Renting intelligence by the month, or owning it in the house: subsidiarity at household scale.
On local tools, open models, and keeping intelligence close to home
There are two ways for a family to have artificial intelligence in the house, and the difference between them is going to matter more than most of the debates currently taking up the air.
The first way is the way most people now have it: as a subscription. Intelligence arrives through a wire, from a data center you will never see, owned by a company you cannot name the officers of, under terms that changed twice since you signed up. It is metered. It is monitored, your questions are data, and data is retained, reviewed, and monetized in ways described honestly in documents no household has ever read. It can be repriced, throttled, re-personalized, or withdrawn. What you have, in the old language, is a service. What you are, in the old language, is a tenant.
The second way is quieter and, at the moment, less convenient: the model lives in the house. It runs on a machine on the desk or the kitchen counter. It works when the internet does not. It answers to no terms of service because there is no service. Your questions remain in your home the way conversations at a kitchen table are supposed to remain in your home. It is usually a step behind the best rented intelligence, sometimes two steps. It is also, in every way that compounds over a childhood, yours.
I have spent enough time thinking about centralized power in the large, empires, agencies, platforms, that I have come to believe the household is where the principle gets tested first. Catholic social teaching calls it subsidiarity: matters should be handled by the smallest, lowest, most local authority competent to handle them. We usually apply the idea to governments. Apply it to intelligence. Every question a family can competently answer in its own house, with its own tools, under its own judgment, is a question that does not need to be asked of an empire, and does not become part of an empire's memory.
The rented model is better. I want to be honest about that, because the argument for local tools is often made by people pretending the gap isn't real. The frontier systems are remarkable, and there are tasks for which nothing local will do. But "better" is doing quiet work in that sentence, better at the task, this month, for this benchmark. The history of technology suggests the gap is a tide, not a wall. What required a mainframe became an application. What required a data center is becoming a download. The ordinary pattern of computing is that capability sinks, steadily, toward the person. The only question is whether our habits, laws, and markets let it land there, or whether they keep it fenced in the cloud on a meter.
Habits matter more than we admit. A generation of children is learning right now where thinking happens. If every hard question is sent away, typed into a box, answered from elsewhere, by a system that also reports the question, then thinking itself starts to feel like something that happens elsewhere, in buildings owned by other people, who are kind enough to rent it back. That is a formation, whether we chose it or not. The kitchen table used to be where a family worked things out: homework, budgets, letters, arguments, plans. A tool that sits on that table, owned like the other tools in the house, keeps the working-out at home. A service that replaces the table relocates the family's mind.
None of this requires refusing the cloud. It requires refusing dependency, the condition where the rented thing becomes the only way the household can function, and its owner knows it. Dependence is the point at which a customer becomes a subject. We would recognize the arrangement instantly in any other domain. A man who cannot repair his own tools, read his own contracts, or teach his own children without a monthly permission from a distant corporation is not being served by modernity. He is being kept by it.
So the practical counsel is small, the way most durable counsel is. Learn the local tools, even while they are clumsier. Keep copies of what matters on machines you control. Teach the children that the box answers questions but the family decides, that a model's output is a draft, not a verdict, and that privacy is not paranoia but the ordinary dignity of a home. Support the open projects, the downloadable models, the stubborn people building intelligence that can be inspected and kept, because they are building the difference between a population that uses AI and a population that is used by it, politely, at $20 a month.
Empires are not usually seized. They are subscribed to.
The alternative is not heroic. It is a machine at the kitchen table, doing ordinary work for the people who live there, under their authority, inside their walls. That is what keeping intelligence close to home means. It is a small arrangement, and like most small arrangements faithfully kept, it decides what the large ones will be allowed to become.