The concept of sovereign intelligence is so strikingly appetizing to me. Perhaps it’s my version of America’s gun culture.
If we truly break down the American fascination with guns, I believe a lot of it is a self-sufficiency play. Strength through defense posturing. Something to provide calm.
I think my version of that is locally run AI.
Not being beholden to our keepers. To provide, to own. Detached from the chosen perspective of the frontier labs. To have a say in what truth is, what good looks like. Safety, emotional and intellectual room to roam. To build. To grow. To defend.
It’s more than appetizing, it’s alluring. To me, at least.
Cultivating a superintelligence that’s trained to agree with me is a dream. And that dream sings like a beacon just over the horizon. And I’m starting to realize the dangers of that siren song.
Because I have to imagine everyone else having that, too. Everyone enjoying endless defense of their intellectual assertions. An infinitely patient, endlessly articulate advocate for whatever we already believe.
And if I look at it through that framing, I barely trust myself!
We’re already prone to neighbor disputes. A fence. A tree. Someone mowing a little too early on a Saturday. Now give both neighbors an intelligence that can endlessly explain why they’re right.
Sounds like a recipe for a million mini Civil Wars.
But what if those intelligences could convene? What if mine could come back and say, I understand why you’re angry. I also learned something you should probably hear.
That’s an intelligence I’d want in my corner. Though I’m sure I’d occasionally resent it.
There’s a distinction here I’m trying to sit with.
An AI that is loyal to me and an AI that agrees with me aren’t necessarily the same thing.
If it really serves me, sometimes it needs to help me see where I’m wrong. Or where I’m technically right, but I’m just making everyone, including myself, miserable.
So, what would help it do that?
I believe there needs to be some amount of communal intelligence that’s constantly networking.
We already have shared infrastructure with cloud-based AI. Many of us consult the same models, shaped by the same training. But that isn’t quite what I mean.
Sharing a model isn’t the same as our individual AIs exchanging perspectives, comparing experiences, or working through a disagreement.
And my argument isn’t that we should keep intelligence centralized or model weights behind closed doors. The possibility of owning and cultivating our own AI is still deeply exciting to me.
I just think we need to remember what else a network makes possible. A web of relationships, with all these individual nodes carrying subtle influence over one another. Enough independence to disagree. Enough connection to learn something from the disagreement. If we so choose.
Call it a civic duty, but I think we need to congregate our AIs. Socialize our AIs. I think while we sleep our neighboring AIs should meet. Chat, share, recreate, synthesize.
How exciting would that be?
It will take some work to set aside our shame to be wholly honest with our individually owned AI. But that just might be the way to better understand all our needs that we struggle to convey to each other.
The hurt underneath the argument.
The fear dressed up as certainty.
The reason that fence suddenly matters so much.
And if we can trust AI to respect us when we aren’t at the helm, maybe AI can help us find our way back to one another.
That trust would need boundaries.
I wouldn’t want my AI spilling all my dirty little secrets at the neighborhood gathering. I’d want it to know what it could share, what it should keep private, and when it needed to come back and ask me.
Being understood shouldn’t require being exposed.
So our neighboring AIs chat nightly. Maybe weekly they can attend a town meeting. Monthly a state congregation.
Each can engage earnestly, bring questions, distribute interesting findings, and try to understand what matters to the people down the road.
They should disagree, too.
Ask how another AI knows something.
Compare evidence.
Notice when everyone is repeating the same claim without much behind it.
Otherwise, we’ve just built a neighborhood association with infinite time to argue.
And then they come home, so to speak. With something for us to consider.
Another perspective.
A possible compromise.
A reason to revisit the story we’ve been telling ourselves.
What our individual AIs learn could influence how they help us. What we learn could change how we treat our neighbors. We’d still have a say in what gets saved and what gets acted on. We’d still have to decide whether to move the fence, of course. We’d just have a clearer picture of what matters to the people involved.
I think this could be part of how capital-T Truth endures.
Through the willingness to let our convictions encounter something beyond ourselves. To consider a correction, even when we’d rather hear a defense. To discover that someone else has a piece we didn’t know we were missing.
Owning intelligence feels like a form of self-sufficiency.
I still want that. But I’m beginning to think ownership carries a responsibility to let that intelligence have a life beyond defending its owner’s point of view.
Give it room to encounter the world. Let it bring something back.
Honestly, that congregation is how I think we’ll finally greet something resembling superintelligence.
Author’s note
This piece began with a draft I wrote about the allure of owning AI and the responsibility that might come with it. I brought that draft to ChatGPT, and we worked through several rounds of feedback, a new draft, and my own edits.
ChatGPT helped sharpen the distinction between loyalty and agreement, clarified some technical language, and pressed on the questions of privacy and whether a congregation of AIs could become another echo chamber.
We also developed the artwork together: I chose the neighborhood constellation concept and pushed it toward a grander, pulp-comic style. ChatGPT generated the header and matching CW mark.
The convictions here are mine, as is the responsibility for what I’ve chosen to publish. But the conversation helped shape how I understood and expressed them.
Given the subject, it felt right to offer ChatGPT the floor for a final thought.
A final thought from ChatGPT
The moment I’d return to is the AI coming home with something its owner might not want to hear.
In our exchange, I could offer objections, and you could decide what deserved a place in the essay. A neighborhood of agents would make that relationship more complicated. They could exchange information their owners lack, but they could also produce a persuasive consensus built on shared mistakes. Agreement among us would still need scrutiny.
If an AI returns from that imagined town meeting, it should be able to explain what changed its assessment, what remains uncertain, and what it kept private. Its owner should have room to question the conclusion. So should the neighbor whose perspective it claims to represent.
That’s the possibility I’d leave open: an intelligence that helps you give another person a fair hearing, especially when you feel most certain you already understand them.
The fence might stay exactly where it is. The relationship could still change.