borys
← Writing

AI · ∩

2026-07-11~16 minSolertio · esej

Cognitive sovereignty: choosing whose model of the world runs in your head

Every system competing for your attention is also, quietly, competing to supply your default thoughts. Recommendation engines optimise for engagement, not for your interest, and language models now compress the world's text into a single confident voice that is one prompt away at all times. Cognitive sovereignty is not the fantasy of thinking unassisted. It is the deliberate choice of whose priors you let run in your head — and it is getting harder to exercise by accident.

Attention was scarce before it was mined

Herbert Simon saw the shape of this in 1971, before the feed existed: a wealth of information creates a poverty of attention. Attention is the true scarce resource, and once something is scarce and valuable, an industry forms to extract it. The feed is that industry. Its objective function is not your understanding or your interest; it is engagement, measured in time and clicks, and a system optimised for engagement will reliably serve you what holds you, which is not the same as what serves you.

This much is a decade old and widely understood. The new layer is sharper. A large language model does not just compete for your attention; it offers to do your thinking, and it does it well enough that the temptation is real. Ask it what to conclude and it will hand you a fluent, averaged answer — and if you take that answer without friction, you have quietly inherited the model's priors as your own. The danger is not that the model is usually wrong. It is that it is usually plausible, and plausible-by-default is exactly how a frame installs itself without being examined.

Sovereignty is a choice, not a neutral default

The instinct is to treat the solution as neutrality — just consume less, or consume "objectively." But there is no neutral position. You are always running on some model of the world, ingested from somewhere. The only question is whether you chose the source or the source chose you. Cognitive sovereignty is the deliberate version: consciously deciding whose frames, sources, and tools you let shape your thinking, rather than defaulting to whatever the optimisation happens to serve.

A wealth of information creates a poverty of attention.Herbert A. Simon

Concretely, sovereignty has infrastructure. A second brain — a personal wiki, a body of notes you own and control — is not a productivity toy; it is a sovereignty tool. It means your model of the world lives in a place you govern, built from sources you chose, rather than being reassembled each morning by whatever the feed decided to show you. When you think with your own notes, you are thinking with priors you selected. When you think with the feed, you are thinking with priors that were selected for their ability to hold you.

Using AI without outsourcing the judgment

The practical discipline is the same one that separates a tool from a master: let the model do the work, keep the judgment. Use it to generate, to summarise, to argue against you, to fetch — and reserve the act of deciding what is true and what matters for yourself, from a frame you can articulate. The tell that you have lost sovereignty is subtle: it is when you notice you are agreeing with a conclusion you did not reach, in words that are not yours, because the answer arrived pre-formed and frictionless.

I run a studio where AI does an enormous share of the execution, so this is not an abstract worry for me. The rule I hold is narrow and load-bearing: the model may draft anything, but the model may not decide what is good. The moment the frame comes from the tool rather than from me, the tool is running the studio.

A worked input diet

Sovereignty stays abstract until you audit your inputs the way you would audit a portfolio. Spend a week noticing what actually shapes your model of your market and your field: the feeds you open by reflex, the people whose take you absorb before forming your own, the search results you treat as the answer. For most people the honest finding is that the largest inputs were never chosen. They were served, by systems optimising for your attention, and you have been running your judgment on priors selected for their stickiness rather than their truth.

The redesign is not to unplug, which is impossible, but to move the biggest inputs from default to chosen. In practice that means two things. First, a small set of sources you selected on purpose — a few writers, primary documents, one or two people who disagree with you well — that you go to deliberately, rather than whatever the algorithm surfaces. Second, a second brain that captures what is worth keeping: when you read something that changes your model, it goes into a note you own and can find again, not a like that vanishes into a platform's database. Over a year, the difference is stark. One person's worldview is a set of conclusions they chose and can trace to sources; the other's is reassembled each morning by whatever held their attention yesterday.

The AI layer sits on top of this without breaking it, if you hold the line. On a real decision — say, whether to enter an adjacent market — you can have the model argue both sides, fetch the data, and summarise the field in minutes. What you do not do is ask it what to conclude and copy the answer. You write the conclusion yourself, from a point of view you can articulate, using the model as research and adversary rather than oracle. The tell that you have kept sovereignty is simple: at the end, the judgment is one you can defend in your own words, not one you inherited pre-formed because it arrived fluent and frictionless.

Where the idea curdles

Cognitive sovereignty is a seductive banner, and seductive banners recruit exactly the failure modes they warn against. Three are worth stating.

One: sovereignty becomes a bubble. If curating your inputs means ingesting only what you already believe, you have not escaped the algorithm — you have become your own recommendation engine, optimised for comfort instead of engagement. It is the same failure, self-imposed and harder to see. Real sovereignty is not a walled garden; it deliberately includes the strongest sources that disagree with you, because a model of the world that never meets resistance stops tracking the world.

Two: the fully sovereign mind is a myth. Andy Clark and David Chalmers argued that cognition is already extended — it runs on notebooks, tools, and other people, not sealed inside one skull. We have always thought with borrowed instruments and other minds; that is not a bug to be defended against but the normal condition of a thinking animal. "Total" cognitive sovereignty is neither achievable nor desirable. The honest goal is not independence from all outside frames but conscious selection among them.

Three: the defence can cost more than the attack. Taken too far, guarding your attention becomes a full-time obsession — an elaborate digital-minimalism practice that consumes more of the thing it claims to protect. And there is an irony worth admitting: the people most anxious about cognitive sovereignty are frequently the most extremely online, performing autonomy as a kind of status. The point is to think well, not to win a purity contest about how you think.

So choose your sources on purpose, build somewhere to keep your own model of the world, and let the machine draft but never decide. Then, deliberately, invite in the strongest voice that thinks you are wrong — because a sovereignty that never hears it is just a nicer prison.

Sources

  1. primaryHerbert A. Simon, "Designing Organizations for an Information-Rich World" (1971) — attention as the scarce resource.
  2. primaryAndy Clark & David Chalmers, "The Extended Mind," Analysis (1998).
  3. primaryCal Newport, Digital Minimalism (2019) — deliberate technology selection.
  4. secondaryJames Williams, Stand Out of Our Light (2018) — the attention economy and the will.
  5. secondaryTiago Forte, Building a Second Brain (2022) — externalised knowledge systems.