The Source Is Not the Idea
You read an argument that would have moved you, but the person making it is wrong about something else — politics, a public feud, AI safety, whichever — and you find yourself unable to take the argument seriously. You haven't evaluated it. You've evaluated the source. This happens to you constantly, and you mostly do not notice.
It also happens in the other direction. Someone whose name you trust says something, and you accept it, and the acceptance feels like evaluation. It isn't. Your trusted source's conviction has passed through your own confidence and come out feeling like a belief you arrived at. You did not arrive at it. You inherited it.
Those are the visible versions of the move. The version worth naming is what the same reflex does the rest of the time, the way provenance is doing most of the work in most of your beliefs, most of the time, without your noticing.
The Tag and the Belief
Provenance is a useful thing. In a world where you cannot evaluate everything from scratch, it is a reasonable triage signal. It tells you whose conviction you are running on. It tells you how much weight to put on a position, whether to update easily or stand firm, and whether to defer to people who disagree. Used this way, provenance is a working part of how you think. It is honest about what it is.
Sometimes the tag should dominate. If a source has repeatedly lied, fabricated, or operated in bad faith, discounting their claims is not laziness. It is memory. The error is not using provenance. The error is letting it close the question rather than telling you how to open it carefully.
The failure is when the tag becomes the belief. When "Harvard says it" stops being a reason to read carefully and becomes the reason you don't have to read at all. When a credentialed person's conviction begins to feel indistinguishable from your own. That is the moment you have stopped thinking and started bookkeeping. The argument doesn't get evaluated on its merits. It gets evaluated on the standing of its maker. If the maker is in good standing, you accept. If they are in bad standing, you reject. The evaluation has been outsourced to your tribal accounting.
This is not thinking. This is sorting.
The Map You Did Not Draw
Now ask yourself how much of your map of the world is built this way.
Your view of the economy. Your view of the energy transition. Your view of which technologies matter, which institutions are decaying, and which companies or countries are rising. How much of that is conviction you earned? How much is laundered from a small set of trusted names you would have to admit if pressed?
For most people, on most topics, the honest answer is that almost all of it is borrowed. That is fine. You cannot do everything. The dishonest answer is to deny it, to mistake the borrowed map for one you drew yourself. The cost of that dishonesty is that you cannot tell when your sources are wrong, because you have no independent purchase on the territory. When consensus shifts, you shift with it, and you tell yourself you updated. You did not update. Your sources updated. You copied them.
This is what it means to know the names of the people who believe a thing rather than to believe it yourself. It is a respectable failure mode. Almost everyone is in it almost all the time. The respectable part does not make it less of a failure.
What the AI Rejection Reveals
What has gotten loud lately is the AI rejection. People won't read AI-written articles. Open-source projects refuse AI-generated contributions on policy, before review. The reasoning shifts when you press on it: quality, copyright, environment, contamination, and some of it is real. Maintainers have limited time. Generated work can be sloppy at scale. Licensing questions are not imaginary. Skepticism toward AI-generated contributions is warranted.
But those are reasons to define a review standard, not reasons to replace review with a provenance ban. The revealing move isn't skepticism. It treats the origin as dispositive before the artifact has been examined. "I'll review your AI-assisted code with extra care because slop is cheap" is doing the work. "We measured the review burden and the merge rate and we're closing this channel until the economics change" is also doing the work — that ban is empirical, not tribal. "We don't accept AI contributions, period" is replacing the work with a name check, and most current bans are that version with measurement language painted on after.
That second move exposes something the credential version always concealed. The credential filter, when it works as intended, produces real epistemic value; peer review filters, reputation accumulates, and accountability matters. When it fails, it fails the same way the AI ban fails: by treating origin as a substitute for evaluation rather than an input to it. The AI ban is the cover-off version of the same move. The operational reasons it appeals to, such as slop, time, and licensing, are concerns that review can handle. The ban does not solve them. It moves them out of sight and replaces the epistemic work with a sorting check.
The code case sharpens this further. Code is among the most evaluable artifacts humans produce. It can compile or fail. Tests can pass or fail. Review can inspect concrete behavior. The bar is not perfectly objective, but it is far less mystical than taste or politics or literary judgment. The infrastructure to apply that bar exists. The competence exists. And then it gets suspended for AI-generated code in favor of a name check. The most evaluation-capable population on earth retreating to the laziest possible filter, and discovering they were never quite as evaluation-capable as they thought.
The Move
The move is unglamorous and slow.
Pick the small number of things you actually care about. The things where you would feel embarrassed to be wrong, where being wrong would change a decision you make. Go deep on those. Read the primary sources. Build the model. Run the numbers. Get to the point where you would back yourself against your own trusted sources, because you have done enough work to see where they are simplifying or motivated or just out of date.
For everything else, and that will be most things, hold your beliefs at the right weight. Be honest that you are running on borrowed conviction. Mark them internally with their actual provenance, so that when someone challenges you, you know whether you are defending a position you have earned or a position you have inherited. The first you can defend. The second you should be willing to drop the moment your sources do.
This is the discipline. Not "evaluate everything from scratch", nobody can do that. The discipline is knowing which of your beliefs are yours and which are borrowed, and being honest about the difference. Provenance as a tag on a borrowed belief is fine. Provenance as a substitute for the belief you should hold yourself is fatal.
The work you do on the small number of things you actually care about is the only thing that is actually yours. Everything else is rented, and the rent comes due whenever the world shifts and your sources are wrong before you are. The discipline that helps with the rented part is not pretending you own it. Diversify your creditors. Track when they agree because they are independently right and when they agree because they are reading each other. A person who has done real work on three things and admits it is a more reliable thinker than a person who has done real work on three things and pretends to have done it on thirty.
Pick your three. Do the work. Carry the rest as borrowed.
