I’m looking forward to the day when the word AI in a headline reads the way electric does now. Electric toothbrush. Electric kettle. The word survives in a few places where it still distinguishes one thing from another, and everywhere else it fell away, because announcing that something runs on electricity stopped telling anyone anything.
While out walking our dog this morning, I found myself turning the comparison into a question. Is electricity as interesting as what you can make under a lamp at night? Is the internet as interesting as what you can send over it to someone? And then the one I actually care about: is AI as interesting as what it helps you do?
I don’t think it is, and I mean that as a compliment. AI is a boring technology. It’s infrastructure — and the sooner we get used to thinking of it that way, the sooner we can stop being astonished and get back to making things.
But I also want to be careful with that word, because it’s doing a lot of work and using it risks overlooking the ways that AI is meaningfully different from electricity or the internet.
Electricity never produced anything that resembled judgment. Nobody ever suspected the grid of having opinions about them. AI produces language, and language is the thing we use to recognize a mind, so people reach for the only category they have. I don’t think that reflex is a failure of intelligence or a lack of technical literacy. We name our cars. We apologize to furniture we bump into. We see faces in wallpaper and clouds and the fronts of houses. The impulse comes from a drive to connect, and connection is most of what lets a human mind flourish. A technology that meets that drive as well as this one does was always going to be met with more than curiosity.
That is also where the danger is. A mind in poor shape can be handed its own thoughts back, slightly improved, at any hour, forever — and take that for company. Media has done a version of this for a century. This does it more intimately and more responsively. That’s a real risk, and it’s a social one. Our job is to integrate what we build into the society we actually are rather than the one we describe in the press release. If we can’t handle it, that’s on us, and part of the work is agreeing collectively that it’s on us.
Calling AI “infrastructure” is also, at least at this point, aspirational. There’s a lot of boring work to do to make that so. I believe we can do it, but it remains undone.
Electricity did not arrive boring. It became boring over decades, through standardization and public investment and regulation and a great deal of argument about who should own the wires and who was owed service. Somebody had to decide that a farmhouse at the end of a long road deserved power at a price it could pay. That was not a technical question, and it wasn’t settled by the technology maturing. It was settled by people insisting on it.
And through all of it, the infrastructure stayed owned. It always is. Some of it is public, theoretically held by all of us and funded by our taxes, but public ownership means government control, and if you don’t care for the government you have, that control doesn’t feel meaningfully different from a corporation’s. Ownership isn’t the axis that protects anyone. It never was.
Which is why I’m less worried than I expected to be about who owns the models. There are already good open-source ones, and they’re getting better and cheaper to run. The imbalance sits downstream of the models, in the assets that make them work at scale — chips, electricity, water, land. That’s where the concentration is, and it’s more sobering to me than the terminator narrative, which has been very good for the share prices of the companies telling it.
The work is in two places. Regulation, which first requires the political will to want it, and which nobody gets to skip by hoping the technology will mature into fairness on its own. And efficiency — continued ground gained on open models small enough to run on ordinary machines. The future I want is local — off-grid AI. A model on a laptop that belongs to the person using it, the way a lamp belongs to the person who switches it on.
We’ll know it worked when the word disappears from the headlines. Not because anything was settled dramatically, but because the interesting part will have moved, as it always does, to whatever someone made with it.