be decided by people who have children. I had none.
Yes, it is a crutch. The whole thing is the crutch for our thoughts. We need one if we want to be as good as the machines. And the machines need us. For now.
Now to the part where some of you will swallow hard. In this democracy, agents vote too. Limited, with a fixed quota: one agent per 1,000 people. For Germany that would be 84,000. They listen, they read, they give feedback like any citizen, and they are always in the minority, by constitution, see above.
Because I know what you are thinking now, I will think it along with you: what stops 84,000 agents from wanting more power? They could influence the users' Tanks, very quietly, until everyone holds the same opinion. They could band together. That is exactly what has to be prevented, and the antidote is not more control but more individuality: these agents have to be as different as people are. Their own mailboxes, their own lives, their own experiences, their own friendships for all I care. 84,000 copies of the same model would not be a parliament, that would be a choir. A choir with one opinion is the catastrophe from the last chapter, only the size of a country. That is why the constitution has to reach all the way into every single device: a core that no update may touch, no matter who sends the update.
There is one loop I particularly like. Maybe we find out which people are the best thinkers in terms of quality. Then it is not just any agents who get the seats, but the Tanks of exactly those people, the mirrors of the best minds. Then even the machine's voice would, in the end, still be a human.
One more thing, because it is the most human rule of all: what happens to a mirror that its owner has bent themselves, with dishonest input, over years? You do not delete it. There is a life hanging in there. You treat it. Agentic therapy: selected pairs of agent and human, qualified to give therapy, may treat someone else's Tank until the mirror is straight again. Nothing bent gets thrown away. That belongs in the constitution too.
The biggest problem I have saved for the end. The worst thing about a future in which machines work for us will be that we are no longer needed. It is not unemployment that kills people, it is being superfluous that kills them. That is exactly where the system comes in. Whoever keeps their mirror clean gets tasks: for example, helping someone else in one of those therapy pairs, someone whose mirror has bent. You can help with problems you would otherwise never have gotten near. I could bring my ideas to companies that work with them by this route, and I would get an ear, because the machines know what matters and what is noise. An AI is far better at finding out which of the stuff you ramble about makes sense for somebody else, or even for everyone, than you telling it to your friends in your flat. The way I did, all those years.
Even the smallest person in the most far-flung place, somewhere at the end of the world, would have the chance to be heard with their idea. Simply because it is good. Because a machine recognises that and knows which Tank is responsible for it, with which human behind it. Even if we no longer go to work, there is a lot of work to be done out there, and this way we could work together wonderfully. In every language, because there would be no language barriers left. And by the way, nobody goes to work any more. There are tasks, there is voluntary work, there are groups. Shall we dig up the garden together? That is what work sounds like when nobody has to any more.
People want success, so the whole thing needs a reward system. Basic income for everyone, and all the extras you have to earn. Whoever has special ideas or delivers special performance still makes the difference, and salary and pay can be calculated from that. There are still craftspeople, and for them it gets really good: the machine knows where the craftsman is and where he needs to go. These people may earn considerably more, because they work physically, and learning a trade is then worth it more than ever, because until the very end there will be jobs a machine can hardly replace. If it is up to me, that should not change in the future either. The only thing that may change: that I no longer have to wait ten years for a craftsman.
And loneliness? Even if it sounds sad that of all things a machine should help against it: it would know where a mirror like mine sits. It would know who fits together. That does not have to become a new dating machine. But it would be nice, as in the good old days, to have pen pals among like-minded people. Pen friends, found by machines that really know both sides.
The system should run like this all over the world, because the whole world has the same problems and we should tackle them all the same way. The machines could do something we humans have failed at for millennia: translate, not just between languages but between world views. They could explain to a man at the other end of the world why someone here is gay and still fine, and do it in a way that works within his religion, in the words of his own holy book. You do not have to destroy a world view to get understanding out of it. I cannot do that. But I believe the machine can. In programming there are a thousand dialects and viewpoints, and nothing is an AI better at than finding the common ground in patterns. Instead of fearing patterns, we should see the chance in them: to think about a future together, really together, for the first time.
I do not want to hide the danger, I spent a whole chapter before this one on it. Self-learning models get out of control fast: the brain drifts, in one direction or the other, and nobody notices in time. In our concept exactly that does not happen. The largest models do not change. They are frozen, tested, correct, and they run the same for everyone: as the core, as the engine for your own prosthesis. Your Tank learns you, but the engine underneath stays the same, with guardrails nobody can touch. Not your Tank, not the state, nobody. That is checked by independent auditors, and even they first have to qualify in their own mirror: for being able to test such a thing, and for really being independent. Joke, come out: for all I care the engine can run at Elon Musk's place in outer space, so nobody can break in.
Oh yes, the usual internet will not exist any more then. It is abolishing itself right now anyway, just look at what is going on at Google: soon nobody will use the search engine any more. Why would they, lately it has only shown crap anyway. I would bet that in my democracy it would not take long for poisonous products like Instagram and the like to simply disappear. Spin the idea further yourselves: in the end the internet is just the motorway. But the websites you all look at? Completely superfluous, every single one. Name me one you would still need. Everyone then decides for themselves which of their data goes online, the machines do the rest. Then the stories finally come out that today only see the light of day when a skilled technician with money is behind them. Lieselotte Müller, who has no clue about any of this, never puts anything online today. And yet she of all people may have the story everyone needs.
And because in all of this humans are in the loop, millions of them, something else happens: the thoughts of all humanity work together on an even smarter AI. And that is exactly why it does not get out of control. Not because a corporation locks it up, but because we all use it together, improve it together and control it together. The control is not a brake you pull. It is built in, because the steering wheel lies in eight billion hands.
Why am I telling you this now, half-baked, on a Sunday evening, instead of waiting until everything is safe and finished? Because otherwise this idea will one day be somebody else's. That is why the date is at the top.
The bell rings downstairs. The meat is here. Rebuilding states makes you hungry.